Sardine named a Leader in The Forrester Wave™: Financial Crime Management Solutions, Q3 2026

Learn More
The Saturday Fraud Strategist

AIで開発が低コストに。不正対策ベンダーは、今なお何のために必要なのか?

64 min

ベンダー業界では常に開発競争が繰り広げられてきましたが、今年はAIがその中心です。一方、私が話を聞く現場の実務者たちは、以前よりもはるかに多くのものを自分たちで開発しています。数週間前、あるリスク管理責任者から、彼女のチームが社内で開発したものについて話を聞き、その内容に感銘を受けました。それをきっかけに、ベンダーを取り巻く環境が根本から変わりつつあるのではないか、そして1、2年後にも私たちはベンダーを必要としているのだろうかと考えるようになりました。

この問いを一緒に考えるため、Netanel Kabalaを招きました。Netanelと私がPayPalで出会ったのは17年前です。テルアビブのFraud SciencesチームがPayPalに買収された後、彼が不正対策アナリストとして働き始めた頃でした。その後、彼は暗号資産の購入を可能にするオンランプサービスの先駆けとなったSimplexを共同創業し、同社の買収後はNuveiでAIとアナリティクスを統括しました。最近では、決済事業者やフィンテック企業が意思決定システムを自社で管理・運用できるよう支援するDulyを共同創業しています。今回は、不正対策システムを内製するか、ベンダーから導入するかというテーマを、ベンダー疲れ、チームが最初に内製するもの、基盤モデルや意思決定モデルがもたらす変化など、さまざまな角度から考えます。結論は単純なイエスかノーにはなりませんが、それが誠実な答えだと思います。その代わりに、どの領域を内製し、どの領域で外部製品を導入し、どの領域で共同開発するのが合理的か、その見取り図を描きます。

このエピソードで取り上げる内容:

  • ベンダー疲れがチームを自社開発へと向かわせる理由と、それがベンダーとの関係に及ぼす影響
  • これまでに最も成功している内製開発が、プラットフォーム全体を置き換えるものではなく、専門家一人の作業時間を短縮する小さなツールである理由
  • ベンダーがエンドツーエンドのソリューション提供からデータ提供へと軸足を移しつつある状況と、大規模なチームがネットワークデータやデバイスインテリジェンスを特に重視する理由
  • AIコーディングで手動レビュー用ビューアを簡単に構築できる理由と、ルールエンジンの開発がまったく異なる種類のプロジェクトである理由
  • 一夜にして作られる数十ものモデルから、誰も想定していなかったレイテンシーやドリフト監視の問題まで、AIが生み出す低品質な成果物がリスク管理チームにどのような形で現れるのか
  • 基盤モデルと意思決定モデルによって、特徴量や異常検知はどう変わり、リアルタイムの不正検知モデルを誰がデプロイできるようになるのか
  • セグメンテーションが最適な出発点である理由と、KYC(本人確認)や制裁リスト照合のような汎用機能の構築が落とし穴となる理由
  • エンジニアリング面の支援、リスク管理の専門人材の確保、経営陣の賛同などの観点から、自社が内製化に取り組む準備ができているかを判断する方法
  • プロダクトマネージャー、DevOpsエンジニア、バックエンドエンジニアを兼務するアナリストが、保守の罠に陥る理由

次のような方におすすめのエピソードです:

  • 不正対策やリスク管理チームを率いており、何を内製し、何を引き続き外部から調達するかを検討している方
  • ベンダーとの付き合いに疲れを感じ、どの取引関係を維持する価値があるのか、より明確な基準で考えたい方
  • AIを活用して社内で何かを開発し、それが規模の拡大に対応できるか知りたい
  • 短期間の概念実証にとどまらず、開発ロードマップに何が必要かを経営陣に説明する必要がある
  • 基盤モデルや意思決定モデルが、特徴量レイヤーやベンダーデータをどのように変える可能性があるのかを知りたい
エピソードの概要と重要なポイント

ベンダー疲れがチームの内製化を後押し

Netanelは、業界のさまざまなチームから同じ話を聞いています。誰もが自社開発へのプレッシャーを感じており、それはAIモデルが実務をこなせるようになるずっと前から始まっていました。今ではAIモデルにその能力が備わり、ベンダー疲れと実現可能性が結びついています。彼が見てきた中で最も成功している自社開発の事例は、有能なチームメンバーが作業時間を短縮するツールを作るといった、小規模なものです。長年、ベンダーは判断や保証、スコアを完成されたパッケージとして提供してきました。チームには、それ以外のものを使いこなす専門知識や余力がなかったからです。今ではチーム側が、もっと細かな単位でも扱えると考え、カスタマイズ可能な部品を求めています。私の見方では、ベンダーは今後ますますデータ提供者として捉えられるようになる一方、負荷の大きいリアルタイム処理のエンジニアリングは、引き続きベンダーが担うことになるでしょう。両者の関係は変化しており、その姿は業界ごとに異なるものになるはずです。

ビューアーの構築は容易だが、ルールエンジンはそうではない

Netanelは、情報を表示するビューアーと、実際のアクションにつながる製品を区別しています。手動レビュー用の操作画面を作るのは簡単な部分です。AIによるコーディングで容易にカスタマイズできるうえ、チームはベンダーが変更対応の優先順位を上げてくれるのを待つことにうんざりしているからです。一方、ルールエンジンの構築は別の仕事です。リアルタイム性能、高可用性、効率的なデータアクセス、そしてテストとシミュレーションが求められるバックエンドエンジニアリングです。Simplexでは、当初からPythonでルールエンジンを構築していたため、チームは戦略を自動的に切り替えることができました。多くのチームは、決済を審査するために複数のソースからデータを集めるなど、手作業の分析から着手します。ベンダーのアルゴリズムを丸ごと置き換えたいと考えるチームはほとんどありません。彼はまた、私にも心当たりのあるリスクとして、AIが粗製乱造する低品質な成果物、いわゆる「AIスロップ」を挙げました。新たな不正の問題に直面したチームは、一晩で20個のモデルをリリースできてしまいます。その結果、レイテンシやドリフトの監視が、誰も想定していなかった課題になるのです。

基盤モデルが変えるのは特徴量のレイヤーであり、業務全体ではない

私はNetanelに、基盤モデルはベンダーにとって脅威になるのかと尋ねた。ベンダーが競争優位を築いてきた、手作業で設計する特徴量のレイヤーを置き換えてしまうからだ。彼は、標準的なLLMがすでにその作業の多くを担うようになったと考えている。データセットからどのような特徴量を抽出すべきかLLMに尋ねれば、専門家チームが生み出すものの約80%は得られるというのが彼の見立てだ。彼は、基盤モデルの真の価値は高度な異常検知にあると見ている。私は、これまでに見た中で最も優れた成果は、過去の履歴がある既知の対象に関するものだったと付け加えた。より大きな変化をもたらすのは、必ず何らかの決定を下す必要があり、その判断にLLMを用いる半決定論的な意思決定モデルかもしれない。こうしたモデルは、外部データ、ネットワークデータ、特徴量ストアの価値を高める。彼が使った料理人のたとえが、私の心に残った。食材がそろっていることと、料理が予想より早く煮立ったときや、シミュレーションに含まれていなかった攻撃が起きたときにどう対処すべきかを知っていることは、別なのだ。

ターゲット層のニーズが満たされていない領域で開発する

Netanelの最初のアドバイスは、セグメンテーションについて考えることです。新規ユーザーとリピーター、地域、商品、決済方法によって、その傾向は大きく異なります。成果が出ていない領域や採算性の低い領域を探し、別のベンダー、AI、独自のソリューション、新たな人材の採用など、何かを試してみましょう。私からも付け加えるなら、こうした領域こそ内製化が最も効果を発揮するところです。規模の大きな企業には、どのベンダーにも対応できない、その企業特有のセグメントがあるはずです。盗難カードへの対策はそれに当たりません。制裁対象者のスクリーニングやKYC(本人確認)も同様です。そこに時間を費やすのは、陥りがちな落とし穴です。2つ目のアドバイスは、自社の事業にとってリスク管理がどれほど中核的なものかを問うことです。そのうえで、過去6〜12か月のロードマップを振り返り、リスク管理の仕組みが障壁となって遅れた施策がいくつあったかを数えてみてください。内製するなら、意思決定を最小単位の要素に分解し、それぞれを自社で開発するか、外部から調達するかを判断しましょう。

準備態勢とメンテナンスの落とし穴

Netanelは、リスク管理部門の責任者の多くが自分のチームを厚く信頼しているものの、それが盲点になり得ると述べました。彼が勧めるのは、組織の体制を見極めることです。必要なエンジニアリング能力はあるか。投資を回収するまでに先行コストがかかるプロジェクトを経営幹部は支援してくれるか。リスク管理の人材を継続的に確保できるか。私はそこに、チームが最も見落としがちな点を付け加えました。作るのは簡単で、行き詰まるのは保守の段階です。アナリストが、自分のツールのプロダクトマネージャー、DevOpsエンジニア、バックエンドエンジニア、UXエンジニアをすべて兼ねる羽目になることもあります。それでもアナリストとしての仕事がなくなるわけではありません。結果として、プロジェクトの保守が止まってしまうのです。経営陣の理解と支持を得るには、2日で概念実証を作るのとは別のスキルが必要です。リスク管理チームも、内製化の落とし穴とは無縁ではありません。

