Hande
False positives in know-your-business (KYB) screening are an unavoidable property of fuzzy-matched sanctions, PEP, and adverse media checks. The real lever is reducing how much analyst time each one costs, not the alert count itself.
Key takeaways
False positive rates in KYB screening are 90–95% by design. McKinsey, PwC, and ACAMS all converge on this range. Better matching helps at the margin; analyst time per alert is the real lever.
Four operational moves carry most of the impact: bulk dismissal from the list view, side-by-side comparison of the screened entity and matched profile, automatic ignoring of recurring hits on profiles a human has already cleared (the Wolfsberg Group's "Good Guys list" or whitelisting approach), and policy-level knockouts on unambiguous sanctions hits.
Match assessment is where the volume sits, not impact assessment. Mechanical comparison of attributes between the screened entity and the matched profile can be automated. The institution's policy decision should run on a deterministic policy engine.
Defensibility is the new regulatory bar. Under the European Anti-Money Laundering Authority (AMLA) framework, operational since 2025, every dismissed alert needs a recorded rationale. Firms that store decision logic in the system, not in analyst memory, are at a structural advantage.
Duna AI virtual screening assistants reduce false positives by approximately 70% across the queue (source: Duna AI memo).
What is a false positive in KYB?
A false positive in KYB is a screening alert that fires against the wrong entity. A KYB analyst onboarding a new business runs the company and its ultimate beneficial owners (UBOs) against sanctions, PEP, and adverse media databases. The vendor returns a hit. The hit belongs to a different person with a similar name, a different company in a different jurisdiction, or an article unrelated to financial crime. The analyst dismisses it and moves on.
A false positive wastes analyst time. A false negative — a real match the system missed — is a regulatory breach. The two failure modes pull in opposite directions, which is why screening engines err toward over-alerting.
What is the average false positive rate in KYB screening?
False positive rates in KYB and AML screening run between 90% and 95% of all alerts. McKinsey reports that more than 90% of transaction-monitoring alerts at most banks turn out to be false positives, with only one or two of every hundred alerts acted upon. PwC research puts the range at 90–95%. ACAMS estimates up to 90% for sanctions screening specifically, with compliance teams spending the bulk of their investigation time on dismissals. Duna's own observation across customer screening queues is closer to the upper bound — 95–99% — depending on the vendor and the configuration.
Why are KYB false positive rates so high?
Three structural causes compound:
Input data is incomplete. Most KYB screening uses a name and a country, sometimes a date of birth or a national ID. Without strong identifiers, the engine cannot disambiguate "John Smith, Netherlands" from every other John Smith in the world. Better input data starts at onboarding — see first-time-right data collection.
List data is messy. Sanctions and adverse media sources contain incomplete profiles, transliteration variants, and entries that are not actually about financial crime. A five-year-old regulatory fine can score the same severity as a sanctions designation. The Wolfsberg Group notes that "weak aliases" and low-quality identifiers in source lists are a primary driver of avoidable hits.
Fuzzy matching is the default for good reason. Vendors tune matching logic to be permissive — to catch transliterations, spelling variations, and partial data. Exact matching would miss real risks. The system therefore over-alerts by design. Academic research into natural-language-processing methods for sanctions screening confirms that purely string-based matching cannot resolve this without sacrificing recall.
Can you eliminate false positives in KYB?
No. Any tuning aggressive enough to remove false positives would also remove true positives, which is a regulatory breach. The achievable target is reducing the time analysts spend on each false positive, not the alert count itself. Done well, the same compliance team reviews more entities, with shorter cycle times, and produces a more defensible audit trail.
What is the difference between match assessment and impact assessment?
A KYB analyst reviewing a screening hit does two sequential checks. Most tools collapse them into one, which is why review takes so long.
Match assessment — is this the same entity? The analyst compares attributes between the screened business or person and the matched profile: name, date of birth, address, nationality, identifiers. If key attributes clearly conflict, the alert is a false positive. If they align, it is likely a true match.
Impact assessment — does this matter? If the entities match, the analyst then asks whether the finding is credible, whether it indicates financial crime risk, and whether it crosses the institution's anti-money-laundering (AML) policy thresholds. A confirmed match to a five-year-old corporate fine is a different outcome than a confirmed match to an active sanctions designation.
Most of the volume sits in match assessment. Automation has the highest leverage there because the comparison is mechanical and the input data is structured. Duna runs impact assessment on its policy engine — translating compliance rules into code so the impact decision is deterministic and auditable.
How do you reduce false positives in KYB? Four operational moves that work
1. Bulk dismissal from the list view
Forcing analysts into a detail view for every alert turns a five-second decision into a one-minute decision. Letting analysts dismiss confirmed false positives from a list — with one decision applied across many alerts — is the single largest time saving in most review queues. See how Duna implements this in Decide.
2. Side-by-side comparison of screened entity and matched profile
Many false positives are obvious at a glance: different country, different date of birth, different industry. Showing the screened entity and the matched profile next to each other, in the list view rather than only in the detail page, removes the click-through tax on the obvious dismissals.
