---
title: From Snapshot to Journey: Holistic Fraud Detection in the Age of AI, with Tal Yeshanov
source_page: https://www.sardine.ai/media/fraudology/episodes/holistic-fraud-detection
canonical: https://www.sardine.ai/media/fraudology/episodes/holistic-fraud-detection
format: text/markdown
date: 2026-07-30T09:14:00.000Z
description: Holistic fraud detection means looking past the transaction. Orchestration platforms, account takeover signals, and why domain expertise still beats AI alone…
---

**Quick links:** [Human page](https://www.sardine.ai/media/fraudology/episodes/holistic-fraud-detection) · [Home](https://www.sardine.ai) · [Customers](https://www.sardine.ai/customers) · [Blog](https://www.sardine.ai/blog) · [Demo](https://www.sardine.ai/demo)

---

# From Snapshot to Journey: Holistic Fraud Detection in the Age of AI, with Tal Yeshanov

**Published:** 2026-07-30T09:14:00.000Z

Holistic fraud detection means looking past the transaction. Orchestration platforms, account takeover signals, and why domain expertise still beats AI alone…

Welcome back to Fraudology.
In this episode, I'm sitting down with Tal Yeshanov. Someone I've known for a very long time in this industry, and one of the sharpest risk leaders I know. Tal's path into fraud started almost by accident at Google and YouTube. Then took her through building fraud programs at Eventbrite, before its IPO, and Uber during its earliest hockey-stick growth years. Tal has spent her career building holistic fraud detection systems from scratch, in industries where there was no playbook to follow.
For years, fraud teams operated off a snapshot. Device at checkout. IP at checkout. Did the payment information match? Tal walks through why that single-moment view is no longer enough. And why the shift toward an orchestration platform, one that pulls in customer journey risk signals from the moment a user lands on your site rather than just the moment they transact, is where modern fraud programs are actually headed.
The deeper theme of this episode is what happens when you stop treating fraud detection as a scoring exercise, and start treating it as a full picture. Tal shares a personal story about a rule she built early in her career that was, on paper, flawless. It caught the exact triangulation fraud pattern it was designed for. It also caught a company executive, because his girlfriend used his credit card in a different city. That's false positive reduction in fraud detection in its most human form, and it's a direct argument for upstream fraud prevention data collection: pulling in more signals earlier in the journey instead of adding more rules at the transaction point.
What you'll hear in this episode:
How Tal moved from Google and YouTube into building fraud programs at Eventbrite and Uber with no existing playbook to follow.
Why holistic fraud detection means tracking customer journey risk signals from first visit to transaction, not just a snapshot at checkout.
How an orchestration platform unifies device, IP, email, and behavioral data that used to live in separate point solutions.
A real story about false positive reduction in fraud detection, including a rule that was technically perfect and still failed a legitimate customer.
How the same holistic approach extends to account takeover detection, including typing cadence, autofill behavior, and device history.
Why domain expertise vs AI in fraud isn't a competition, and how agentic AI is being used right now to query databases, support customer service teams, and triage escalations.
A candid conversation about AI replacing fraud jobs, including a real example of a company that prematurely laid off its fraud leadership.
Why fraud team tribal knowledge doesn't transfer to an AI model, and what companies risk losing when it walks out the door.
Tal's fraud leadership philosophy, built on transparency and empathy, and why it creates teams that stay in touch for years.
The ongoing industry shift from risk team vs fraud team naming, and why more companies are choosing the broader term.
You should listen to this episode if you:
Work in fraud detection, risk operations, account security, or trust and safety.
Are evaluating an orchestration platform or trying to move your fraud program beyond point-in-time scoring.
Want a practical, non-hypothetical look at where agentic AI actually fits in fraud operations today.
Are a fraud leader worried about AI replacing fraud jobs on your team, or trying to make the case for why domain expertise still matters.
Care about fraud leadership philosophy and want to build a team that stays loyal and stays sharp.
### Episode notes
Holistic fraud detection starts with the customer journey, not the transaction
For most of fraud's history, detection meant a snapshot. Device at checkout. IP at checkout. Did the AVS and CVV match. Tal explains why that model is running out of runway, and why the future belongs to an orchestration platform that tracks customer journey risk signals from the moment someone arrives on a website or app, not just the moment they buy.
A transaction-only snapshot misses everything that happens before the purchase.
Customer journey risk signals include how someone entered the site, how they interact with it, and even how they hold their device.
An orchestration platform unifies data that used to live in separate device, IP, email, and phone vendor tools.
This shift allows workflows to run across the entire customer lifecycle instead of a single rule firing at one moment.
False positive reduction in fraud detection is just as important as catching fraud
Tal tells a story about a rule she built early in her career that caught triangulation fraud with near-perfect precision, and also caught a company executive whose girlfriend had used his card in another city. It's a sharp, human example of why false positive reduction in fraud detection matters as much as catching bad actors, and why upstream fraud prevention data collection, gathering more signals earlier, produces better decisions than adding more rules at the point of transaction.