最後のまとめ

不正対策を内製するか、ベンダーのソリューションを購入するかは、単純にイエスかノーで答えられる問題ではありません。意思決定基盤のどの部分を自社で保有すべきか、どの部分は外部から購入したほうがよいかという問題です。これをうまく進めるチームは、まず自社のセグメントを見極め、内製したものの保守コストを見積もり、コードを一行書く前に経営陣との認識をそろえます。今回の対話は、確信よりも問いを多く残して終わりました。これほど速く変化し続けているテーマには、それがふさわしいと思います。

参考資料とリンク

私の、そして願わくば皆さんの大好きなテーマについて、もっと語り合いたいですか?ニュースレター「サタデー・フラウド・ストラテジスト」にぜひご登録ください。

つながりましょう:Netanel Kabala | LinkedIn
Duly共同創業者

つながりましょう:Chen Zamir | LinkedIn
「The Saturday Fraud Strategist」ホスト
フィンテック企業による、よりスマートな不正対策の構築を支援
『不正対策担当者のためのAI実践ガイド』共著者

Episode transcript
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:00
If you go and ask an LLM this is my data set
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:02
which features I should derive from?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:05
You'll have I I dare say 80% of the value that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:08
would be created by a team of professionals.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:11
I'm not sure you you even need data, training data I mean.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:15
I think that there's vendor fatigue.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:17
Everyone has pressure or desire or motivation to build.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:21
Whether to buy, I think that the real promise or square depends
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:25
on which side of the fence you're on is the new
Chen Zamir
Chen Zamir
00:36
What's up everybody?
Chen Zamir
Chen Zamir
00:37
Welcome to another episode of the Saturday Fraud Strategist.
Chen Zamir
Chen Zamir
00:40
With me today, Netanel Kabala.
Chen Zamir
Chen Zamir
00:42
Netanel, we've been uh we we've known each other for
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
00:46
17 years, I believe. Yeah.
Chen Zamir
Chen Zamir
00:49
Yeah. Okay, you said it. Yeah, that's true. That's true.
Chen Zamir
Chen Zamir
00:52
And you know we've known each other for a long time.
Chen Zamir
Chen Zamir
00:55
Uh funnily enough I I'll let you introduce yourself and tell you tell
Chen Zamir
Chen Zamir
01:00
a bit about uh your own journey in the fraud prevention uh space.
Chen Zamir
Chen Zamir
01:04
Uh but I'll say you know from my side uh that
Chen Zamir
Chen Zamir
01:06
our journeys are very very similar.
Chen Zamir
Chen Zamir
01:09
Uh maybe yours is a bit more successful.
Chen Zamir
Chen Zamir
01:11
Uh but yeah I'll I'll let you introduce yourself and
Chen Zamir
Chen Zamir
01:14
tell tell our audience a bit more about yourself.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:18
I'm uh I started uh my fintech journey uh in
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:20
PayPal uh 17 years ago where I met Chen.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:23
Um I started as a fraud uh analyst uh as part of uh PayPal's
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:27
Tel Aviv fraud science fraud sciences acquisition.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:30
Joined briefly after the acquisition.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:33
Uh I spend most of my time on seller side risk uh creating
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:37
uh different strategies models for uh PayPal's sellers.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:40
Um and after five years where I became also a manager and
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:44
a group manager, I left PayPal to pursue the uh my uh founder
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:48
dreams and uh co-founded a company called Simplex.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:52
Uh Simplex was basically the first crypto
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:54
on-ramp provider uh globally.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
01:56
Um, we allowed different businesses in the crypto space, uh, most
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:01
notably exchanges and wallets to provide their customers with,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:05
uh, uh, the option to purchase crypto via uh, uh, payment cards
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:09
and later on via other payment methods all with a full uh,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:13
chargeback guarantee and the full end to end solution.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:16
So I was in charge of everything related to
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:19
risk, compliance, data science and so on.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:22
Uh um the company was acquired by Nuvei in 2021.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:25
And I then spent uh three and a half years in charge
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:29
of Nuvei's AI and analytics department.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:31
Um and after a small break and some uh consultancy I
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:35
recently uh co-founded a new company called Duly.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:38
Duly is uh basically trying to help uh different businesses uh most
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:43
notably uh payment providers and uh FinTech to own their
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:46
decisioning stack uh and to build uh the capabilities that they
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
02:50
need to grow and and be as successful as top tier companies in that space.
Chen Zamir
Chen Zamir
03:43
That's an amazing story.
Chen Zamir
Chen Zamir
03:44
Uh as I said, very similar to my own, a tiny bit more successful.
Chen Zamir
Chen Zamir
03:48
You know the reason why I wanted to bring you on the show and the
Chen Zamir
Chen Zamir
03:53
topic that uh we wanted to discuss today, also to an extent ties
Chen Zamir
Chen Zamir
03:57
uh to to what you guys are thinking of doing uh with Duly now,
Chen Zamir
Chen Zamir
04:01
but uh there's an arms race right in the vendor space.
Chen Zamir
Chen Zamir
04:05
And there was always an arms race in the vendor space.
Chen Zamir
Chen Zamir
04:08
It's it's it's forever the dynamic of the field.
Chen Zamir
Chen Zamir
04:12
Evidently like very obviously in the last year
Chen Zamir
Chen Zamir
04:15
uh it was all about AI in the last year.
Chen Zamir
Chen Zamir
04:17
And so vendors are all trying to develop
Chen Zamir
Chen Zamir
04:20
these new capabilities around AI.
Chen Zamir
Chen Zamir
04:22
And this is interesting because this is what I'm seeing uh
Chen Zamir
Chen Zamir
04:26
on the practitioner side on like in the teams themselves.
Chen Zamir
Chen Zamir
04:30
Uh and with uh we had uh Lilitha Rao uh from Imprint
Chen Zamir
Chen Zamir
04:33
uh just a few weeks ago here on the pod.
Chen Zamir
Chen Zamir
04:36
And and you know she described what they've built in
Chen Zamir
Chen Zamir
04:39
house which was you know very very impressive.
Chen Zamir
Chen Zamir
04:43
Uh a lot of the clients that you know I I meet and talked to today they
Chen Zamir
Chen Zamir
04:47
are doing much more building than they've been doing in the past.
Chen Zamir
Chen Zamir
04:52
And so I'm thinking to myself while you know vendors
Chen Zamir
Chen Zamir
04:55
are kind of like trying to reach for the frontier
Chen Zamir
Chen Zamir
04:58
this moving frontier uh ever moving frontier.
Chen Zamir
Chen Zamir
05:01
Um I'm thinking is the ground shifting beneath vendors fit here?
Chen Zamir
Chen Zamir
05:06
Um would we actually need vendors in one or two years?
Chen Zamir
Chen Zamir
05:09
Uh would they have their own uh place?
Chen Zamir
Chen Zamir
05:12
And you know of course the reason why I'm asking all of that is because
Chen Zamir
Chen Zamir
05:16
building software in general has become very cheap, very accessible.
Chen Zamir
Chen Zamir
05:21
Uh much smaller teams with much less budget can now um
Chen Zamir
Chen Zamir
05:25
you know like go after much more ambitious projects.
Chen Zamir
Chen Zamir
05:28
And fraud prevention was always an and still is an ambitious project.
Chen Zamir
Chen Zamir
05:33
I really wanted to you know bring you uh to the pod and
Chen Zamir
Chen Zamir
05:36
discuss you know where where is the industry going?
Chen Zamir
Chen Zamir
05:40
And are we going to see this shift?
Chen Zamir
Chen Zamir
05:42
So I'll start by asking like what's your view on
Chen Zamir
Chen Zamir
05:45
this change, and and and how do you see it?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:00
The change is significant I I would say.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:02
During the uh new company's ideation process I
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:05
talked to uh many people from many companies.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:08
And the common thread was that everyone one has
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:11
pressure or desire or motivation to build.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:14
Whether to uh than to buy.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:16
It started as an external pressure I believe way before
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:20
AI was mature enough to actually perform these tasks.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:23
But now the models have uh improved so much that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:26
some of these things are actually useful.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:29
Um so I I think that um there's vendor fatigue.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:32
That's one thing uh that happened you know um.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:35
And together with these AI capabilities people
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:38
are starting to ask can I build it myself?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:41
Can I build part of it myself?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:43
Uh and I think everyone is struggling with this.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:46
Some of the most more successful attempted that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:49
I've seen are uh mostly isolated I would say.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:52
Usually a very strong team member did something
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:56
uh that only he or she could use.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
06:58
Maybe some other power power users.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:00
But it mostly improves or reduces the
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:03
handling time uh of different tasks.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:05
Rather than uh replace a complete system.
Chen Zamir
Chen Zamir
07:43
Yes.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:43
Um so I think that's where the the market is is at right now.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:47
Um you know that in in the past vendors were
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:50
uh offering the the complete package.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:52
To this day many many of the vendors in the
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:55
space are offering an end to end solution.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
07:58
Um which was mandatory uh until uh recently.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:01
Uh give me a decision, give me a guarantee, give
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:04
me a a score, give me something that I could offload
