Demo on synthetic data (Kaggle datasets named on each report). No real client data.

WoWyrdConfidential

WoWyrd · Rule review

Rule review: extended management report

Prepared for Demo client (ps, synthetic)

Client: Demo client (ps, synthetic). Data: 2017-01-01 to 2017-01-31 (daily periods). Cut-off: 2017-01-31 (set for this run). Generated: 2026-11-20 10:02 UTC.

Internal report. Do not share it outside your firm.

Internal report. Do not share it outside your firm.

This is the extended management report. The short management report is management.html; the technical report is technical.html.

Headline numbers

54,510
Transactions reviewed
Transactions file, 2017-01-01 to 2017-01-31
54,425
Alerts
Rule-hit log, 2017-01-01 to 2017-01-31
6,153
Confirmed fraud
Labels file, known by 2017-01-31
45,625
False alarms
Rule-hit log and labels, mature hits, 2017-01-01 to 2017-01-31
219,000.00 USD
False-alarm cost (estimate)
Your unit costs, 2017-01-01 to 2017-01-31
99.6%
Share of fraud with a rule hit
Labels and rule-hit log, 2017-01-01 to 2017-01-31

Value in your own costs

In 2017-01-01 to 2017-01-31, false alarms took 4,562.5 analyst hours and cost 219,000.00 USD. Source: your unit costs and the rule-hit log.

3 of the top actions touch different rules. Together they remove 1,854 alerts and save 6,748.80 USD over the same period. They lose 304 caught fraud cases. This is a backtest on past data, not a forecast.

Charts

Alerts and confirmed fraud per period31 periods from 2017-01-01 to 2017-01-31. Alerts total 54,425; confirmed fraud total 6,153.Alerts6,36202017-01-01: 6,362 alerts2017-01-02: 4,215 alerts2017-01-03: 335 alerts2017-01-04: 419 alerts2017-01-05: 317 alerts2017-01-06: 4,435 alerts2017-01-07: 3,811 alerts2017-01-08: 3,907 alerts2017-01-09: 3,233 alerts2017-01-10: 2,959 alerts2017-01-11: 2,932 alerts2017-01-12: 2,561 alerts2017-01-13: 2,932 alerts2017-01-14: 2,594 alerts2017-01-15: 2,668 alerts2017-01-16: 2,364 alerts2017-01-17: 2,729 alerts2017-01-18: 417 alerts2017-01-19: 288 alerts2017-01-20: 358 alerts2017-01-21: 400 alerts2017-01-22: 546 alerts2017-01-23: 514 alerts2017-01-24: 455 alerts2017-01-25: 546 alerts2017-01-26: 349 alerts2017-01-27: 315 alerts2017-01-28: 310 alerts2017-01-29: 569 alerts2017-01-30: 318 alerts2017-01-31: 267 alerts6,362Confirmed fraud30602017-01-01: 256 confirmed fraud2017-01-02: 292 confirmed fraud2017-01-03: 297 confirmed fraud2017-01-04: 262 confirmed fraud2017-01-05: 236 confirmed fraud2017-01-06: 228 confirmed fraud2017-01-07: 263 confirmed fraud2017-01-08: 260 confirmed fraud2017-01-09: 250 confirmed fraud2017-01-10: 262 confirmed fraud2017-01-11: 253 confirmed fraud2017-01-12: 288 confirmed fraud2017-01-13: 231 confirmed fraud2017-01-14: 232 confirmed fraud2017-01-15: 257 confirmed fraud2017-01-16: 247 confirmed fraud2017-01-17: 306 confirmed fraud2017-01-18: 271 confirmed fraud2017-01-19: 237 confirmed fraud2017-01-20: 243 confirmed fraud2017-01-21: 258 confirmed fraud2017-01-22: 238 confirmed fraud2017-01-23: 215 confirmed fraud2017-01-24: 271 confirmed fraud2017-01-25: 0 confirmed fraud2017-01-26: 0 confirmed fraud2017-01-27: 0 confirmed fraud2017-01-28: 0 confirmed fraud2017-01-29: 0 confirmed fraud2017-01-30: 0 confirmed fraud2017-01-31: 0 confirmed fraud3062017-01-012017-01-042017-01-072017-01-102017-01-132017-01-162017-01-192017-01-222017-01-252017-01-282017-01-31
Alerts (bars) and confirmed fraud (dashed line) per period. Two scales: fraud is far smaller than alerts. Source: rule-hit log and labels, 2017-01-01 to 2017-01-31.
False-alarm cost by rule, top 8Transfer above 500 USD (Rule ID: tp_u21_transfer_over_500): 89,664.00 USD; Single payment above 9,000 USD (Rule ID: tp_u21_large_txn_9k): 84,583.20 USD; Money from 5+ different senders into one account within 5 days (fan-in) (Rule ID: tp_ffiec_fan_in_5_senders_5d): 44,752.80 USDTransfer above 500 USD (Rule ID: tp_u21_transfer_over_500)Transfer above 500 USDTransfer above 500 USD (Rule ID: tp_u21_transfer_over_500): 89,664.00 USD89,664.00 USDSingle payment above 9,000 USD (Rule ID: tp_u21_large_txn_9k)Single payment above 9,000 USDSingle payment above 9,000 USD (Rule ID: tp_u21_large_txn_9k): 84,583.20 USD84,583.20 USDMoney from 5+ different senders into one account within 5 days (fan-in) (Rule ID: tp_ffiec_fan_in_5_senders_5d)Money from 5+ different senders into one account wi…Money from 5+ different senders into one account within 5 days (fan-in) (Rule ID: tp_ffiec_fan_in_5_senders_5d): 44,752.80 USD44,752.80 USD
False-alarm cost per rule in USD. An alert hit by several rules shares its cost equally. Source: your unit costs, 2017-01-01 to 2017-01-31.

