Use caseBSA/AML compliance

High-risk customer AML review

Each quarter, 286 high-risk files come due. Every one starts with a hunt for what expired or changed, and the hunt eats half the review.

Every file is pre-checked before a reviewer opens itHealth data sent by mistake leaves the recordTwo people sign every high-risk review

The assistantPre-check against your review rules

Good standing certificate and signature cards are valid. Three items are outdated, starting with the manager's passport, expired on 10/14/2026.

source · 9 documents on the customer file
Actions on this review
The pre-check on the file

Items outdated or activity to explain · the AI lists each one

01

The problem

The short version

On January 4, 286 review files open, and 41 of them are already overdue. Most are hard for the same reason: the file itself, built over years, with an ID that expired last month, a declaration from two springs ago, and an activity nobody has set beside the profile since the last review. The reviewer spends the morning discovering that, while the overdue count grows. Half of each review is an inventory done by hand, before any judgment starts.

Now the inventory runs ahead of the reviewer. The AI pre-checks every file in the campaign against your review rules, lists what is outdated with its document, and sets the activity beside the declared profile. Luis starts from a list of facts and spends his time on the judgment, which stays with two people.

Review REV-27-0193 · ActivityViewer
What a year changed at Gulfport. The activity statement on the left, the pre-check count on the right, before Luis has read a page.
Review campaign Q1 2027Case management
The quarter on one line. All 286 reviews are pre-checked the morning they open. 118 will need something from the customer.

A BSA team has seen every one of these screens before. What it has not seen is this order: every file in the campaign is pre-checked before assignment, the customer is asked once through the advisor who knows them, and the rating is confirmed by a reviewer and signed by a second. The nine steps below follow one of the 286, Gulfport Import Export, from the campaign run on January 4 to the second signature on January 19.

02

The campaign

The quarter's reviews open at once, and each file is checked before anyone reads it.

Step 1Workflow

The quarter's review cases open in one run

On January 4, your scheduler calls the repository API: one review case opens for every customer rated High whose review date falls in the quarter. Priya Natarajan, BSA officer, finds 286 of them, with the 41 overdue from last quarter at the top of each queue.

The workflow assigns them to her five reviewers by branch and by workload.

Review campaign Q1 2027Case management
A home page report, built by Priya, counts the cases by status.
The campaign starts the same morningNobody builds the list in a spreadsheet. The review cases exist, assigned, on the day the quarter opens.
Step 2AI check

Each file is pre-checked before assignment

With 286 files, the AI check runs asynchronously. It reads each customer file against the review rules your team wrote in plain language.

Are all IDs valid on the review date?Is the ownership certification under 12 months?Are the latest financial statements on file?

Each result lands on the case as a status and its reasoning. Luis Ortega's queue can be sorted by it, so he sees which files need the customer and which he can review today.

My review queueCase management
A file marked Complete still needs a reviewer. The pre-check tells him where to start.
The inventory is done before the reviewThe reviewer no longer opens a file to find out what it lacks. That list is on the case when he arrives.
Step 3Viewer

Open the file and read the pre-check

Luis opens Gulfport Import Export, a food wholesaler trading with North Africa. The pre-check lists three items: the manager's passport expired in October, the ownership certification is from March 2025, and the 2025 financial statements are missing. Each line opens its document in the viewer.

Passport_Haddad.pdfViewer
Each outdated item points to the document and the date that triggered it.
The reviewer confirms in minutesThree items, three documents, three dates. Luis checks each one at its source before anything goes to the customer.
03

The file is brought up to date

Luis needs new documents and two explanations, so he goes through the advisor who knows the customer.

Step 4Assistant

Set the activity beside the declared profile

The yearly activity statement is filed on the case. Luis asks the assistant to compare it with the profile recorded at the last review. The answer names two differences, each with both sources.

Review REV-27-0193Case management
The assistant reports differences. Whether they are suspicious is Luis's call.
Time it on your next file: how long does it take to set two documents side by side, before the reviewer can even form a view?
The questions are factualLuis puts two figures to the customer, each with its source. Nothing in the file says suspicion before a person has looked.
Step 5Assistant

One request, sent by the advisor

Luis asks for a draft listing the three documents and the two questions. Megan Fowler, the relationship manager, softens the opening and sends it under her name, since Sam Haddad knows her. The case moves to Waiting for customer, with an SLA timer on the step.

Request to customerCase management
The draft comes from the pre-check and the questions. The advisor's edits are kept.
The customer hears it from someone they knowThe request is complete and personal. The reviewer prepares it, the advisor sends it, and the customer answers once.
Step 6Viewer

The reply comes in, and health data is set aside

Sam Haddad uploads a 9-page scan: passport, certification, financial statements, invoices for a market stall paid in cash, and contracts with two olive oil suppliers. Page 6 is an employee's medical certificate, sent by mistake. The AI flags it as personal data the review does not need.