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:07
the work to and and make a decision upon.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:10
Um, and today what I'm hearing from companies is no,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:13
we we we didn't have the expertise nor the bandwidth
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:17
to consume anything other than that in the past. But now we can.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:21
If you give us something more granular, we can handle it.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:25
Uh, and we can uh uh actually help it will
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
08:28
help us customize what we have in house.
Chen Zamir
Chen Zamir
08:51
Yeah, it's uh it's I I I see it as well where
Chen Zamir
Chen Zamir
08:54
there's a I don't want to say confusion.
Chen Zamir
Chen Zamir
08:56
Because it's not the right word, but I think the market
Chen Zamir
Chen Zamir
09:00
is still adjusting to understand who is supplying the
Chen Zamir
Chen Zamir
09:03
data and who's supplying the decisioning platform.
Chen Zamir
Chen Zamir
09:07
Um and I don't think, I I I think it used
Chen Zamir
Chen Zamir
09:09
to be relatively straightforward.
Chen Zamir
Chen Zamir
09:12
Where you know like uh uh businesses would supply their
Chen Zamir
Chen Zamir
09:15
own data, vendors would enrich it with their own kind of
Chen Zamir
Chen Zamir
09:19
like network data their their specialized features.
Chen Zamir
Chen Zamir
09:22
And provide the entire kind of like you know
Chen Zamir
Chen Zamir
09:25
like analytical uh decision layer.
Chen Zamir
Chen Zamir
09:28
Whether this is ML, whether this is a rule engine combination.
Chen Zamir
Chen Zamir
09:32
Um and today things are much more fluid and I would like I think
Chen Zamir
Chen Zamir
09:36
as you also said uh I also see the breakdown not between
Chen Zamir
Chen Zamir
09:40
like data and platform but sometimes between use cases.
Chen Zamir
Chen Zamir
09:43
Some use cases you know are maybe easier
Chen Zamir
Chen Zamir
09:46
to build, some use cases are less easy.
Chen Zamir
Chen Zamir
09:49
Um and so I see that there is like I would say a lack of uh real, I
Chen Zamir
Chen Zamir
09:53
would say doctrine or guidelines behind like for practitioners.
Chen Zamir
Chen Zamir
09:57
Uh living vendors aside, um as to how to approach it.
Chen Zamir
Chen Zamir
10:01
And how to decide whether to build or buy.
Chen Zamir
Chen Zamir
10:04
Uh and I think this is like in our conversation I'm guessing at the
Chen Zamir
Chen Zamir
10:08
end of it uh I'd like to for us to kind of like maybe suggest some
Chen Zamir
Chen Zamir
10:13
of these concepts for how to to manage these decisions now and
Chen Zamir
Chen Zamir
10:17
try to understand like what was the role that vendors played.
Chen Zamir
Chen Zamir
10:21
Uh from a practitioner's perspective in their platform,
Chen Zamir
Chen Zamir
10:24
and how do you think that has evolved?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
10:45
So as I said before.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
10:46
I think that in the past uh vendors were expected
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
10:49
to provide a wholesome solution.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
10:51
Uh an end to end solution that is uh I I would say a kind of a
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
10:55
replacement to old school uh uh consultancy or contractual work.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:00
Um just you know provide me with a decision or provide
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:03
me with a score and I take it from there.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:06
Depends on the companies or the the the customer's maturity
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:10
level and uh what they already have uh in place.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:13
Today when I speak to uh potential customers they're uh
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:17
looking more at network effects and network data.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:20
That's something that has become very important like I'm
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:24
using this vendor because I I I don't have the I can zoom
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:28
out and see where this customer has been before.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:31
Uh they came to to me, uh for things like device intelligence
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:35
for very uh specific use cases that uh are now more I would
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:39
more interesting or more important to to customers.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:42
Especially the you know the midsized and and large customers.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:46
Uh and I think that still similar to the smaller ones,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:50
like small customers still use vendors in very much the
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:54
same way it has happened over the last few years.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
11:57
They still want this assurance, they still want the expertise,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
12:01
they still want the the the the capabilities that they offer.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
12:05
But the the bigger they are, the harder they look at which
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
12:09
specific uh end points could actually bring me value.
Chen Zamir
Chen Zamir
12:48
I agree with everything that you said.
Chen Zamir
Chen Zamir
12:50
Um but you you kind of like relate uh uh mainly to the backend
Chen Zamir
Chen Zamir
12:54
capabilities, the technological or analytical capabilities that vendors offer.
Chen Zamir
Chen Zamir
12:59
But uh some of what vendors give you as a as a as
Chen Zamir
Chen Zamir
13:03
a team as a practitioner team is uh is just UI.
Chen Zamir
Chen Zamir
13:06
And if you look at vendors and how like the more legacy
Chen Zamir
Chen Zamir
13:09
one started, um you see a pure SaaS UI plays, right?
Chen Zamir
Chen Zamir
13:13
Uh a cockpit to allow you to do manual review, a rule engine that
Chen Zamir
Chen Zamir
13:17
allows you to uh codify your heuristics uh in a like not so uh
Chen Zamir
Chen Zamir
13:21
technical environment like an SQL or or even like hardcoded uh uh rule engine.
Chen Zamir
Chen Zamir
13:27
Um how do you see these more UI elements and
Chen Zamir
Chen Zamir
13:29
how practitioner teams look at them today?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
13:49
Um well let's separate between viewers
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
13:51
and uh actionable uh products. Okay.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
13:53
So I think that viewers uh uh what to refer to as a
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
13:57
cockpit or you know manual review tool, uh more and
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:00
more companies are starting to build their own.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:03
Uh and not because it necessarily looks better or performs
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:07
better, but it because it's more customizable.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:10
Uh many customers have expressed frustration from the fact that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:15
they want to change something, and then they have to ask, and wait
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:19
for it to get prioritized on the vendor's end, and so on.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:23
So um and it's much easier to to codify using uh bite coding
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:27
or you know the coding is it's kind of obsolete on its own.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:31
I think that it's actual coding these days.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:34
Um So I I so so this is something that companies have
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:37
have already moved to sometimes it's a mess.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:40
Like you know different team members have different
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:43
viewers, each one changing tweaking their own.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:46
It's not necessarily a good thing it makes it harder to onboard
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:51
new members, uh to share results, to work together.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:54
But um it is the direction that everyone one is going.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
14:58
To things like rule engines for example are much more uh difficult.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:02
I would say sometimes they're stuck deep within legacy systems.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:06
Uh and I do think that companies need to untangle
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:10
it and eventually build their own.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:12
Like at Simplex we had our own Python based rule
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:15
engine from the get-go uh 12 years ago, so.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:18
But but it proved to be pivotal because then you can
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:21
be much more flexible and build different strategies
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:25
and build auto switches that uh can uh switch between
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:28
different strategies on certain uh conditions.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:31
And uh uh you know the sky the sky is the limit.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:35
So uh, there I don't see uh the same movement yet.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:38
I think it usually it it is perceived as part of the system,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:42
for practitioners, unless the company you know as a as a
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:46
overall strategy decides to overhaul the the system.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:49
And uh the the entire thing and then uh the engineering team will
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
15:54
will actually uh will lean towards a proprietary solution.
Chen Zamir
Chen Zamir
16:50
Yeah, I think it's an interesting point because
Chen Zamir
Chen Zamir
16:53
like I I I I like how you separate it.
Chen Zamir
Chen Zamir
16:55
Like viewers, Yeah, I mean basically you just pull
Chen Zamir
Chen Zamir
16:58
data from a data mart and you present it.
Chen Zamir
Chen Zamir
17:01
But a rule engine more than it is a UI like tool or UI uh uh product.
Chen Zamir
Chen Zamir
17:06
It's a you know it's it's a it's a backend engineering module that
Chen Zamir
Chen Zamir
17:10
needs to be real time, needs to be like high availability, needs to
Chen Zamir
Chen Zamir
17:15
like know how to access the data schema in a in an efficient way.
Chen Zamir
Chen Zamir
17:19
Uh needs testing and simulation capabilities.
Chen Zamir
Chen Zamir
17:22
Um and probably needs also to interface with a lot of other systems
Chen Zamir
Chen Zamir
17:26
as well, including the models the viewers themselves.
Chen Zamir
Chen Zamir
17:30
So yeah I I I I I get what you were saying and I also
