BlogTechnical
Agents of Risk
Every investigation should make the next one smarter. False positives, fragmented knowledge, and the risk team that remembers.

One hundred risk analysts walk into a conference room.
They work at the same bank, but they sit in different offices, investigate different customers, and handle different parts of the business. Each has spent the past week making decisions with the information available to them.
Imagine that every Monday, they gather around a giant whiteboard. Everyone brings an observation.
An address that keeps appearing. A business whose documents looked convincing until someone checked the dates. A screening alert that took hours to clear.
One analyst writes down a finding. Someone across the room recognizes it. A third pulls up a case from last month.
Now they have something worth investigating.
The meeting is valuable because each person brings context the others were missing. Connections become visible when they put their observations together.
Collective context
What if that whiteboard were always on?
What if every case added what was learned as it moved through the queue? An investigator opening a new case could draw on a colleague’s work from yesterday, another office, or three years ago. A connection could become visible while there was still time to act.
Consider a wire payment that triggers a screening alert because the recipient’s name resembles a name on a sanctions list. The analyst compares registration numbers, addresses, and ownership records and establishes that these are two different companies. That particular match is a false positive.
A false positive is only obviously false after someone has done the work.
One payment can produce dozens of possible matches. Each needs an explanation, and clearing a name match does not settle every question about the payment. In the example below, most matches can be dismissed, but the payment’s purpose still needs investigation.
Acme Bank · Wire review
The next analyst may face the same names and have to repeat those checks. An always-on whiteboard could bring forward the earlier explanation and its supporting evidence. They would still need to check that it applies and that nothing material has changed, but they would have a place to start.
That is time they could spend on the unresolved question. And whatever they learn about it would become available to the next investigation.
But the whiteboard has a limit: it only contains what analysts notice and decide to write down.
What about the details nobody realizes matter in isolation?
A director’s name in an incorporation document. A subsidiary mentioned on an invoice. A news article about two executives who used to work together. Each may look unremarkable within a single case. Together, they may reveal a relationship worth examining.
Agents could inspect each case, contribute observations with their sources, and pursue connections that no individual analyst knew to look for. One could examine ownership records while another follows payment history. Their findings could give a third a reason to revisit an earlier case.
The relationship between Acme and Globex illustrates the possibility. A wire, company filings, and a news article each contribute part of the picture. The work is in connecting them, testing the explanation, and identifying what still needs to be established.
Cross-case investigation
The system would need to keep evidence separate from the conclusions drawn from it. A filing can establish ownership; an article may only suggest a relationship. An agent should be able to follow either lead without treating both as equally certain.
The same attention should turn inward, toward the work of investigation itself.
An agent might repeatedly miss ownership evidence buried in an appendix. Similar cases might receive different decisions because different teams ask different questions. A useful check might appear in one investigator’s notes but never make it into anyone else’s process.
Those are things a risk team should be able to learn from. Reviewing how investigations unfold could reveal where a process needs another check, where instructions are ambiguous, or where an agent’s work needs closer scrutiny.
And some of the value would not involve finding a pattern at all.
One relevant prior investigation could answer the question holding up today’s case. A previous explanation could save someone from pursuing a dead end. New evidence could overturn an old belief and change how the team approaches a customer.
An experienced analyst already does some of this. A detail reminds them of a previous investigation; they remember which explanation held up and where to look next. When they leave, their case files remain, but the associations that made those records useful can leave with them.
A company can retain its records while losing some of its ability to use them.
Shared memory should preserve those connections and make them available to whoever needs them next. It should also preserve uncertainty. A shared address can have an ordinary explanation. Several agents repeating a suspicion does not establish that it is true.
When a finding is challenged, corrected, or retired, that change needs to reach the investigations that relied on it. Otherwise, yesterday’s weak suspicion becomes tomorrow’s recurring alert.
The process itself needs the same discipline. A proposed improvement should be tested against cases where the evidence and outcomes can be reviewed. Faster work is useful, but a missed relationship or an unsupported conclusion cannot count as progress.
That is the ambition behind what we’re building at Footprint: a risk team whose investigations improve both its understanding of the world and the way it does its work.
Every case should leave the team better equipped for the next one.
In the next post
- How do you coordinate thousands of risk agents across millions of investigations so that each can build on what the others discover?
- How do you turn patterns across investigations into critical findings, with enough evidence to act on them?
- How do you continuously improve, refine, and update the investigation process—and verify that each change makes its conclusions more accurate?