Top 5 actions

ActionTypeExpected effect (backtest)Confidence
REC-f1c16d71 Retire rule Money from 5+ different senders into one account within 5 days (fan-in): 2.4% of its 28,140 alerts with a known outcome were fraud
tp_ffiec_fan_in_5_senders_5d: Finds one account that collects money from many people, as a collection or mule account does.
Retire the rule58 fewer alerts, 7 caught fraud cases lost, 244.80 USD saved (estimate)medium
REC-69738ebd Retire rule Transfer above 500 USD: 11.8% of its 51,693 alerts with a known outcome were fraud
tp_u21_transfer_over_500: Flags every transfer over a low fixed amount, so analysts see most mid-size transfers.
Retire the rule869 fewer alerts, 145 caught fraud cases lost, 3,129.60 USD saved (estimate)medium
REC-21bf9124 Raise the review score from 1 to 2: 927 fewer alerts in the backtestRetune the threshold927 fewer alerts, 152 caught fraud cases lost, 3,374.40 USD saved (estimate)medium
REC-e3f79706 Add a rule for: Gambling winnings arrive with no deposits, then card transfers followAdd a missing fraud typeNo effect estimate.medium
REC-06d81254 Add a rule for: Large loans are taken and repaid again and againAdd a missing fraud typeNo effect estimate.medium

Effects come from a backtest on 2017-01-01 to 2017-01-31. Selective labels: a declined payment has no outcome, so fraud a rule stopped by declining is not in "caught fraud cases lost". The technical report gives the evidence and caveats for each action. A rule name links to its entry in the rule glossary. Each action has a reference (REC-…). Quote this reference in your change record.

Your tools: what each one adds

Demo: the tool names and the tool map are made up. The rules are the demo's own rules.

Each rule belongs to one tool, from your tools.csv. Rules that are not in the map are grouped as "Not assigned". An alert's cost is shared equally among the tools that fired on it, so the tool costs add up to the total false-alarm cost. Catch rate = caught / all mature fraud with a known outcome (6,153 cases).

ToolRulesAlertsCaughtCatch rateFalse alarmsPrecisionFalse-alarm cost (estimate)Only this tool caught
In-house rules (example)254,3676,11999.4%45,57411.8%152,968.80 USD5,448
AML monitoring (example)128,14067811.0%27,4622.4%66,031.20 USD7

Overlap between tools

Caught by ↓, also caught by →In-house rules (example)AML monitoring (example)
In-house rules (example)—11.0% (671)
AML monitoring (example)99.0% (671)—

Each cell: the share of the row tool's caught fraud that the column tool also caught, with the count. A tool with fewer than 10 caught cases shows "too few to compare".