Chen Zamir
Chen Zamir
17:34
I I like the the distinction that you made.
Chen Zamir
Chen Zamir
17:36
Uh that that definitely clarifies uh the RNRs.
Chen Zamir
Chen Zamir
17:39
Um I'm I'm I'm asking myself, like I don't
Chen Zamir
Chen Zamir
17:42
know how much uh how much you say it.
Chen Zamir
Chen Zamir
17:45
But when practitioners are thinking about
Chen Zamir
Chen Zamir
17:47
building, where do teams usually go?
Chen Zamir
Chen Zamir
17:50
Uh is it more to replace current capabilities that they are
Chen Zamir
Chen Zamir
17:54
you know receiving at the moment uh from their vendors,
Chen Zamir
Chen Zamir
17:57
or could receive from a vendor if they if they buy it?
Chen Zamir
Chen Zamir
18:01
Or is it more to cover some gaps that vendors just never gotten to?
Chen Zamir
Chen Zamir
18:06
Uh that or maybe because they couldn't.
Chen Zamir
Chen Zamir
18:08
Like is it new capabilities that they never had, um or
Chen Zamir
Chen Zamir
18:12
is it really to replace capabilities from vendors?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
18:36
I could say unfortunately uh companies
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
18:38
are starting with the manual task.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
18:40
Uh and how they can automate uh whatever specialists
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
18:44
or analysts have done in the past.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
18:46
Uh so you might call it uh new capabilities because it it wasn't
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
18:50
possible to automate some of these uh things in the past.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
18:54
But I think that uh
Chen Zamir
Chen Zamir
19:01
Can you give an example just so so like what
Chen Zamir
Chen Zamir
19:03
do you mean by an analytical process?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:06
I mean uh uh reviewing a payment um going and you
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:09
know logging into different data sources.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:12
Or you know things that need pulling different
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:15
uh files, screenshots, whatever.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:17
Um in order to make a decision or to
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:19
to document a decision or audit.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:21
So that's something that has already gone agentic to some extent.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:26
Um, I treat it as new capabilities because it hadn't added a new
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:30
layer of data, but it allowed us to automate this layer of data.
Chen Zamir
Chen Zamir
19:48
Mhm.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:50
Uh other than that, I see many companies don't necessarily want
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:54
to replace vendors, whether they're afraid to do so or they
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
19:58
need these uh unique uh capabilities that we've discussed,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:02
like network data, or you know data is is king today.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:05
Uh but in terms of uh algorithms, maybe
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:08
I can build a better one for me.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:10
Not necessarily a better one than the vendor's algorithm,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:14
but maybe it suited it suits my business,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:16
my way of thought, my uh risk appetite.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:19
So uh most of the companies try to blend in right now.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:22
Try to see okay maybe I have different segments
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:26
or pockets where I don't need this the vendor.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:29
The classic example is returning users.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:31
Like I already know something about this user.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:34
I have to verify, it's not account take over, but maybe
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:38
I can do it, maybe I can do it in another way, maybe
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:42
with another vendor with different economics.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:45
Uh but the blanket solution is already starting
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
20:48
to you know to unravel at this point.
Chen Zamir
Chen Zamir
21:15
I like this example and and the algorithms
Chen Zamir
Chen Zamir
21:17
um is definitely a juicy point.
Chen Zamir
Chen Zamir
21:19
Because this is like a very want to believe a very kind of like
Chen Zamir
Chen Zamir
21:24
core rudimentary capability that everybody can relate to.
Chen Zamir
Chen Zamir
21:27
When you buy a rule engine from a vendor, that's what you buy.
Chen Zamir
Chen Zamir
21:32
You you buy a rule engine, you don't buy the rules, you
Chen Zamir
Chen Zamir
21:35
don't buy the capability for producing your rules.
Chen Zamir
Chen Zamir
21:39
Now, you might, I don't know, some vendors might give you like managed
Chen Zamir
Chen Zamir
21:43
services or you might have in-house uh an in-house team that does
Chen Zamir
Chen Zamir
21:48
it or uh a consultant a consultant team that does that.
Chen Zamir
Chen Zamir
21:51
But the point is that the the product
Chen Zamir
Chen Zamir
21:54
that you buy, is the engine itself.
Chen Zamir
Chen Zamir
21:56
And researching rules, monitoring rules, fixing, tweaking,
Chen Zamir
Chen Zamir
22:00
deprecating rules, all of these are processes um that
Chen Zamir
Chen Zamir
22:03
a require expertise, b require access to data.
Chen Zamir
Chen Zamir
22:06
But I'm trying to understand whether teams would
Chen Zamir
Chen Zamir
22:10
that be the next thing that they go after?
Chen Zamir
Chen Zamir
22:12
Like being able to conduct these analytical processes,
Chen Zamir
Chen Zamir
22:16
inhouse, because no one is offering them.
Chen Zamir
Chen Zamir
22:19
Or would they really go after replacing the
Chen Zamir
Chen Zamir
22:22
the kind of like the software itself?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:34
I think that most if not all practitioners today are being
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:37
assisted by AI to make these uh uh these decisions or
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:41
these processes, even if it's an offline process.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:44
Like consulting with my favorite uh chat uh and
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:47
coming back with the the rule syntax or the decision
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:51
or the policy or the tweak or whatever.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:53
Um so the market is kind of ready for it.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
22:56
Um I think that there's a tension for for employees between
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:00
uh building these uh the systems that uh replace them rather
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:04
than being assisted by them and getting time back.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:07
So I don't see a rush to completely replace
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:10
uh this decisioning system. That's the truth.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:13
But but I would say that what what I do see happen is
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:17
uh what we see happen everywhere, which is AI slop.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:20
Um and it already happens in fraud and risk as well.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:24
Uh multiple rules, tons of them that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:26
you couldn't produce in the past.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:28
Tons of models which was unheard of in the past.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:31
Like uh you see companies that you know that have a
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:35
new fraud problem, and suddenly 20 new models uh are
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:38
built overnight and and released to production.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:41
Uh which creates latency problems, which most aren't aware of.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:46
Like vendors take so much time uh trying to optimize this
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:49
and uh that's something that is usually not on the practitioners
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
23:54
or the data uh data scientists uh radar at all.
Chen Zamir
Chen Zamir
24:32
Yeah, I mean also monitoring drift across 20 models is a
Chen Zamir
Chen Zamir
24:35
different task than doing it for one to three models.
Ad Break
24:40
Hey folks, I want to take a quick break
Ad Break
24:41
to speak about today's sponsor, me.
Ad Break
24:43
If you're finding this useful, do me a small favor, like and subscribe.
Ad Break
24:47
It really helps with the algorithm.
Ad Break
24:49
And if you're not already on my newsletter, The Saturday
Ad Break
24:51
Fraud Strategist, you should definitely check it out.
Ad Break
24:54
Every Saturday, I break down fraud trends, real cases, and strategies
Ad Break
24:58
from my own experience of building and operating fraud prevention systems.
Ad Break
25:01
Whether you're new to the space or a system practitioner,
Ad Break
25:04
I'm sure you'll find it interesting. Check the link in the description.
Ad Break
25:08
Takes 2 seconds. You'll thank yourself next Saturday.
Ad Break
25:10
Now, back to the video.
Chen Zamir
Chen Zamir
25:12
This is really like where I'm where my head
Chen Zamir
Chen Zamir
25:14
is at in the last couple of weeks.
Chen Zamir
Chen Zamir
25:17
Uh both because we had um you know a lot of uh PR.
Chen Zamir
Chen Zamir
25:20
Or no I don't want to call it PR because it's like it it is research
Chen Zamir
Chen Zamir
25:25
right um around foundation models for financial risk.
Chen Zamir
Chen Zamir
25:28
And you know we've seen I mean of course at Sardine we
Chen Zamir
Chen Zamir
25:32
had uh we've published one Revolute published a very interesting
Chen Zamir
Chen Zamir
25:36
uh research uh Plaid had um released one.
Chen Zamir
Chen Zamir
25:39
Uh I'm sure that I'm forgetting a couple of others.
Chen Zamir
Chen Zamir
25:42
Uh but I I would say these are like the three kind
Chen Zamir
Chen Zamir
25:45
of like main cases that that come to my mind.
Chen Zamir
Chen Zamir
25:48
And one of the things that I'm like I'm thinking about is
Chen Zamir
Chen Zamir
25:52
that to me one of the main things that foundation models do,
Chen Zamir
Chen Zamir
25:56
and I think maybe how they can be a threat to vendors, is
Chen Zamir
Chen Zamir
26:00
the fact that they replace the feature layer right.
Chen Zamir
Chen Zamir
26:03
I mean the the advantage of vendors uh traditionally
Chen Zamir
Chen Zamir
26:07
was that they are specialists.
Chen Zamir
Chen Zamir
26:09
They they are experts in fraud prevention.
Chen Zamir
Chen Zamir
26:12
And that spec specialization doesn't necessarily
Chen Zamir
Chen Zamir
26:15
um manifest as better algorithms.
Chen Zamir
Chen Zamir
26:17
It manifests as better data and better data means
