What problem does it solve?
Missing a sanctioned party can bring enforcement action and large fines, so screening engines are tuned to match generously. A customer named Mohammed Ali, a ship with a common name or a company whose address contains a sanctioned city all create alerts, and most of them turn out to be false positives. Each one needs an analyst to compare dates of birth, nationalities, identifiers and context against the list entry.
On payments the pressure is time. An instant or cross border payment held for a name match breaks the settlement promise to the customer, and a backlog on a busy day means delayed payroll or trade payments. On onboarding, screening alerts slow account opening. Meanwhile list updates after a new sanctions package can raise alert volumes sharply overnight.
How does it work?
- Parse the record. Names, dates, addresses, identifiers and free text are extracted from the customer record or the payment message, including structured ISO 20022 fields.
- Resolve the match. Entity resolution compares the record with the list entry across transliterations, aliases, name order and cultural naming patterns, and weighs secondary identifiers such as date of birth, nationality and registration numbers.
- Score and explain. A model estimates whether the alert is a true match and lists the factors that support or contradict it, in words an analyst and an auditor can follow.
- Decide under policy. Alerts that meet approved criteria for a clear non match are closed with that explanation stored. Possible and likely true matches go to an analyst, ranked by risk, with the evidence side by side.
- Assure. A sample of automated closures is reviewed by a second analyst, and every list update triggers regression tests on known true and false matches.
- Audience
- Back office
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| False positive reduction | Too few to pool | 60% | 1 | 1 organization |
Value drivers: Compliance quality, Speed and cycle time, Employee productivity, Customer experience.
Indicative value
A bank screening payments and customers with 300,000 name screening alerts a year
USD 157,500 to USD 1.4 million
Screening analyst capacity released per year
How this is calculated
Formula: alerts * minutesPerAlert / 60 * autoClosed * costPerHour. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Name and payment screening alerts per year alerts, alerts per year | 300,000 | 300,000 | The reference bank. |
| Analyst minutes per alert today minutesPerAlert, minutes per alert | 3 | 8 | Editorial assumption for level one review. Replace with your own time study. |
| Share of alerts closed automatically as clear non matches autoClosed, fraction of alerts | 0.3 | 0.6 | Editorial assumption, kept at or below UOB's name screening pilot result on this page (60% fewer false positives on individual name alerts). Replace with results from your own parallel run. |
| Fully loaded analyst cost per hour costPerHour, USD per hour | 35 | 60 | Editorial assumption. Replace with your own. |
What it leaves out: Counts analyst time only. It leaves out faster payment release and onboarding, lower penalty risk, and the cost of validation, list management and the platform.
Who already uses it?
7 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
First National Bank of Omaha (FNBO)
United States · Banking · 2026
FNBO deployed Nasdaq Verafin's Agentic EDD Analyst and Agentic Sanctions Analyst, which remove manual information gathering across multiple systems for enhanced due diligence cases and sanctions alerts. The vendor reports that the bank spent 50% less time on these reviews and alerts and redirected investigator capacity to deeper analysis.
- Handling time reduction: 50%, per case, enhanced due diligence and sanctions alert reviews
"At First National Bank of Omaha, AI agents have begun taking on some of the work of human financial crime investigators, reducing the time that people spend on each case by 50%, according to bank executives."
Claimed by: independent
Standard Chartered
United Kingdom · Banking · 2018
Standard Chartered announced a partnership with Silent Eight to give its financial crime compliance teams machine learning and natural language processing for name screening. The system recommends whether a screening alert is a true or false match and explains the recommendation in a plain English narrative for the analyst. The announcement describes aims rather than results.
No outcome disclosed.
United Overseas Bank (UOB)
Singapore · Banking · 2018
In a six month pilot reported in August 2018, UOB tested Tookitaki's Anti-Money Laundering Suite, with machine learning features co created by the bank, on top of its rule based name screening (against internal and external watch lists) and transaction monitoring, to separate genuine risk from false positives with explainable outputs. UOB reported large false positive reductions on name screening alerts and said it would progressively roll the solution out to customer risk assessment and sanctions screening, which it treats as processes separate from name screening.
- False positive reduction: 60%, six month pilot, name screening alerts on individual names
"For name screening alerts, there was a 60 per cent and 50 per cent reduction in false positives"
Claimed by: organization - False positive reduction: 50%, six month pilot, name screening alerts on corporate names
"For name screening alerts, there was a 60 per cent and 50 per cent reduction in false positives"
Claimed by: organization
AJ Bell
United Kingdom · Wealth and asset management · 2025
UK investment platform AJ Bell uses ComplyAdvantage's AI powered Customer Screening and Ongoing Monitoring, whose matching combines fuzzy logic with machine learning that learns aliases and global naming conventions. AJ Bell's Head of Financial Crime and MLRO says that optimising the system's settings cut alert volume by 82 percent, which lets analysts clear alerts faster and focus on the highest risk areas.
- Alert volume reduction: 82%, customer screening alert volume
"Through optimizing the levers within the system, we’ve been able to reduce our alert volume by 82 percent."
Claimed by: organization
Ratepay
Germany · Payments and cards · 2025
Ratepay, a German provider of white label buy now pay later solutions and part of the Nexi Group, replaced its previous solution with Hawk's Payment Screening, which screens transactions in real time against global sanctions lists, and Hawk's AML Transaction Monitoring, with centralised case management for investigators and auditors. The vendor's story says Ratepay is now planning to add Hawk's AI technology for anomaly detection and false positive reduction, so the AI adjudication step this record is filed under is announced rather than live.
No outcome disclosed.