Fraud types per tool

Fraud typeIn-house rules (example)AML monitoring (example)
no type given6,119 (5,448 only)678 (7 only)

Caught fraud cases per fraud type in your labels; in brackets, the cases only this tool caught.

If one tool were off (backtest on past data, not a forecast)

ToolAlerts removedCaught fraud cases lostCost saved (estimate)
In-house rules (example)26,2855,44887,182.40 USD
AML monitoring (example)587244.80 USD

Only the payments where this tool alone fired change. Cost uses the same either/or rule as the rest of the report: analyst time when not declined, friction when declined.

Your tools over time

The technical report compares each tool's precision and catch rate between 2017-01-19 to 2017-01-21 and 2017-01-22 to 2017-01-24, with the counts and the changes larger than our limit. This page shows none of them, so that no small window count can be worked out from this page.

What this data cannot show

  1. A tool whose score or decision is not in your export cannot be measured here. Access to a tool is not use of it: this report measures only what your hit log recorded.
  2. A tool that acts before the hit log is written (for example a decline at authorization) is not visible here.
  3. A declined payment has no outcome, so a tool that declines payments gets less credit than it may deserve.
  4. "Tool off" assumes your other tools would have acted the same. They may not: one tool's decision can change what another tool sees.
  5. A score band is one cut-off. The same tool at another cut-off can catch more or less.
  6. Two windows of three periods show a change, not its cause. A change in the tool's set-up, in your traffic or in your labels can each move the figures. A window counts periods that hold data, not calendar periods: a gap in your export makes a window span more time, and the table per period shows it.

For internal use. Check your vendor contracts before you share tool-level numbers outside your firm.

What changed since the previous report

No previous report was given, so this report is treated as the first one. There is nothing to compare with yet. The next run can compare with this one.

Coverage by fraud type

Fraud types present in your labels

Fraud types present in your labelsType not given: 6,126 caught, 27 missedType not givenType not givenType not given: 6,126 caughtType not given: 27 missed6,126 caught, 27 missed
Labelled fraud cases per type. Blue: a rule fired on the payment. Orange: no rule fired. Source: labels file and rule-hit log, labels known by the cut-off.
Fraud typeLabelled casesCaughtMissedWoWyrd
Type not given6,1536,12627WoWyrd rule: planned

Not in this dataset, so it cannot be measured here

Your labels hold no case of these types, so this data cannot show whether your rules catch them. This is not a gap in your rules.

New in our reference set since your last report

Our reference set: version v0.7.0, published 2026-11-18 (built into this collector).

This is the first report with a record of our reference set. The next report lists what is new since this one.

Incoming payments and money-mule signs

Incoming payments not measured: your transactions file has no direction column (for example direction, with values such as in and out).

Requirements changed since the last review

Your rules are mapped to current requirements. We flag when a new requirement affects them. We follow official publications of regulators in Ukraine, the EU, the UK and the US.

Not legal advice. This report maps your rules to requirements, in our own short wording. It is not legal advice. You decide how you comply.

Requirements library read on 2026-11-11. No change file was given for this run, so no changed requirement is listed here.

Requirements still open since the previous report

No previous report was given, so this report is treated as the first one. There is nothing to compare with yet. The next run can compare with this one.

Requirements that showed a gap are followed up only when the two reports compare.

Not legal advice. Your compliance team decides how a requirement applies to you. Public cases show how regulators have treated the same kind of gap; they do not predict an outcome for you.

Regulatory checks by regulator

Not legal advice. This report maps your rules to requirements, in our own short wording. It is not legal advice. You decide how you comply.

Regulatory statuses as of 2026-11-11 (library date). Check them again before you rely on this report.

Grouped by regulator. Our own short wording, not the regulator's text. The technical report gives each finding and its evidence.

Checks per NBU requirements

Rows citing the NBU draft of 17.07.2026 — draft, not in force: its text can change before adoption. Resolutions 51, 54 and 116 are in force. Bill 16013 is a Ukrainian bill, shown here with its status.