Chen Zamir
Chen Zamir
26:20
better features in how you calculate velocity.
Chen Zamir
Chen Zamir
26:23
You know the simplest example uh in how you do uh IP piercing,
Chen Zamir
Chen Zamir
26:28
in how you do device ID, but also like how do you
Chen Zamir
Chen Zamir
26:31
how do you um look at identity cohesiveness.
Chen Zamir
Chen Zamir
26:34
How do you look at you know boguseness, um or synthetic ID.
Chen Zamir
Chen Zamir
26:38
So there are a lot of like features like that are
Chen Zamir
Chen Zamir
26:41
very complex and that need a lot of research.
Chen Zamir
Chen Zamir
26:44
And so far they have been handcrafted by specialists
Chen Zamir
Chen Zamir
26:47
mostly on the vendor side mostly not only.
Chen Zamir
Chen Zamir
26:50
Now you have foundation models that build features for free.
Chen Zamir
Chen Zamir
26:54
Uh no one understands what they mean, but
Chen Zamir
Chen Zamir
26:57
they are there and they are free.
Chen Zamir
Chen Zamir
26:59
Um so is that like are we going to see foundation models replacing,
Chen Zamir
Chen Zamir
27:04
not vendors, but maybe the need or the let's say the the gap.
Chen Zamir
Chen Zamir
27:08
Uh maybe bridging the gap in terms of uh SME, or
Chen Zamir
Chen Zamir
27:11
kind of like fraud knowledge domain knowledge, that
Chen Zamir
Chen Zamir
27:14
sometimes practitioner uh teams don't have.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
27:57
That's a great question.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
27:58
I remember that two years ago we discussed uh whether
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:02
LLMs could uh replace fraud models entirely.
Chen Zamir
Chen Zamir
28:06
Yes. And by the way I must say you you reminded me this conversation.
Chen Zamir
Chen Zamir
28:09
You, I must say that you said I don't know if the
Chen Zamir
Chen Zamir
28:12
scoring part but probably the features part.
Chen Zamir
Chen Zamir
28:14
And I said yeah I don't know what you mean by that.
Chen Zamir
Chen Zamir
28:17
But two years later now I know what you mean by that. Oh yeah,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:21
I think that actually the features layer was replaced not by uh
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:25
transaction foundation models necessarily, but by standard LLMs.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:29
Like if you go and ask an LLM this is my data set which features
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:33
I should derive from, you'll have I I dare say 80% of the value
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:37
that would be created by a team of professionals these days.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:41
And uh I think that in a year or so, the some models would
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:45
be powerful enough to provide 95% of the value.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:48
Um which makes it a question of economics at this point.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:52
And how you actually you know get the data engineer these
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
28:56
features but that's that's easier, that's even cheaper.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:00
Um I think that the value of foundation models in the space is
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:04
not necessarily that they saved the feature engineering, but
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:08
it's basically a highly advanced anomaly detection engine.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:12
The ones the one that we always dreamed of.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:15
It's uh we always tried, and I'm sure you tried it in
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:18
the past in your company as well, to to bring unsupervised
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:22
learning to the level of supervised learning.
Chen Zamir
Chen Zamir
29:43
Yes. Yes.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:44
Uh and uh now it's almost doable, if you have enough data.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:47
Um uh whether this new transaction without any other context
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:51
or labels, makes sense within uh uh this time series.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
29:55
Uh so I I would say that's the the real value.
Chen Zamir
Chen Zamir
30:08
Yeah.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:09
Um I don't know where it will go.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:11
I I I think that the real uh uh promise, or threat
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:14
depends on which side of the fence you're on,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:17
is the new uh semideterministic models like
Chen Zamir
Chen Zamir
30:25
Yes. Yes. Yes. Yes. Yes. That was my next point. Yes. Okay. Go go go.
Chen Zamir
Chen Zamir
30:29
Yeah. Yeah. Please.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:31
That's, I, when we had the discussion on whether LLMs could replace
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:35
fraud models, this is the was this was the missing piece.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:39
Um a model that is actually um dedicated to a decision, uh that is
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:43
forced to make a decision at the end but uses an LLM to to make
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:47
judgment, uh will probably at some point will be uh the new XG
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:52
boost or the new uh gold standard for uh fraud models I would say.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:56
That's uh imagine that you can, you know we grew
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
30:59
up on the fraud sciences story based approach.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
31:02
Uh and then we kind of moved from it as a industry,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
31:06
but maybe we'll go back to stories.
Chen Zamir
Chen Zamir
31:26
Well I I think I think we can spend like
Chen Zamir
Chen Zamir
31:28
three hours discussing this topic.
Chen Zamir
Chen Zamir
31:30
Uh because there there are so many aspects to it.
Chen Zamir
Chen Zamir
31:34
Um, I'll say first regarding foundation models,
Chen Zamir
Chen Zamir
31:37
I I I I I completely agree with what you say.
Chen Zamir
Chen Zamir
31:40
I'll just say that the anomaly detection part is re like as far as I've
Chen Zamir
Chen Zamir
31:45
seen so far, Um, and we've seen it again both in the Sardine research,
Chen Zamir
Chen Zamir
31:49
in Revolute research, and it in Plaid's research, it's an anomaly
Chen Zamir
Chen Zamir
31:54
detection within a known entity, an entity that I have history on.
Chen Zamir
Chen Zamir
31:58
Um and whether the new event, you know, makes sense is consistent
Chen Zamir
Chen Zamir
32:02
uh with how this entity usually acts and behaves.
Chen Zamir
Chen Zamir
32:06
I haven't seen it yet on, you know, like I I would
Chen Zamir
Chen Zamir
32:09
call it like horizontal horizontal data problems.
Chen Zamir
Chen Zamir
32:12
Um an acquirer, right, or a merchant that doesn't
Chen Zamir
Chen Zamir
32:15
necessarily have accounts, right?
Chen Zamir
Chen Zamir
32:18
So, we just see users coming in and and transacting.
Chen Zamir
Chen Zamir
32:21
And then anomaly detection would be much more not on how a single user
Chen Zamir
Chen Zamir
32:26
uh acts, but how you know like a cluster of users uh that might be
Chen Zamir
Chen Zamir
32:30
relatively small within the context of my volume, how they behave.
Chen Zamir
Chen Zamir
32:35
Uh but yeah I I I definitely agree with your with your
Chen Zamir
Chen Zamir
32:38
point that for the first use case, uh foundation models
Chen Zamir
Chen Zamir
32:42
are extremely powerful, extremely powerful.
Chen Zamir
Chen Zamir
32:45
As as as we go to the kind of like decision models right,
Chen Zamir
Chen Zamir
32:49
that's uh that's an interesting naming convention, um I think
Chen Zamir
Chen Zamir
32:53
what is interesting for me is not only the fact that they
Chen Zamir
Chen Zamir
32:56
are much more fit uh like fit for purpose here.
Chen Zamir
Chen Zamir
33:00
And and this is not only their accuracy,
Chen Zamir
Chen Zamir
33:02
this is also the latency, right?
Chen Zamir
Chen Zamir
33:04
And the cost, uh which to me were always kind
Chen Zamir
Chen Zamir
33:07
of like the the the real pain points, right?
Chen Zamir
Chen Zamir
33:10
Even with foundation models, foundation models
Chen Zamir
Chen Zamir
33:13
don't do the scoring themselves.
Chen Zamir
Chen Zamir
33:15
And a lot of what you want to do there needs to happen offline
Chen Zamir
Chen Zamir
33:20
because you don't necessarily you can't necessarily
Chen Zamir
Chen Zamir
33:23
run all of these anomaly detection in real time.
Chen Zamir
Chen Zamir
33:26
And the session models solve all of that.
Chen Zamir
Chen Zamir
33:29
But I think what came to my mind in the last week or so
Chen Zamir
Chen Zamir
33:33
is that actually the most powerful thing that decision
Chen Zamir
Chen Zamir
33:36
models give you, is that anyone can roll them out.
Chen Zamir
Chen Zamir
33:40
You don't need to have a data science team.
Chen Zamir
Chen Zamir
33:42
You don't need to have a data science infrastructure to run
Chen Zamir
Chen Zamir
33:46
like all of these simulations and all of this data.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
34:01
I'm not sure you you even need data the training data. I mean,
Chen Zamir
Chen Zamir
34:05
You can open a a a shop on Shopify tomorrow.
Chen Zamir
Chen Zamir
34:07
And as a as as a as the seller, like as a single person
Chen Zamir
Chen Zamir
34:11
seller, you can deploy a pretty decent real time fraud
Chen Zamir
Chen Zamir
34:15
model uh that doesn't cost you almost anything.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
34:30
Yes
Chen Zamir
Chen Zamir
34:31
It's it's it's pretty it's pretty insane.
Chen Zamir
Chen Zamir
34:33
So I'm like I'm thinking what, yes vendors would probably
Chen Zamir
Chen Zamir
34:37
be the like first ones to capitalize on this.
Chen Zamir
Chen Zamir
34:40
But I'm trying to think like a year or
Chen Zamir
Chen Zamir
34:43
two ahead, how would that change?
Chen Zamir
Chen Zamir
34:45
How practitioner teams, uh build internally
Chen Zamir
Chen Zamir
34:48
and what they buy externally.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
34:52
In order for that to happen, it will need to
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
34:54
be assimulated by the large LLM providers.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
34:57
And it already happens like OpenAI released something
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:01
very similar, a decision model of their own.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:04
I assume that entropic will do the same.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:06
And then when you write to ChatGPT solve my fraud
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:10
problem, it will probably create uh this new uh decision
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:13
model endpoint, and say here run this.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:16
Uh or it's already baked into your app.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:18
Um and I think that will happen s sooner rather than later.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:22
And then it depends what you fill it with.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:25
Uh and to my earlier point, uh vendors are um you know it
