Context & Why Capturing Contextual Memory Is Crucial in the AI Era

I stumbled across something unexpected while researching the app logic behind an account takeover spike for a consulting client: a nearly complete product plan written by the fraud team almost two years earlier. It described much of what we were trying to figure out and the surrounding context, including how the team had thought about the problem and what they had planned to build. Nobody I spoke to remembered it existed. Finding it saved us a lot of time, but it also exposed a bigger problem. The organization had already done much of the thinking. It simply had no way of knowing that knowledge was there.

Ask a fraud manager what happens when an experienced investigator leaves the company, and they probably won’t say, “We lose all their knowledge.” There will be a handover. There will be documentation, links to Confluence pages, investigation reports, dashboards, and repositories. The rules and models are still there. On paper, the organization has everything it needs to continue. What is often missing is the context. What problem was the team trying to solve? What did they observe afterward, and what eventually led them to keep, modify, or reverse the decision? Without that context, the artifacts tell you what changed, but not what the team learned along the way.

As Jared Gruenberg, founder of Stingray Fraud Intelligence, a case management platform built specifically for fraud teams, put it: “The artifacts remain; the story connecting them often doesn’t.”

Fraud Has a Memory Problem

Fraud prevention is rarely a sequence of independent decisions. A new attack leads to an investigation. The investigation leads to a rule, model, or process change. That change produces new evidence, which leads to another decision. Over time, fraud teams build up a body of knowledge about what they have seen, what they tried, what worked, and what failed.

The problem is that this knowledge rarely exists in one place. The production code may be there. A Confluence page may describe the rule or model. An investigation report may explain what happened. But the analytical trail is often missing: the queries used to build the original dataset, the hypotheses the team considered, the models they rejected, and the reasoning behind the final decision.

This is more than a documentation problem. It is a knowledge problem. Research on fraud investigations makes a similar distinction. Petter Gottschalk of BI Norwegian Business School describes knowledge as facts combined with interpretation, context, and reflection. In fraud prevention, knowing what happened is not the same as understanding what the organization learned from it.

The Context + Intuition Problem

Fraud investigators don’t make every decision by following a documented procedure. With experience, they develop a mental library of patterns, exceptions, and warning signs. Some of that knowledge is difficult to articulate because it comes from seeing hundreds or thousands of cases over time.

I saw this firsthand in an investigation involving a fraud ring that was opening fake accounts in one European country and transferring money from another. The ring was never formally documented, but the researcher who had investigated it remembered the pattern. When similar activity appeared later, that memory allowed the team to narrow the investigation quickly.

This kind of intuition is valuable, but it creates another layer of the knowledge problem. The organization may have the transaction data and the investigation tools, but the reasoning that connects one case to another can exist only in someone’s head. When that person is no longer involved, the organization may have to rediscover what they already knew.

So how do you capture knowledge that investigators may not even think to document? As Jared puts it: “The honest barrier is that nobody has time to document. In my experience, if capturing the reasoning isn’t part of the workflow at the moment of the decision, it just doesn’t happen.”

Then AI Arrived

Fraud teams have always had a knowledge problem. AI did not create it. It has simply made the problem harder to ignore.

AI is changing fraud prevention from a collection of tools that support human decisions into systems that can increasingly perform parts of the investigative and decision-making process themselves. That makes organizational knowledge a different kind of asset.

AI doesn’t necessarily need more data. Fraud organizations already have plenty of it. What AI needs is context: an understanding of what happened, what the organization decided, why it decided it, and what happened afterward.

Without that context, AI can find patterns across thousands of cases without understanding their significance. It may recommend an action without knowing that the same approach failed before, or mistake an old exception for a new fraud pattern. As AI takes on more investigative and decision-making work, the cost of missing context becomes greater.

Capturing Context Becomes a Competitive Advantage

The answer is not to ask fraud teams to document more. McKinsey estimates that knowledge workers already spend about 20% of their time searching for and gathering information. The goal should be to make knowledge capture part of the work itself, supported by workflows and tools that capture and connect the reasoning as decisions are made, so that finding what the organization already knows does not depend on knowing where to look.

When a rule, model, threshold, or process changes, the reasoning behind the decision should be captured at that moment. Investigation findings, including failed hypotheses and rejected approaches, should become part of the record without requiring a separate documentation exercise.

This turns individual experience into institutional memory. It gives the next investigator, analyst, or AI system access not only to what the organization did, but to what it learned.

In an AI-driven fraud environment, the advantage will not come from having more data or another model. It will come from preserving what the organization has learned and making that knowledge available when the next decision is made. The organizations that can preserve their memory will be able to build on every investigation instead of starting from zero.