Rules without code: the guided rule builder

Monitoring rules should belong to the compliance team that answers for them, not to whoever can read a configuration file. This release brings a guided rule builder and a new rule library to FlowSentry, so analysts can describe a pattern, see exactly what the rule will do, test it against history and send it for approval, all without writing code.

The rule library

Rule library (/monitoring/rules) lists every rule your institution runs: the built-in scenario pack (about 25 typology scenarios, such as structuring, mule fan-in, rapid movement, velocity, wires to high-risk jurisdictions and round amounts) together with your own rules.

Search by id or name, and filter by status, severity, typology, transaction type, channel or what the rule looks at. Tiles show what is Active, Awaiting approval, My drafts and Built-ins switched off. Each row shows the rule's alerts over the last 30 days and its last change.

Each rule page opens with In plain English, then its scope, window, thresholds, conditions, cool-off after an alert, score weight, versions with Compare, and its backtest and performance.

The create rule wizard

Select Create rule (/monitoring/rules/new). The first question is "How do you want to start?"

• Describe it in plain English. Write what you want to catch, for example several cash deposits just under the reporting threshold within a day. The AI drafts the rule and lists Notes from the AI draft. You review every part before anything is saved.
• Start from a built-in scenario and adapt it.
• Start from a blank rule.

Then work through five steps:

1. Basics: name, typology (with Suggest from typology), severity and score weight.
2. What to watch: transaction types, channels, whether to add up activity for the customer, account or counterparty, and the window to look back over.
3. Conditions: for example the number of transactions or the total amount, with filters, combined as All of or Any of.
4. Thresholds: every number becomes a named threshold with a currency. Add calibrations for other base currencies.
5. Review & test: a sentence describing the whole rule, any problems flagged with "Fix the problems above to submit.", and a backtest showing Projected alerts.

Save draft at any step; only you can see a draft. When you are ready, Submit for approval. Experienced users can open Advanced and Edit as JSON.

Every number is a named threshold

The builder never buries a number inside a condition. Each one becomes a named threshold with its own default, which your institution can tune and which appears on the rule page. When an alert fires, the analyst sees each threshold against the actual value, so the reason for the alert is never hidden.

Test before you submit

The backtest replays your stored transactions through the rule exactly as the live engine would, and shows projected alerts, distinct subjects and an estimate of false positives based on past dispositions. If the rule would have stayed silent, the page says so: "The rule would have stayed silent over this window." The Backtest page (/monitoring/backtest) lets you replay any rule version at any time.

Tuning the built-ins for your institution

Select Tune on a built-in scenario to change its thresholds for your institution. Your tuned version keeps the built-in's id and replaces it for your institution only. Retire your version and the original comes back.

If a built-in does not fit your business, Switch off removes it from your monitoring and Switch back on restores it. Both go through approval.

What stays governed

• Nothing runs until a second person approves it. A rule version moves from Draft to Awaiting approval to Active only when an MLRO, Deputy MLRO or Administrator approves it, and never the author.
• Approval needs a fresh second factor. The approver confirms with their authenticator code or passkey.
• Rejecting or retiring needs a reason. The author can withdraw a submission until it is decided.
• The AI drafts; it never activates. A plain-English draft is a starting point you review, and it goes through the same approval as any other rule.
• Every version is kept. Rule versions (/monitoring) shows every version and its maker-checker status.

How to try it

1. Open Monitoring in the rail and select Create rule.
2. Choose Describe it in plain English and describe a pattern you already investigate by hand.
3. Read the Notes from the AI draft and check each step.
4. On Review & test, read the sentence and the projected alerts.
5. Save draft, or Submit for approval if you are happy.

Note: Only roles that can author rules see Create rule. Others can read rules in the library.

Related

• Build a monitoring rule
• Switch a built-in scenario off or on
• Analyst and senior analyst
• Release notes