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:29
will crystallize the importance of external data points.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:33
Um and large uh data pools and network data.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:36
Um it will also uh require more feature stores of the sort.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:40
Because um let's say that you want to include the history
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:43
of a you know a lengthy history of a merchant or a customer
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:47
within these models, uh that obviously the SLA will
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:51
be affected, or it will be just too many tokens.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:54
Uh so the economics won't work.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:56
So you need to pre-calculate uh like we're
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
35:59
doing for the past two decades.
Chen Zamir
Chen Zamir
36:34
Yes.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
36:35
Uh but in in a more efficient and um I would say LLM friendly way.
Chen Zamir
Chen Zamir
36:41
I uh yeah I I I tend to agree.
Chen Zamir
Chen Zamir
36:43
Maybe I I would say that uh you know maybe maybe I was
Chen Zamir
Chen Zamir
36:46
very, it sounded uh very uh kind of like um deterministic,
Chen Zamir
Chen Zamir
36:50
uh and and and convinced what I said about Jev.
Chen Zamir
Chen Zamir
36:53
It's indeed the experience that I've uh seen, um you know were
Chen Zamir
Chen Zamir
36:57
based on data that you know like enriched data right.
Chen Zamir
Chen Zamir
37:01
Um you don't necessarily need to train it right so you
Chen Zamir
Chen Zamir
37:04
don't need like the like millions of consortium rows.
Chen Zamir
Chen Zamir
37:08
Um but you like I think the feature, uh sorry the
Chen Zamir
Chen Zamir
37:11
features themselves, like basically being able to
Chen Zamir
Chen Zamir
37:14
extend uh the data that you have per event.
Chen Zamir
Chen Zamir
37:17
That is probably that that would become the the
Chen Zamir
Chen Zamir
37:20
main point here, the main performance driver.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
37:36
But but I would al also say that um it's it's still not enough.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
37:40
It might might be enough for this uh Shopify uh merchant.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
37:44
But for more more complex uh multi-party um businesses,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
37:47
uh decisioning is still a challenge.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
37:50
Let's say you have a perfect decision for every merchant,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
37:53
every transaction, every uh onboarding. Uh what do you do with it?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
37:58
Uh it's the same of, you know, getting the ingredients to
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:02
a chef's meal and trying to cooking it on your own.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:05
You won't get get the same uh results
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:07
even if you have AI assisting you.
Chen Zamir
Chen Zamir
38:25
Yes.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:25
Uh on the side. So
Chen Zamir
Chen Zamir
38:27
What what what why do you think that's the case?
Chen Zamir
Chen Zamir
38:30
Like what what would the chef, in this example,
Chen Zamir
Chen Zamir
38:33
uh have uh that that you wouldn't necessarily?
Chen Zamir
Chen Zamir
38:36
Is it the interoperability between like the different
Chen Zamir
Chen Zamir
38:39
components of the of the platform? How like what is it?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:44
The know how of what to do in case things go unexpectedly.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:47
Um um that's the to to to chef's point.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:50
What happens if it boils a minute before it's supposed to boil?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:54
What do I do?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:55
What if I uh to to to uh compare our uh industry uh what do I do
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
38:59
if the charge back rate is higher than expected this month?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:03
What do I do if I have a fraud attack that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:06
wasn't expected in the simulation?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:08
Because I I it's the first to to the point of you know building
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:13
a fraud prevention system without any expertise.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:16
This is the first time I've ever experienced it.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:19
What do I do with it?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:20
Um, how do I perform um stepups or challenges?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:24
Um, how do I retrain? How do I monitor drift?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:27
Uh, how do I ultimately make a decision
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:29
that makes sense for my business?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:31
I think that's the next uh phase that everyone that tries
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:35
to uh build these uh decision models or computation models
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
39:39
depends of the size will have to go through.
Chen Zamir
Chen Zamir
40:09
So I I I tend to agree that I think there are two elements here, right?
Chen Zamir
Chen Zamir
40:13
One is is just the expertise itself, and that is probably something
Chen Zamir
Chen Zamir
40:18
that vendors would anyways not be able to uh to impart on you.
Chen Zamir
Chen Zamir
40:22
Uh but one of that is just you know generally
Chen Zamir
Chen Zamir
40:25
speaking like a high engineering bar. Uh and to make sure that, Yeah.
Chen Zamir
Chen Zamir
40:29
And and and I think what vendors have here that you as a practitioner
Chen Zamir
Chen Zamir
40:34
would never have, they they have a portfolio of clients.
Chen Zamir
Chen Zamir
40:38
They've seen a lot of things go wrong uh both mechanically,
Chen Zamir
Chen Zamir
40:42
like technically in the system, but also uh in terms of
Chen Zamir
Chen Zamir
40:45
you know like weird fraud patterns and and so on.
Chen Zamir
Chen Zamir
40:49
And so they probably already smoothed out a lot of wrinkles in
Chen Zamir
Chen Zamir
40:53
their system that it will take you like years of bumping into
Chen Zamir
Chen Zamir
40:57
them again again until you understand, um a what you need to
Chen Zamir
Chen Zamir
41:01
do and b that you also like you get the uh the resource.
Chen Zamir
Chen Zamir
41:05
Because usually these things are not like uh easy to fix.
Chen Zamir
Chen Zamir
41:08
It's usually like as I said it's like there's a high standard
Chen Zamir
Chen Zamir
41:12
engineering effort that is required and it's not something that
Chen Zamir
Chen Zamir
41:17
you can like easily easily conduct on your own many times.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:30
Um definitely.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:31
Um you know to be the devil's advocate, at some point
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:34
LLM will absorb some of it and we'll figure it out.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:37
I think that we're, we don't want to sound
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:40
like uh designers from two years ago.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:43
That said no that LLM can do my my job.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:45
So I I I do think I I do want to be
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:48
you know uh transparent about it.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:50
But yeah, I think that eventually uh you understand your business.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:54
Even if it's you know your Claude agent that understands your
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
41:58
business at some point and have the contact has the context.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
42:02
But um but it means that you can't uh
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
42:05
take a one-size-fits-all solution.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
42:07
You need to figure out uh your path and how you do things.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
42:11
And u whether it will work in terms
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
42:13
of latency or not that we'll see.
Chen Zamir
Chen Zamir
42:37
Yeah, I agree.
Chen Zamir
Chen Zamir
42:38
So we mentioned uh different use cases, different components,
Chen Zamir
Chen Zamir
42:42
different uh views on each component. Are we looking at the UI?
Chen Zamir
Chen Zamir
42:46
Are we looking at the analytical components?
Chen Zamir
Chen Zamir
42:49
Are we looking at the uh the software itself and the
Chen Zamir
Chen Zamir
42:52
architecture like from a technical perspective?
Chen Zamir
Chen Zamir
42:55
Let's try and kind of like now package it to a more
Chen Zamir
Chen Zamir
42:59
coherent um I don't want to say guidelines, because
Chen Zamir
Chen Zamir
43:02
I I don't think that we can be pres prescriptive.
Chen Zamir
Chen Zamir
43:05
Uh I mean we want to be prescriptive but I
Chen Zamir
Chen Zamir
43:08
think I think it's too early to do that.
Chen Zamir
Chen Zamir
43:11
But how do you suggest teams today, uh that
Chen Zamir
Chen Zamir
43:14
are thinking about building more.
Chen Zamir
Chen Zamir
43:16
I think everybody is building already,
Chen Zamir
Chen Zamir
43:18
they're thinking about building more.
Chen Zamir
Chen Zamir
43:21
What would be In your mind, the first few areas?
Chen Zamir
Chen Zamir
43:24
You would say to them, look, you you want to probably look
Chen Zamir
Chen Zamir
43:28
uh around these areas, and maybe where not to look at.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
43:41
One thing that we constantly see is uh um
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
43:43
everything related to segmentation.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
43:46
Um I think uh many many of the many customers
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
43:49
miss the importance of segmentation.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
43:51
And the first thing I would advise them is try to
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
43:54
think of the different segments that behave completely
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
43:58
uh differently within your uh system.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:00
The obvious example is new versus returning user,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:04
or any tenure based uh uh segmentation.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:06
But there are so many of them you know. Geo products, so on.
Chen Zamir
Chen Zamir
44:26
Payment methods, business versus uh consumer accounts.
Chen Zamir
Chen Zamir
44:29
Yeah there there are so many yeah
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:31
Try to find pockets that are either underperforming, or are
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:34
uh the economics don't work as well as in other segments.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:38