Luis splits the scan in the viewer, files the business pages in their slots and removes the certificate. The action is recorded with his name.

Scan_documents_january.pdfViewer
The flag is a suggestion. Luis decides what leaves the record.
The AML record keeps only what it needsA page that should never have been filed is caught before it is, and the removal is part of the trail.
04

Rated and signed

The file is current again. Luis confirms the rating, and Priya countersigns.

Step 7AI check

The pre-check runs again on the new documents

With the new pages in their slots, the analysis runs again. The documents pass. The activity point stays open on purpose: the explanation is for a person to accept.

AI analysis resultsAI console
The earlier warnings stay in the history, with the documents that cleared them.
The documents stop being the topicWhat is left for the reviewer is the judgment on the activity, which is the reason the review exists.
Step 8Workflow

The reviewer confirms the rating and writes his conclusion

Luis reads the invoices and the contracts. He accepts the explanation, adds Tunisia to the declared countries, keeps the rating at High because of cash volumes and cross-border flows, and writes his conclusion. He asks for a summary to hand over.

Review REV-27-0193Case management
The rating is a field the reviewer sets, with his reason next to it.
The conclusion is written onceThe reviewer's reasoning and its sources travel together. The second reviewer reads them without reopening the whole file.
Step 9History

The second reviewer signs and the next date is set

Priya opens the case from her signing queue, reads the summary and the two supporting documents, and signs. The customer file gets its next review date, twelve months out. Her campaign report moves Gulfport Import Export to Done.

Review REV-27-0193 · HistoryCase management
Both signatures, the request, the removal and the rating sit on one trail.
The examiner's question is answered from the fileWho reviewed, who signed, what was asked and what came back. The campaign count and every trail behind it are in one place.
05

The verdicts

Gulfport came back with three outdated items. Any file in the campaign comes back one of three ways, judged against your review rules, and each way starts different work.

Complete

Every document is current

IDs valid, certification recent, financials on file. The pre-check records each date it read.

Then: the reviewer looks at the activity, rates, and sends for signature.
Attention required

Items to renew or explain

An expired ID, an old declaration, a difference with the declared profile. Each item names its document.

Then: one request through the advisor; the check runs again when the reply arrives.
Not eligible

A policy rule fails, and you know which

A document your policy requires that the customer can no longer produce. The AI names the rule that blocks.

Then: the BSA officer reviews the file and decides what follows.
Where the AI stops

It rates nobody. It files no SAR. It closes no relationship. Across the nine steps on this page, the rating and the signatures belong to two people, each with their own rights and their own trace.

The AI pre-checks, compares, flags and drafts. The control is in the review rules and in the second signature, so it holds when the campaign is three hundred files long.

06

What it saves

One campaign of 286 files, with one of them followed end to end. A smaller portfolio or stricter rules would move these counts.

286 filespre-checked before a reviewer opens the first one.
1 requestto the customer for three documents and two questions, sent by the advisor.
0 pagesof health data left in the AML record after the reply came in.

Where these numbers come from. They are counted on the scenario described on this page, with example data. They are not a customer result.

Three principles behind the figures

Zero blind openingsProductivityEvery file arrives with its list of outdated items. The reviewer starts from facts and their documents.
Zero extra dataData protectionWhat the customer sends is checked for personal data the review does not need, and a person removes it.
Decision supportTwo people decideComparisons and summaries cite their sources. The rating and both signatures stay human.
07

Built for

A reviewer, a BSA officer and an architect all sign off before a campaign runs this way. Each of them is looking for something different.

Business operations

AML reviewers

Mornings vanish into finding out what a file is missing.

  • Each file arrives with its outdated items listed.
  • The activity is set beside the declared profile, with sources.
  • The customer request is drafted for the advisor to send.
  • Your conclusion travels with its documents to the second reviewer.
Risk and compliance

BSA officers and privacy officers

Examiners open with one question: how many reviews are overdue?

  • The campaign and its progress are a report you build.
  • The AI rates and signs nothing. Two people do.
  • Personal data outside the review purpose is flagged before filing.
  • The review rules are yours, written in plain language.
IT and architecture

CIOs and architects

Each campaign still means an extract, a spreadsheet and a mail merge.

  • Review cases are created through the API from your customer data.
  • The pre-check runs asynchronously when the campaign is large.
  • Permissions apply before the AI reads a file.
  • Deploy it in your own data center or in the cloud, as your AML data policy allows.

A good first step

Begin with the check your reviewers do first on every file, such as ID validity on the review date, and run it across next quarter's campaign before assignment. One quarter is enough to compare queues with and without it.

Your customer files, review cases and documents stay in the systems that hold them today.

Your move

See it on your own review files

Bring a few anonymized high-risk files and your review checklist. We will pre-check them in front of you and leave the judgment with your reviewers.

The screens on this page are drawn to show one step at a time, with example data.

Uxopian Software helps organizations put AI to work across the systems they already run, without giving up control of their data, their infrastructure or their critical processes.