Dear engineer,
Somebody just walked over with a request that sounds simple: “Analysts need more context when an alert opens. Can you add it to the screen?”
I’m writing because I’ve seen how this goes. It starts with good intentions and ends with a screen that scrolls for a page and a half. So here are a few things I wish someone told every team at the start.
First, what decision does this screen support?
An AML analyst opening an alert is trying to decide one thing: is this activity explained, or does it need to go further? Every item on the screen should help with that decision. If a field wouldn’t change what the analyst does, it’s decoration.
Ask three analysts to cross out the fields they never use. You’ll be surprised how little survives.
Second, “nothing found” has two meanings
This is the one that bites developers hardest. Imagine the screen shows “Previous cases: 0”.
Does that mean there are none? Or that the search couldn’t find them, because the earlier case was filed under a trading name and your query used the legal name?
A human reading “0” will assume the first. So the screen quietly teaches analysts that no history exists, when it only means none was found.
My suggestion: give every section three honest states: found, none found and could not check. Make “could not check” visible and boring, a small grey label. It takes ten minutes to build and prevents a lot of false comfort.
Third, show how sure you are
Entity matching is messy. Two records might be the same customer, or they might not. Don’t merge them silently, and don’t hide them either. Show “possible match”, show the evidence behind it (same address, same phone number), and let the analyst confirm or reject.
Their answer is data too. Store it, so the next analyst doesn’t redo the same check.
Fourth, order by consequence, not by database
Put what changes decisions at the top: how this activity compares with the customer’s normal pattern, the outcome of the last case, any open questions. Keep raw records one click away.
Think in three layers: the verdict frame, the supporting evidence, the drill-down. If a field lives in the top layer, it should earn its place.
Why I care about this
When context is hard to see, analysts do what anyone would. They open other tabs, search again, copy identifiers and rebuild a picture the institution already holds. Better detection doesn’t fix that. A better first screen does.
I wrote about the compliance side of the problem here: why alerts arrive without context.
Over to you
I’d really like to hear how you’d approach this, especially if you’ve built something similar.
Have you built an internal tool where “no data” was ambiguous? How did you handle it?
If you could show only five things at the top of an investigation screen, what would they be?
Hot take: a screen that shows less but is trusted beats a screen that shows everything. Agree, or fight me?
Warm regards,
Sonal
Top comments (0)