And try to think of, okay what can I do here.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:41
Uh um maybe I can not necessarily build it in house, but maybe
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:45
change uh something maybe uh experiment with other vendors, with
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:50
AI, with both, with uh custom solutions, with hiring someone, whatever.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:54
But that's something that many are are missing.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
44:57
I would also uh suggest to look at uh um how pivotal
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:01
is risk and fraud for your business. Let's say you're a merchant.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:05
Fraud is very painful.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:07
And for an e-commerce merchant, it could determine the fate of the
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:11
business, but it's it's still a sub function, I would say.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:15
Uh the growth usually comes from other places.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:18
And for all these table stakes you have to take care
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:22
of it and increase the approval rate, and so on.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:25
That opposed to other businesses for example fintechs uh
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:29
that fraud and risk are are very uh core uh functions.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:32
Um and if you're offloading a core function Ask
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:35
yourself, is this the right thing to do?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:38
Maybe it is, maybe it's not, maybe
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:40
it's right for some of the cases.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:42
And the best way to do it practically is to look at the road map
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:47
from the last 6 months or 12 months, and ask yourself how many
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:51
things were either delayed or deprioritized because it would
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:55
be too difficult to perform given our current risk setup.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
45:59
You have to have this flexibility as
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
46:01
a as a business in order to grow.
Chen Zamir
Chen Zamir
46:48
And and so for for for such cases for such businesses,
Chen Zamir
Chen Zamir
46:51
you would suggest to build more or to buy more?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
46:58
Today, to build more. Definitely.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:00
Uh to build the foundations uh whatever it is you
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:03
you offloaded uh to a vendor in the past.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:06
Try to think of the atomic units of the decision
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:09
that needs to be made, and whether you can produce
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:12
these atomic units yourselves or buy them.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:15
Uh and build everything around it.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:17
And have full flexibility, more complexity definitely,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:21
u but assuming that we see the trajectory of AI uh for
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:25
coding and where it's headed, assuming you can handle
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:28
it with roughly the same headcount in the future.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:31
Uh increase the complexity because it will provide you
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:35
more flexibility, more growth opportunities uh uh and
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
47:38
you can cater the business needs uh much better.
Chen Zamir
Chen Zamir
48:01
What do you what do you, from a team size or not team size but from
Chen Zamir
Chen Zamir
48:05
a team perspective, whether it's the size whether it is the domain
Chen Zamir
Chen Zamir
48:09
expertise whether it is the talent the skill set that you have.
Chen Zamir
Chen Zamir
48:14
What do you think are the prerequisites for a team to
Chen Zamir
Chen Zamir
48:17
actually start building in this you know this space?
Chen Zamir
Chen Zamir
48:21
Which is very you know it is very demanding uh both
Chen Zamir
Chen Zamir
48:24
technically and from a from uh theme perspective.
Chen Zamir
Chen Zamir
48:27
Um what are the prerequisites?
Chen Zamir
Chen Zamir
48:29
What like how should a leader look at their team and how should
Chen Zamir
Chen Zamir
48:33
they know if they are ready to start building themselves?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
48:48
Most of the risk leaders that I know have uh
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
48:50
very high trust in their uh in their team.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
48:53
Uh and I know it's it's sometimes a blind spot, but I would
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
48:57
actually advise to look elsewhere to look at the organization and
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:01
ask yourself, do does the organization have the the proper
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:05
capabilities and expertise to help me build what I want to be?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:09
Uh uh what about the product engineering?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:12
Uh will the uh sea level uh exact uh uh
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:15
support what I I want to achieve here?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:17
Uh obviously it will cost more in the beginning,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:21
until we reach uh economic benefit.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:23
Uh does the the organization have the tolerance to do it?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:27
Um second thing I would say, um access to risk talent.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:30
Um even though AI allows us to do uh more with less, we would still
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:35
need these core members to to actually uh make it happen.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:39
So, it's not only and assuming we're looking for years
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:42
ahead, it's not only who I have right now, but can
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:45
I get uh uh the right people continuously?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:48
Uh can I fill up the ranks?
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:50
Um it depends on location, on the size of the company,
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:54
on the industry that you're operating in.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:56
Um um usually for for companies that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
49:59
don't work remote, location is key.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
50:01
Uh that's something that I hear from from many
Chen Zamir
Chen Zamir
50:44
Um super interesting.
Chen Zamir
Chen Zamir
50:45
Yeah, I you said some uh some you got some very good points there.
Chen Zamir
Chen Zamir
50:49
Uh but since we are uh rounding up, and coming to the takeaways uh
Chen Zamir
Chen Zamir
50:54
section, I I I I I want to kind of like go go at it uh in order.
Chen Zamir
Chen Zamir
50:58
Uh I'll say first Netanel, super interesting conversation.
Chen Zamir
Chen Zamir
51:02
I think that you know a lot of it is very uh how
Chen Zamir
Chen Zamir
51:05
should I say, uh very exploratory in nature.
Chen Zamir
Chen Zamir
51:08
And you know I think that two years ago when AI started
Chen Zamir
Chen Zamir
51:12
to come out, I was much more or I'd say differently.
Chen Zamir
Chen Zamir
51:15
I was much less uh um imaginative uh in how I saw the future.
Chen Zamir
Chen Zamir
51:19
And I try today not to fall into the same into the same trap.
Chen Zamir
Chen Zamir
51:23
Um because things are changing, and it's very very hard to keep
Chen Zamir
Chen Zamir
51:27
up with the with the changes, and it's very very hard to you
Chen Zamir
Chen Zamir
51:31
know like think about how the future might be in two years uh
Chen Zamir
Chen Zamir
51:36
when we don't know how it would look like in two months.
Chen Zamir
Chen Zamir
51:39
Um but I think there are a few things that uh I really liked um in in
Chen Zamir
Chen Zamir
51:44
what you said that already kind of like present a couple of trends here.
Chen Zamir
Chen Zamir
51:49
Um, I'll start by saying that, you know, we've opened
Chen Zamir
Chen Zamir
51:52
the conversation with a question of, you know,
Chen Zamir
Chen Zamir
51:55
whether we will need vendors in the future.
Chen Zamir
Chen Zamir
51:58
And if you expected a yes or no answer, then, you know,
Chen Zamir
Chen Zamir
52:02
you probably didn't uh listen to my previous podcast
Chen Zamir
Chen Zamir
52:05
or ever uh read what I what I'm usually writing.
Chen Zamir
Chen Zamir
52:08
I think reality is a bit more complex than that.
Chen Zamir
Chen Zamir
52:12
Um, and so of course the the answer is no, not a yes or a no.
Chen Zamir
Chen Zamir
52:16
It's more about how the relationship between
Chen Zamir
Chen Zamir
52:19
uh vendors and businesses uh would change.
Chen Zamir
Chen Zamir
52:21
Uh I think it also you know going to one of your last points right
Chen Zamir
Chen Zamir
52:26
now is, um that the change and the relationship would probably also
Chen Zamir
Chen Zamir
52:30
look different, uh dependent on um in which industry you work.
Chen Zamir
Chen Zamir
52:34
But I would say I think one of the first points that you touched upon
Chen Zamir
Chen Zamir
52:39
is the shift between who's considered to be the data provider,
Chen Zamir
Chen Zamir
52:43
and who's considered to be the uh the main platform builder.
Chen Zamir
Chen Zamir
52:47
And while I am not sure that in two years time we
Chen Zamir
Chen Zamir
52:50
would not see any vendor platforms and we would see
Chen Zamir
Chen Zamir
52:54
only kind of like in-house built platforms.
Chen Zamir
Chen Zamir
52:57
I actually don't think that's the case.
Chen Zamir
Chen Zamir
52:59
I think that it would be a bit more balanced.
Chen Zamir
Chen Zamir
53:02
Uh but I think uh a lot of the uh kind of like core engineering
Chen Zamir
Chen Zamir
53:07
uh components that need like a lot of reliability, uh especially
Chen Zamir
Chen Zamir
53:11
the real-time ones, I think you know they the mostly
Chen Zamir
Chen Zamir
53:14
vendors would still end up servicing these.
Chen Zamir
Chen Zamir
53:17
But still the the uh the balance between who's a data provider
Chen Zamir
Chen Zamir
53:21
and who actually builds part of the platform it would change.
Chen Zamir
Chen Zamir
53:25
And it would be uh I I I think vendors would be seen more and more as
Chen Zamir
Chen Zamir
53:30
data providers, uh rather than um you know like capability enablers.
Chen Zamir
Chen Zamir
53:34
Um there would be core capabilities that
Chen Zamir
Chen Zamir
53:37
they would probably forever enable.
Chen Zamir
Chen Zamir
53:39
Uh but a lot of other capabilities would be um would
Chen Zamir
Chen Zamir
53:43
