Synthetic. Auditable. Benchmarked.

Every configuration choice should be deliberate.Know the evidence behind it.

We test your screening system with synthetic data and show what each setting detects and what it costs in alerts. Our findings and recommendations are based on that test, so any change should be confirmed in your own production environment before you choose the balance that fits your risk appetite and business.

  • Independent validation testing

    An objective, third-party check on your screening engine's performance.

  • Threshold tuning without real customer data

    Safe calibration before go-live.

  • Health checks for audit reporting

    Ready-made evidence for your compliance file.

  • Regulatory readiness

    A methodology-versioned record to support supervisory reviews.

The measurement gap

Most firms have a screening system. Few can prove it performs.

  • 01

    No safe test data

    Real customer data cannot be used freely for testing, and few firms maintain a versioned, auditable test set.

  • 02

    Tested once, then assumed

    Many systems are tested at implementation and rarely afterwards. Few teams can evidence to a regulator how theirs performs today.

  • 03

    Nothing to compare against

    Without a common method, there's no objective way to compare performance with others.

  • 04

    Blind spots in production data

    It only covers the name variations your customers happen to have, and only shows the alerts raised, never the matches missed below the line.

What we measure

Know how well your screening really works

Every test of the same type follows the same methodology: the same variation types, in the same proportions. Each Health Check is therefore compared with other Health Checks, and each Breadth Check with other Breadth Checks, across clients, vendors and test runs.

  • Effectiveness

    Does it catch what it should?

    How reliably your system detects sanctioned names, from exact matches to deliberately altered spellings, and whether it correctly clears non-sanctioned names that should not match.

    → Risk assurance

  • Efficiency

    How much noise does it create?

    How many alerts it raises to find the real matches, and how that changes as the threshold moves.

    → Lower alert volume, and the analyst hours and cost behind it

  • Benchmark

    How does it compare?

    How your current settings and vendor perform against anonymised results from other firms tested in the same round.

    → Perspective on where you stand today

  • Diagnosis

    What should change?

    Which variation types, fuzziness levels and entity types cause your misses and alerts and, where needed, detailed customised testing on those findings (Round C). A tuning change is confirmed by re-running Round A or B.

    → Continuous improvement

“We do not simply report where your screening stands; we explain why it performs as it does and what can be improved.”

From our testing

What testing has shown clients

Configuration comparison

Same names. Same threshold. Only the configuration changed.

61 → 190

exact matches of 200

24% → 63%

sanctioned test names caught

Situation
A client was screening with an in-house configuration and had alternative settings in mind, but no evidence to choose between them.
Outcome
Exact matches caught rose from 61 to 190 of 200 (31% to 95%), and sanctioned test names caught overall from 188 to 504 of 800 (24% to 63%). Total alerts rose from 2,265 to 4,709, but fewer alerts were raised for each sanctioned name found.
Dr Haibo Zhang

Who we are

Built by Compliance Analytics, led by Dr Haibo Zhang

Designed the testing methodology behind this portal and is a pioneer of synthetic data testing for sanctions screening systems. Formerly Global Head of Analytics for Financial Crime and Threat Mitigation at HSBC, with 20+ years across global banks, consultancies and vendors including NICE Actimize and Fortent. Has served as a Skilled Person monitor for a major US brokerage firm.

About Compliance Analytics →

Source lists

Sanctions lists we work with

Standard datasets are currently built from OFAC or the EU consolidated list. Datasets from any other list below are available on request.

  • OFAC: OFAC Specially Designated Nationals (SDN) List (standard)
  • UN: UN Security Council Consolidated List (on request)
  • UKSL: UK Sanctions List (FCDO) (on request)
  • DFAT: Australia DFAT Consolidated List (on request)
  • CANADA: Consolidated Canadian Autonomous Sanctions List (SEMA / JVCFOA) (on request)
  • EU: EU Consolidated Financial Sanctions List (FSF) (standard)
  • SECO: Switzerland SECO Consolidated Sanctions List (on request)

How testing works

Two standard rounds, repeated regularly

Two standardised rounds for regular assurance, and a bespoke service for specific questions.

Round A1,000test names

Health Check

Exact matches, low to high fuzziness and non-sanctioned names, covering typographical errors, phonetic spellings and abbreviations.

Round B10,000+test names

Breadth Check

Low-fuzziness variations across around 15 categories for each entity type (individuals and companies), tested at scale to identify weaknesses category by category.

Round CBespoke

Where Rounds A and B identify scope for improvement, or a specific question needs testing

Expert Optimisation Service

A detailed, customised dataset built on your Round A and B findings or a specific question, such as how your system handles transliterated names from a particular country. It is scoped individually rather than forming part of the regular cycle.

Built from official sanctions lists

Each standard dataset is currently built from a single list, OFAC or the EU consolidated list. Datasets built from the UK Sanctions List or any other list shown above are available on request.

Ongoing assurance, not a one-off

Sanctions authorities update their lists frequently and at irregular intervals, sometimes several times within a fortnight. We recommend testing quarterly, or every six to twelve months for smaller firms. Each test confirms that your screening continues to perform after those changes and provides a dated record as evidence.

How a test round works

  1. Step 1

    We issue the dataset

    A new synthetic dataset is prepared each month and made available in your workspace, with a notification when it is ready.

  2. Step 2

    You screen the dataset

    Download the dataset, screen it through your system and upload the results.

  3. Step 3

    Report and benchmark

    We analyse your results and send you a report. Enterprise clients and Round A pilot participants also receive a 30–45 minute walkthrough of the findings. Round B adds an anonymised peer benchmark once enough firms take part.

Ready for your next test?

Data hosted in the EU (Ireland), 10-year retention by default, methodology-versioned reports and anonymised peer comparison.