Run a customer risk review

Every customer in FlowSentry has a risk rating that you can take apart factor by factor. This tutorial follows a typical review: you spot a customer whose rating looks too low, understand why, test whether the model itself should change, propose an override for a second person to approve, and then work the enhanced due diligence (EDD) case that a High rating opens.

Before you start

Step	Who can do it
Read ratings and the risk register	Anyone who can read customers (Analyst, Senior analyst, MLRO, Deputy MLRO, Risk manager, KYC reviewer, Relationship manager, Senior approver, Auditor)
Use the simulator and draft model versions	Risk manager
Propose an override	Risk manager or MLRO
Approve an override or activate a model version	MLRO or Deputy MLRO — not the proposer — with step-up
Work an EDD checklist	KYC reviewer, Risk manager, MLRO or Deputy MLRO
Approve an EDD case	Senior approver, MLRO or Deputy MLRO — not the submitter

How ratings are produced

You never score customers by hand. FlowSentry rates customers when they are loaded or their profile changes, and a half-hourly backfill rates anyone missed. Each rating combines five factor groups — customer, geography, product, channel and behaviour — into a 0–100 score that maps to a band: Low, Medium, High or Very high. Floor rules can lift the band regardless of score; for example, a domestic PEP is never rated below Medium.

Until your institution activates its own model, ratings use the platform reference model, labelled "Platform reference · uncalibrated".

Step 1: Find customers to review

1. Open Customers › Customer risk (/risk).
2. Use the figures at the top — rated customers, High band, reviews due, overrides and open EDD cases — as filters.
3. Filter the register by effective band, model, reviews due within 30 days, PEPs only or overridden only.
4. Check Band mix for the spread of bands and the movements over the last 90 days.
5. Open a customer to go to their risk tab.

Step 2: Read the explainable score

On the customer's Risk tab (/customers/{id}?tab=risk):

• Current standing shows the effective band, the score, the model, the next review date and any EDD case.
• How the score is made up is a waterfall with one bar per factor. Each bar is worth group weight × factor weight × the factor's 0–100 score, with a plain-English reason. The bars add up to the overall score. Any floor rules that raised the band are listed.
• Score history shows every score, what triggered it and who asked for it.

Select Explain this rating for an AI reading of what drives the band and what would change it. It is advisory and sends nothing identifying to the model.

If the customer's details have just changed, Rescore now (risk model managers and approvers) recalculates the rating and records your name in the history.

Step 3: Test a model change with the simulator

If the issue is the model rather than one customer — for example, the geography weight is too light for customers with high-risk jurisdictions — test it first.

1. Open Customers › Risk models (/risk/models) and go to the What-if simulator.
2. Choose a Starting model: one of your versions or a platform reference model.
3. Edit the Factor group weights (they must add up to 1.00) and the Band cut-offs.
4. Select Simulate. FlowSentry re-scores the stored customer snapshots and shows the band-shift matrix — how many customers would move from each band to each other band — and the customers who would move. Nothing is saved.
5. If the result is right, select Save as draft version and explain Why these changes. A different person (MLRO or Deputy MLRO) reviews it and selects Activate with step-up. Doc pack downloads the version's governance documentation.

Step 4: Propose an override

When the model is right in general but misses something about one customer, propose an override.

1. On the customer's risk tab, select Propose override.
2. Choose the Override band.
3. Write a Reason for the override of at least 20 characters — what you know that the model does not.
4. Set Expires on, at most one year away.
5. Select Submit for approval.

The override waits for a checker. The proposer cannot approve it.

Step 5: The checker decides

1. The MLRO or Deputy MLRO sees the override under Overrides on /risk, in My approvals and in their notifications.
2. They review the reason and the breakdown and select Approve override, completing step-up, or Reject with a reason.

Once approved, the customer shows Overridden with the new effective band in the risk register and the customer register. The calculated score is still recorded beside it. The override ends automatically on its expiry date.

Step 6: Work the EDD case

A customer whose effective band becomes High gets an EDD case automatically.

1. Open Customers › EDD workbench (/risk/edd). Cases are listed by status: Open, Awaiting approval, Approved, Rejected.
2. Open the case and work through the Checklist — for example source of funds, source of wealth, purpose of the relationship and enhanced monitoring. For each item, record the evidence you saw and select Mark complete.
3. When the checklist is complete, select Submit for approval.
4. A senior approver — not you — selects Approve relationship or Reject with a reason.

Tips

• Overrides are for exceptions. If you find yourself overriding many customers the same way, change the model instead.
• Use the Customer risk dashboard (/dashboards/customers) to watch bands, PEPs, overdue reviews and expiring IDs across the book.
• The EWRA page (/risk/ewra) holds the institution-level risk assessment that your customer risk methodology should reflect.

Related

• Risk manager
• Screening and risk FAQ
• Smarter screening and explainable risk