3. Automatic ignoring of recurring hits on cleared profiles
A UBO whose name happens to match a sanctioned individual will trigger the same alert on every periodic monitoring cycle. Storing the analyst's prior decision and automatically ignoring future hits for that same profile — while preserving the original reasoning in the audit record — removes recurring noise from the queue without removing it from the audit trail. The Wolfsberg Group's guidance on sanctions screening recognises this practice under the labels "Good Guys lists" and "whitelisting," and considers it acceptable provided the suppression is documented and reviewable. The benefit compounds over the lifetime of the customer relationship, because every future hit for that profile is suppressed.
4. Policy-level knockouts on unambiguous sanctions hits
True hits against active sanctions designations are a stop, not a workflow question. Encoding them as policy-level knockouts removes ambiguity from the analyst's queue and concentrates human review on the genuinely uncertain cases. Knockouts run on the same policy engine that drives the rest of the case workflow.
How does AI reduce false positives in KYB screening?
AI in compliance is held to a higher bar than AI in sales or support. Explainability, auditability, and zero tolerance for material error are non-negotiable (Duna AI memo). ACAMS guidance on AI in sanctions screening places model validation, drift testing, and decision traceability at the centre of any deployment.
That bar makes large language models more useful for match assessment — entity comparison in messy text, close to what these models are trained on — than for impact assessment, which is policy-dependent and rule-based. The right architecture runs a deterministic policy engine for impact and applies AI to triage match likelihood, with every decision logged, attributed, and reviewable. McKinsey reaches the same conclusion: machine-learning models materially reduce noise in screening and transaction monitoring, but only when integrated into a governed compliance architecture.
Two design tests separate useful AI from theatre:
Can the analyst see why the model classified an alert as a likely false positive — and override it?
Can the institution backtest the model against historical decisions to show the regulator it does not drift?
Duna AI is built around these constraints, with virtual screening assistants that reduce false positives by approximately 70% across the queue (source: Duna AI memo).
What metrics prove false positive reduction is working?
Four numbers tell you whether a KYB review queue is actually improving:
False positive rate (and total false positives reviewed)
Total analyst time spent reviewing hits
Share of monitoring hits automatically ignored on profiles previously cleared
Screened or monitored entities per analyst, per period
Without those four metrics, "we reduced false positives" is a marketing claim. With them, it is a benchmark.
What does the regulator require for false positive handling?
Under the European Anti-Money Laundering Authority (AMLA) supervisory framework, operational since 2025, the test for an effective compliance programme is whether it was reasonable for the alert to fire and whether the firm can demonstrate defensible reasoning for the escalation or the dismissal. Every dismissed alert needs a recorded rationale. Firms that have moved their decision logic out of analyst memory and into the system are at a structural advantage under this standard.
What does Duna do to reduce false positives in KYB?
Duna is the AI-native business identity platform for KYB, KYC, AML, and lifecycle compliance. Plaid, CCV (Fiserv), Moss, and Bol use Duna to onboard businesses 10.6x faster and run their compliance teams 4.8x more productively.
Duna's screening and monitoring layer ships four production capabilities directly tied to the operational moves above:
Analysts dismiss false positives in bulk from the list view, with the screened entity and matched profile shown together so most dismissals do not require opening the detail page (in Decide).
Screening and monitoring workflows are separated; monitoring hits no longer automatically create a Review.
Once an analyst has marked a profile as a false positive, future hits for that same profile are automatically ignored — the Wolfsberg-recognised "Good Guys list" approach — preserving the original reasoning in the audit record while removing the recurring noise from the queue.
Policy-level knockouts on unambiguous sanctions designations remove ambiguity from the analyst's queue, running on Duna's policy engine.
On top of these, Duna AI virtual screening assistants reduce false positives by approximately 70% across the queue (Duna AI memo). Every decision is logged with the analyst's reasoning preserved, which is the defensibility standard the EU AMLA framework requires.
Frequently asked questions
What does KYB stand for? KYB stands for know-your-business. It is the regulatory process of verifying a business customer's identity, ownership, and risk profile before and during a commercial relationship — the business equivalent of KYC (know-your-customer). See Duna's guide to KYB verification for the full definition.
What is a false positive in AML screening? A false positive in anti-money-laundering screening is an alert that matches a customer or business to a sanctions, PEP, or adverse media record that is not actually about that entity. False positives waste analyst time but do not constitute a regulatory breach. False negatives — missed real matches — do.
Is a false positive a compliance failure? No. False positives are an unavoidable property of fuzzy-matched screening. A compliance failure is dismissing a true match (false negative), or dismissing a true match without a recorded rationale — a defensibility failure under the EU AMLA framework.
What is the typical false positive rate for sanctions screening? Industry research from McKinsey, PwC, and ACAMS places the typical false positive rate at 90–95% across sanctions, PEP, and adverse media screening.
Can AI eliminate false positives in KYB? No. AI can substantially reduce the time analysts spend on each false positive, particularly in match assessment. Duna's virtual screening assistants reduce false positives by approximately 70% without eliminating them.