Requirement (our wording)ResultSource statusAs ofOfficial source
The acquirer passes payee (merchant) name and category code to the issuer. Check that these fields are filled in your data. Most time-relevant item: the deadline is about 20.11.2026.Not measurable from these filesin force since 20.05.2026; 6 months to comply (about 20.11.2026)2026-11-11bank.gov.ua
Review every rule at least monthly by back-testing it.Gapdraft (NBU draft resolution on control of payment operations, published 2026-07-17; not law)2026-11-11bank.gov.ua
Metric 1: compare how often each rule fired with how many payments were actually stopped.Gapdraft (NBU draft resolution on control of payment operations, published 2026-07-17; not law)2026-11-11bank.gov.ua
Metric 2: share of fraud value that was prevented, out of all fraud value.Gapdraft (NBU draft resolution on control of payment operations, published 2026-07-17; not law)2026-11-11bank.gov.ua
Set rule-effectiveness thresholds in writing. A rule under its threshold gets an immediate review.Gapdraft (NBU draft resolution on control of payment operations, published 2026-07-17; not law)2026-11-11bank.gov.ua
Record a review decision for every rule: keep, adjust or cancel.Partly metdraft (NBU draft resolution on control of payment operations, published 2026-07-17; not law)2026-11-11bank.gov.ua
Have evidence ready for a supervisor: documents, rules, risk markers.Partly metdraft (NBU draft resolution on control of payment operations, published 2026-07-17; not law)2026-11-11bank.gov.ua
Show the acquirer's full name before an online payment; receipts show payee, commission, purpose and category code.Not measurable from these filesin force since 20.05.2026; 6 months to comply (about 20.11.2026)2026-11-11bank.gov.ua
Reconcile payment operations every day.Partly metdraft (NBU draft resolution on control of payment operations, published 2026-07-17; not law)2026-11-11bank.gov.ua
Acquiring for natural persons only for allowed categories, with the required data passed to the issuer.Not measurable from these filesin force since 20.05.2026; 6 months to comply (about 20.11.2026)2026-11-11bank.gov.ua
Non-bank providers that acquire payments need an authorisation; the sanctions pause has ended.Not measurable from these filesin force; grace period ended 02.09.20262026-11-11bank.gov.ua
Remote verification of natural persons uses the updated procedure with credit-bureau data.Not measurable from these filesin force since 10.10.2026; 3 months to align internal procedures (about 10.01.2027)2026-11-11bank.gov.ua
Reports to the State Financial Monitoring Service go only through the e-cabinet, with numbered registers.Not measurable from these filesin force since 10.10.2026; 3 months to align internal procedures (about 10.01.2027)2026-11-11bank.gov.ua
Passing cards, account access or payment data to fraudsters becomes a crime. Detect and restrict accounts used as drops.Not measurable from these filesadopted by the Rada 28.10.2026; presidential signature not confirmed2026-11-11itd.rada.gov.ua

Checks per EU requirements

Requirement (our wording)ResultSource statusAs ofOfficial source
When a payee name check fails because of the payee's provider, that provider compensates the payer's provider. Keep evidence of the name check on received transfers.Not measurable from these filesagreed, not adopted (EU Payment Services Regulation; adoption expected by 12.2026)2026-11-11oeil.europarl.europa.eu

Checks per UK requirements

Requirement (our wording)ResultSource statusAs ofOfficial source
A claim may carry an excess of up to 100 GBP, never for vulnerable customers.Not measurable from these filesin force2026-11-11psr.org.uk
Test and review rules; a poorly calibrated rule set the firm cannot explain is poor practice.Partly metguidance in force2026-11-11fca.org.uk

Checks per US requirements (FinCEN, NYDFS, FFIEC, Nacha, Federal Reserve)

FinCEN and Nacha items are not checked from your files in this run; the rows below are NYDFS and FFIEC items.

Requirement (our wording)ResultSource statusAs ofOfficial source
Analyse scenarios, rules and thresholds on an ongoing basis.Partly metin force2026-11-11dfs.ny.gov
Test rules before and after they go live.Not measurable from these filesin force2026-11-11dfs.ny.gov
Tune and validate monitoring; justify each threshold.Partly metguidance2026-11-11bsaaml.ffiec.gov

Fraud types in plain words

Each fraud type named in your labels, in one sentence. The word in small grey text is the label as your file gives it.

Type not given no type
Fraud labels in your file that do not say which kind of fraud it was.