A rule can be technically effective and still generate costly false positives.
Upstream fraud prevention data collection means capturing behavior long before checkout, not just at it.
Understanding a user's full history and device relationships prevents good customers from being treated like fraudsters.
This same logic directly extends to account takeover detection, including login timing, autofill versus manual entry, and device consistency.
Domain expertise vs AI in fraud isn't a competition, it's a dependency
Tal shares specific, practical ways she's used agentic AI, including querying a company database through Claude instead of writing SQL, using AI to answer routine customer support questions from existing documentation, and using AI to triage competing escalations. But she's clear that domain expertise vs AI in fraud is the wrong framing. Agentic AI can execute a workflow. It cannot design one.
Agentic AI is most effective on repetitive, well-documented tasks like escalation triage or first-line customer questions.
The domain expertise fraud prevention requires is what determines whether the AI is even asking the right question.
Customization matters more than the tool itself. A one-size-fits-all AI approach does not work across fraud programs.
The people who bring creativity and institutional knowledge to design these systems are the ones who remain essential.
AI replacing fraud jobs is already happening, and fraud team tribal knowledge is what's at risk
This is the part of the conversation that got personal for both of us. AI replacing fraud jobs isn't hypothetical. Tal and I have both seen companies prematurely lay off experienced fraud leaders on the assumption that an AI tool could simply take over, only to watch account takeovers and chargeback losses climb within months. What gets lost in those moments is fraud team tribal knowledge, the kind of institutional understanding that lives in a person's head after years in the industry and cannot be transferred to a model.
Premature AI-driven layoffs have led to measurable increases in account takeover and chargeback losses.
Fraud team tribal knowledge cannot be replicated by AI because it was never documented in the first place.
The right question isn't which jobs AI can replace, but which repetitive tasks should be automated so people can focus on higher-value work.
The professionals who will remain essential are the ones who bring the domain expertise fraud prevention actually requires to designing systems, not just running them.
Fraud leadership philosophy and the risk team vs fraud team conversation
We close with Tal's fraud leadership philosophy, built on transparency and empathy, and why the people who've reported to her over the years still reach out for advice long after they've moved on. We also touch on the broader risk team vs fraud team naming shift happening across the industry, as more companies fold fraud, abuse, first-party fraud, and account security under one umbrella.
Transparency and empathy build the kind of trust that keeps former team members engaged for years.
Being direct, done respectfully, is a form of kindness in fraud leadership, not a lack of it.
The risk team vs fraud team shift reflects how many different types of loss now fall under one umbrella.
Strong fraud leadership philosophy matters more, not less, as AI changes what day-to-day fraud work looks like.
### Key takeaways
A transaction-only snapshot misses everything that happens before the purchase; holistic fraud detection requires tracking customer journey risk signals from first visit onward.
An orchestration platform unifies data that used to live in separate device, IP, email, and phone vendor tools, allowing workflows to run across the entire customer lifecycle.
A rule can be technically effective and still generate costly false positives; false positive reduction in fraud detection matters as much as catching bad actors.
Upstream fraud prevention data collection, gathering signals earlier in the journey, produces better decisions than adding more rules at the point of transaction.
The same holistic logic extends directly to account takeover detection, including login timing, autofill versus manual entry, and device consistency.
Domain expertise vs AI in fraud is a dependency, not a competition. Agentic AI can execute a workflow, but the domain expertise fraud prevention actually requires is what determines whether the AI is asking the right question in the first place.
AI replacing fraud jobs is already happening, and companies that prematurely cut experienced fraud leadership have seen account takeover and chargeback losses climb within months.
Fraud team tribal knowledge cannot be replicated by AI because it was never documented. It leaves with the person who holds it.
Fraud leadership philosophy built on transparency and empathy is what keeps former team members engaged and reaching out for advice years later.
The risk team vs fraud team naming shift reflects how many different types of loss, from account security to first-party fraud, now fall under one umbrella.
### Final takeaway
Holistic fraud detection is not just a technical upgrade. It's a mindset shift, from scoring a moment to understanding a journey, from static rules to orchestration platforms, and from fearing AI to knowing exactly where domain expertise still has to lead. The fraud programs that will hold up best are the ones built by people who understand both the technology and the tribal knowledge no model can replace.
Episode resources & links:
Connect with Tal Yeshanov | LinkedIn
Connect with Karisse Hendrick | LinkedIn
Host of the Fraudology Podcast
Award-Winning Cyberfraud Expert
Ecommerce Fraud Prevention Consultant
Startup Advisor, Keynote Speaker, and
Consultant to Fortune 500 merchants