be relegated to to the to the to the business side.
Chen Zamir
Chen Zamir
53:46
Um also I I I think another point that you touched on,
Chen Zamir
Chen Zamir
53:50
and I think this will also be very interesting to see
Chen Zamir
Chen Zamir
53:53
how it would how it would uh develop, is you know when
Chen Zamir
Chen Zamir
53:57
we say data provider what data do we mean by that?
Chen Zamir
Chen Zamir
54:00
And I think there are two dimensions to that that that you mentioned.
Chen Zamir
Chen Zamir
54:05
One is you know the features themselves, and is it ever
Chen Zamir
Chen Zamir
54:09
going to be replaced by for example foundation models. Yes or no.
Chen Zamir
Chen Zamir
54:13
On one hand it's very cheap on the other hand how should
Chen Zamir
Chen Zamir
54:17
you use uh like foundation model uh features in rules.
Chen Zamir
Chen Zamir
54:20
Uh like a human cannot understand them.
Chen Zamir
Chen Zamir
54:23
Or how would you use these features in in manual review?
Chen Zamir
Chen Zamir
54:27
Um even an agent cannot really make sense of them, like an AI agent.
Chen Zamir
Chen Zamir
54:31
Um so I think this would be like this would be very
Chen Zamir
Chen Zamir
54:35
interesting to see how that would develop.
Chen Zamir
Chen Zamir
54:37
Uh on the flip side there's also the the breadth
Chen Zamir
Chen Zamir
54:41
of the data, uh like the consortium data.
Chen Zamir
Chen Zamir
54:43
Um But uh like h the all the patterns and all the intelligence that
Chen Zamir
Chen Zamir
54:48
you get from um outside of your own little box as a as a business.
Chen Zamir
Chen Zamir
54:52
And I think this is something that unless you're like incredibly
Chen Zamir
Chen Zamir
54:56
big, unless you're like top 10 in your in your uh vertical,
Chen Zamir
Chen Zamir
55:00
um you're likely to see much less data and patterns
Chen Zamir
Chen Zamir
55:04
um than you know most vendors out there.
Chen Zamir
Chen Zamir
55:06
Um, so I I really like how yeah, how how you how you
Chen Zamir
Chen Zamir
55:10
kind of like phrase the the data part there.
Chen Zamir
Chen Zamir
55:13
When it comes to what to build, Um, so yeah, of course, I mean,
Chen Zamir
Chen Zamir
55:17
if it's a core function, you want to invest more in that.
Chen Zamir
Chen Zamir
55:21
So you it's not only about, you know, building it.
Chen Zamir
Chen Zamir
55:24
It's also because building is like the easy part. It's the maintenance.
Chen Zamir
Chen Zamir
55:29
It's, you know, it's like servicing uh all of that.
Chen Zamir
Chen Zamir
55:32
Um, and that's, you know, something that hopefully,
Chen Zamir
Chen Zamir
55:36
uh, isn't something that your risk team is doing.
Chen Zamir
Chen Zamir
55:39
Um, but I think what I like the point that you made that I really
Chen Zamir
Chen Zamir
55:43
liked, because I absolutely agree with, is that where you want to
Chen Zamir
Chen Zamir
55:48
focus on in terms of building is not necessarily the your biggest
Chen Zamir
Chen Zamir
55:52
pain point or your biggest kind of like uh loss driver.
Chen Zamir
Chen Zamir
55:56
But it is really to look at your population as segments and understands
Chen Zamir
Chen Zamir
56:00
understand which ones are underserved by your vendor.
Chen Zamir
Chen Zamir
56:04
And in many cases, especially if you're a bigger business,
Chen Zamir
Chen Zamir
56:08
you probably have a very particular segment that is very
Chen Zamir
Chen Zamir
56:12
unique to you and how you run your specific business.
Chen Zamir
Chen Zamir
56:15
And that means that it will have a very particular behavior.
Chen Zamir
Chen Zamir
56:19
Uh both for the good users and the bad users.
Chen Zamir
Chen Zamir
56:22
And also a very particular data signature.
Chen Zamir
Chen Zamir
56:25
Uh and this is something that vendors will, you know
Chen Zamir
Chen Zamir
56:28
these are all like vendors would never solve.
Chen Zamir
Chen Zamir
56:31
And able to solve it on your own now is maybe a more important
Chen Zamir
Chen Zamir
56:35
thing than servicing you know like a more standard use case,
Chen Zamir
Chen Zamir
56:39
like stolen credit cards, that you know there are enough
Chen Zamir
Chen Zamir
56:43
vendors that know how to solve stolen credit cards.
Chen Zamir
Chen Zamir
56:47
Is that how you want to spend your time and effort? Right.
Chen Zamir
Chen Zamir
56:50
Um also by the way let's let's take it like one one step
Chen Zamir
Chen Zamir
56:54
below that in terms of you know rudimentary use cases.
Chen Zamir
Chen Zamir
56:58
Sanction screening, or KYC, is that something
Chen Zamir
Chen Zamir
57:01
that you want now to build?
Chen Zamir
Chen Zamir
57:03
Uh because that's that is very generic and kind of like
Chen Zamir
Chen Zamir
57:06
very table stakes and and and and you you can like easily
Chen Zamir
Chen Zamir
57:10
buy it right and you can also easily build it.
Chen Zamir
Chen Zamir
57:13
And that's that is I think the trap that
Chen Zamir
Chen Zamir
57:16
you want to kind of like avoid.
Chen Zamir
Chen Zamir
57:18
And that's why I really like the fact that you kind of
Chen Zamir
Chen Zamir
57:21
like you talked about how to recognize or identify these
Chen Zamir
Chen Zamir
57:25
niches that are unique to you and are understood.
Chen Zamir
Chen Zamir
57:28
And lastly, I think we also talked a bit about prerequisites and how
Chen Zamir
Chen Zamir
57:33
how do you know whether you're ready to to to build uh these
Chen Zamir
Chen Zamir
57:37
components, especially these more advanced complex whatever you want to call them.
Chen Zamir
Chen Zamir
57:42
Um and from a team perspective, and I think the
Chen Zamir
Chen Zamir
57:46
team is the only thing that matters here.
Chen Zamir
Chen Zamir
57:48
Uh it really it really is the only thing that matters.
Chen Zamir
Chen Zamir
57:52
I to, like I I really liked what you said and that the fact that most
Chen Zamir
Chen Zamir
57:57
leaders and I think I think it's not only about risk leaders.
Chen Zamir
Chen Zamir
58:01
Most leaders are very confident maybe even overconfident about
Chen Zamir
Chen Zamir
58:05
their own uh capabilities and their own team's capabilities.
Chen Zamir
Chen Zamir
58:09
Uh and what they fail to see in such uh like efforts
Chen Zamir
Chen Zamir
58:12
in like when you when you take such a big commitment
Chen Zamir
Chen Zamir
58:16
is that it is never about your team only.
Chen Zamir
Chen Zamir
58:18
Um and especially when it comes down to delivering code, in the end.
Chen Zamir
Chen Zamir
58:23
Uh and and that's something that you know came up in my conversation
Chen Zamir
Chen Zamir
58:27
with Lalitha in my conversation with uh Brian Davis.
Chen Zamir
Chen Zamir
58:31
Um I I think the biggest the biggest uh pitfall that I see teams uh you
Chen Zamir
Chen Zamir
58:36
know fall into, is they start building hectically without a road map.
Chen Zamir
Chen Zamir
58:40
But even worse, without any alignment
Chen Zamir
Chen Zamir
58:43
with the organization around them.
Chen Zamir
Chen Zamir
58:45
How they support their uh how they support them building it, but even
Chen Zamir
Chen Zamir
58:50
more importantly, how they support them in maintaining it afterwards.
Chen Zamir
Chen Zamir
58:54
Uh if your analyst is now building something that's
Chen Zamir
Chen Zamir
58:58
great and it will take them you know a week.
Chen Zamir
Chen Zamir
59:01
Um but if they constantly maintain it and fix bugs, and you know like
Chen Zamir
Chen Zamir
59:05
also like take in new uh requirements and they are now the product
Chen Zamir
Chen Zamir
59:10
manager and the devops engineer and the backend engineer and the
Chen Zamir
Chen Zamir
59:14
UX engineer, all in one they are not an analyst anymore.
Chen Zamir
Chen Zamir
59:18
And this is this is a trap that I see a lot of teams uh fall into.
Chen Zamir
Chen Zamir
59:22
And what happens by the way is not really that the analysts would stop
Chen Zamir
Chen Zamir
59:27
doing their job, it's that they would stop maintaining their project.
Chen Zamir
Chen Zamir
59:31
Um, and and I think this is kind of like your
Chen Zamir
Chen Zamir
59:34
point is like super super important here.
Chen Zamir
Chen Zamir
59:37
And on top of that, of course, if you need all of this support,
Chen Zamir
Chen Zamir
59:41
the first thing that you need is actually leadership
Chen Zamir
Chen Zamir
59:45
buy in and to make sure that you actually get it.
Chen Zamir
Chen Zamir
59:48
And I think that getting this buy in is much more difficult and
Chen Zamir
Chen Zamir
59:52
requires a different skill set than erecting a POC in two days.
Chen Zamir
Chen Zamir
59:56
And the the build trap that we see uh in
Chen Zamir
Chen Zamir
59:59
AI doesn't uh doesn't skip risk teams.
Chen Zamir
Chen Zamir
60:01
And I think that a lot of the teams that I see try
Chen Zamir
Chen Zamir
60:05
to avoid uh the hard part, which is the the sell.
Chen Zamir
Chen Zamir
60:08
The internal sell, uh the convincing, uh the alignment, uh
Chen Zamir
Chen Zamir
60:12
both upwards and and horizontally, believing that if they
Chen Zamir
Chen Zamir
60:16
just build it um it will it will just happen and they would
Chen Zamir
Chen Zamir
60:20
show the the proof and and they will get what they.
Chen Zamir
Chen Zamir
60:23
And usually that's not the case. Uh many times.
Chen Zamir
Chen Zamir
60:26
Yeah, Netanel, Uh super interesting.
Chen Zamir
Chen Zamir
60:29
I uh I would watch this episode two years from now,
Chen Zamir
Chen Zamir
60:32
uh just for the you know laughs and giggles.
Chen Zamir
Chen Zamir
60:35
Uh I'm sure I'm sure I would uh find myself very naive uh in its time.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:22
I would just look at how much gray hair I had back then and now.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:26
I just want to say that you know to to summarize uh
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:29
fraud prevention has been an area where machine learning
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:33
was, you know we were a very early adopters.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:36
Because so complex so much data, these are the same qualities that
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:40
uh cause it to be one of the uh laggards in AI adoption.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:44
Um but I do think that we will catch up with uh the
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:47
other use cases that are already adopting AI more.
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:51
Um and we need to see how we do it uh in
A man with salt-and-pepper hair and a beard, wearing a dark sweater, smiles gently against a white background.
Netanel Kabala
62:53
a in an efficient and smart manner.
Chen Zamir
Chen Zamir
63:08
Yeah.
Chen Zamir
Chen Zamir
63:09
And I must say I mean I already said a year ago nobody
Chen Zamir
Chen Zamir
63:12
talked about building like AI powered capabilities.
Chen Zamir
Chen Zamir
63:16
People were talking about deep fakes and uh whatever, right?
Chen Zamir
Chen Zamir
63:20
Um and now I see the conversation shifting and I
Chen Zamir
Chen Zamir
63:23
see teams are in a very different mind space.
Chen Zamir
Chen Zamir
63:26
So I I already see it uh and I'm glad for it.
Chen Zamir
Chen Zamir
63:29
Uh because that was what was interesting for me all all along. Yeah.
Chen Zamir
Chen Zamir
63:33
Anyway, Netanel, uh really appreciate you joining the pod today.
Chen Zamir
Chen Zamir
63:38
Really appreciating having this conversation.
Chen Zamir
Chen Zamir
63:41
Uh um no, we we finished with a lot more question marks than ex
Chen Zamir
Chen Zamir
63:45
exclamation marks, but I think that's uh you know, the topic warrants it.
Chen Zamir
Chen Zamir
63:50
So, I'm I'm I'm not uh I'm not upset about it.
Chen Zamir
Chen Zamir
63:53
Uh yeah, I'll I'll say thank you one more time,
Chen Zamir
Chen Zamir
63:56
and I'll see you all next Saturday.