HSBC
United Kingdom · Banking · 2024
Silent Eight has supplied HSBC with automation for name screening and adverse media alerts, and in February 2024 the two expanded the partnership to automated alert closure for transaction screening, which investigates and resolves payment screening alerts in real time. No outcome figures are disclosed.
No outcome disclosed.
Mashreq
United Arab Emirates · Banking · 2024
Mashreq selected Silent Eight in May 2024 to automate the adjudication of name screening and adverse media alerts related to sanctions and anti money laundering requirements. Under the plan, false positives are to be investigated and closed quickly and potential true positives escalated to Mashreq analysts. The announcement is a multi year partnership; no results are disclosed.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Historical screening alerts with final decisions and reasons
- Clean customer reference data with secondary identifiers (dates of birth, nationalities, registration numbers)
- Current sanctions and watchlists with version history
- Payment messages with structured party fields where available
Systems to integrate
- Screening engine for customers and payments
- Payment hub and message queues for held payments
- Onboarding and customer data systems
- Case management for escalated alerts
- List management and watchlist data providers
Complexity: High
The matching problem is well understood, but a missed sanctions hit can lead to regulatory fines, as the Starling Bank case on this page shows. Validation, list management, change control and explainability carry more of the effort than the model.
- 1
Tune the engine first
Before adding AI, fix data quality and fuzzy matching thresholds in the screening engine, and measure alert volume per list and per source. Some false positives are cheaper to prevent than to adjudicate.
- 2
Build a labelled test set
Assemble historical alerts with final decisions plus known true matches and synthetic hard cases (aliases, transliterations, partial names) that every model version must pass.
- 3
Run as a recommender
Show the model's recommendation and explanation to analysts for a full quarter and measure agreement, especially on the alerts analysts escalated.
- 4
Automate only clear non matches
Approve auto closure for the band where secondary identifiers clearly contradict the list entry, and never for alerts where the model is uncertain.
- 5
Govern list and model changes together
Re run the regression test set on every list update, model change and threshold change, and keep the results as evidence for auditors.
Guardrails
- The model may close clear non matches but never a possible or confirmed true match
- Every closure stores the explanation, the list version and the model version
- Regression tests on known true matches run before every list, model or threshold change
- Second analyst sampling of automated closures with a hard stop on errors
- Screening coverage monitored so that no customer or payment skips screening when the AI service is down
KPIs to instrument
- Alert volume and share closed automatically, per list and source
- Error rate in second analyst sampling of automated closures
- Time payments are held for screening
- Agreement between model recommendation and analyst decision
- Regression test pass rate on known true matches
Human in the loop
Analysts decide every possible or likely true match and every payment rejection or asset freeze. A second line samples automated closures, and the sanctions compliance officer approves auto closure criteria and every change to them.
Common failure modes
- A true hit closed automatically
- The one failure that matters. Prevent it with conservative auto closure bands, regression tests on known matches and second analyst sampling.
- List update floods
- A new sanctions package can raise alert volumes sharply overnight, and the model has not seen the new entries. Keep capacity plans and re test on every list update.
- Unexplainable decisions
- A score without reasons cannot be defended to an examiner. Require the model to state the identifiers that support or contradict the match.
What are the risks and rules?
EU AI Act
Minimal risk
Sanctions screening by banks and payment firms is not listed in Annex III: point 5 covers credit scoring and life and health insurance pricing, and point 6 covers AI used by or on behalf of law enforcement authorities. It is not a prohibited practice under Article 5, and as an internal tool it carries no Article 50 transparency duty. It still processes personal data at scale, so GDPR applies, and decisions that block a payment or freeze assets remain human decisions.
Rules that apply
Guidance
- A Framework for OFAC Compliance Commitments (US Office of Foreign Assets Control, North America). Lists sanctions screening software or filter faults among the root causes of apparent violations, which is why screening changes need testing and oversight.
- Principles for Using Artificial Intelligence and Machine Learning in Financial Crime Compliance (Wolfsberg Group, Global). The 2022 principles of a group of global banks for using AI and machine learning in financial crime compliance: legitimate purpose, proportionate use, design and technical expertise, accountability and oversight, and openness and transparency.
Controls to put in place
- Written auto closure criteria approved by the sanctions compliance officer
- Stored explanation, list version and model version for every decision
- Regression test set of known true matches, run on every change
- Second analyst sampling with a hard stop on errors
- Model inventory entry with validation and change control
When it went wrong elsewhere
- FCA fines Starling Bank for failings in its financial crime systems and controls. The FCA fined Starling Bank about GBP 29 million in 2024. From 2017 its automated screening system had screened customers against only a fraction of the full sanctions list, which shows why screening coverage and list completeness must be tested independently of any model.
Frequently asked questions
- Can AI clear sanctions alerts automatically?
- Banks let AI close clear non matches under written criteria, with sampling and regression tests. It should never close a possible or likely true match, because a missed sanctioned party can bring enforcement action and large fines, as the FCA's GBP 29 million fine on Starling Bank for sanctions screening failings shows.
- What makes sanctions screening alerts so noisy?
- Engines match generously to avoid misses, names are common and transliterated in many ways, and customer records often lack the secondary identifiers that would rule a match out. Better reference data reduces noise before any AI is added.
- Which banks use AI for screening adjudication?
- Standard Chartered, HSBC and Mashreq have announced screening automation with Silent Eight, and UOB reported a 60 per cent reduction in false positives on individual name screening alerts in a six month pilot with Tookitaki. AJ Bell says it cut customer screening alert volume by 82 per cent with ComplyAdvantage. Most announcements disclose no production results, so insist on your own parallel run.
How to cite this page
Blits.ai AI Use Case Library, "AI for sanctions screening alert adjudication", last verified 26 September 2026, https://www.blits.ai/ai-use-cases/sanctions-screening-adjudication. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published