Name Overlap at Global Scale

Common names produce enormous volumes of spurious matches. There are millions of people globally named “Ahmed Hassan,” “Maria Garcia,” “Li Wei,” “John Smith,” or “Vikram Singh.” Sanctions and PEP lists contain hundreds of thousands of names. Without secondary identifiers, every customer with a common name matches multiple records.

Transliteration & Language Variance

Arabic names can transliterate into English in 10+ different ways (Mohammed / Mohamed / Muhammad / Mohammad). Chinese names appear in Simplified, Traditional, and pinyin. Russian names vary between Cyrillic and multiple Romanisations. Every variant produces a potential match without identity resolution.

Matching Algorithms Tuned for Recall

Screening vendors prioritise finding the real hit over reducing noise, because missing a sanctions match is a far worse outcome than producing a false positive. Algorithms use fuzzy matching, phonetic matching (Soundex, Double Metaphone), token variation, and edit-distance techniques that deliberately over-match.

Sparse Customer Data

If the bank only has the customer’s name and country of residence, there is no way to disambiguate common-name matches. Analysts have to escalate what richer data would have cleared in seconds.

Stale or Over-Broad Source Data

Vendor PEP databases sometimes retain individuals for years beyond relevance, broad role categorisations that aren’t actually PEPs per FATF R12, or low-relevance entries that the bank’s policy would not treat as triggering EDD.

Secondary Identifiers in the Customer Record

The single most effective lever. Capture DOB, nationality, country of birth, address, tax-residency, and — where permitted — identity-document number or partial. The richer the customer record, the faster an analyst can clear partial-name matches on DOB or nationality alone.

Name Variant Handling in Customer Data

Capture known name variants at onboarding — alternate spellings, transliterations, maiden names, former names. The more variants the bank already knows, the fewer spurious alerts its screening produces on legitimate variations of the same customer.

Match Threshold Calibration

Each screening rule has a match-score threshold. Tuning this is a trade-off: lower thresholds catch more true hits but generate more false positives; higher thresholds reduce noise but risk missing matches. Calibration is risk-based — tighter thresholds on low-risk books, looser on high-risk books — and must be documented with a rationale memo that regulators can review.

Known-Cleared Lists (“White Lists”)

Where an alert has been fully investigated, dispositioned as a false positive, and documented with rationale, the specific customer-to-alert pairing can be added to a known-cleared list. Future identical alerts suppress the re-review. Governance: periodic re-audit of known-cleared lists to ensure the cleared status still holds.

Deduplication Within the Vendor Database

A single individual frequently appears as multiple records across vendor databases — under different name spellings, different roles held, different reporting years. Dedup logic collapses these into a single match entity, reducing alert count without loss of coverage.

Role-Relevance Filtering for PEP Database Entries

Most PEP databases include individuals whose role does not meet the strict FATF R12 “prominent public function” test — low-level civil servants, local-government officials, minor SOE executives. Banks apply role-relevance filters: alerts on roles below the bank’s policy threshold either suppress entirely or route to a separate lower-priority queue.

Geography-Based Filtering

A customer with no footprint outside India is very unlikely to be a match for a sanctioned individual in Syria. Geography-aware matching applies nationality and residency filters to prune alerts where the geographic overlap between customer and sanctioned entity is implausible.

Vendor & Algorithm Diversity

Running two parallel vendors (for example Dow Jones and World-Check, or LexisNexis and Refinitiv) provides redundancy and allows each vendor’s specific false-positive patterns to be filtered. Differences between vendor output are themselves valuable signal.

Machine Learning & AI-Assisted Triage

Tier-1 banks and major KPOs increasingly use ML models trained on the bank’s historical dispositioning decisions to pre-score alerts before they reach analyst queues. High-confidence clear predictions are auto-dispositioned with human spot-check; low-confidence ones are escalated with features highlighted. These models must be auditable, reviewable, and not operating in a black box — a regulatory expectation explicitly called out by the FCA and DFSA.

Alert Queue Structure

Most tier-1 KYC operations run tiered queues: high-severity (sanctions hits, Foreign PEP matches) to senior analysts; medium-severity (Domestic PEP, adverse media Tier 2–3) to standard analysts; low-severity (role-relevance alerts, adverse media Tier 4–5) to a lower-priority queue. Volume rebalancing between tiers happens dynamically during the day.

SLA Expectations

Alert SLAs differ dramatically by type. Sanctions alerts typically have a 4-hour SLA at onboarding (faster for payment screening mid-flight). PEP alerts have 24–48 hour SLAs. Adverse media alerts vary between 48 hours and 5 business days depending on source tier. Missing an SLA on a sanctions alert is a material internal incident at every tier-1 bank.

Quality Assurance (QA)

Two QA dimensions matter: (a) false-positive clears that should have been escalations — sampled at typically 5–10% of analyst output and reviewed by Senior Analysts; (b) escalations that should have been cleared — monitored to catch analysts who over-escalate to avoid responsibility. Both patterns are tracked at individual analyst level and feed performance reviews.

Productivity vs Accuracy Balance

The productivity target at an entry-level KYC analyst role is typically 50–120 alert dispositions per day, depending on alert type and book complexity. Accuracy targets are typically 98%+ true-positive retention and 98%+ correct disposition. Productivity and accuracy are both tracked and reported to compliance management monthly. Analysts who hit productivity targets but miss accuracy rarely last; accuracy is the non-negotiable floor.

Scenario 1 — Textbook DOB clearance at Barclays GCC

Priya, a KYC analyst at Barclays GCC Mumbai, receives a sanctions alert: 78% name match between her customer (a Gurgaon-based tech consultant) and a sanctioned individual from a Central Asian jurisdiction. The customer’s DOB is 1989. The sanctioned person’s DOB is 1952. Nationality is also different.

Workflow: Two independent secondary identifiers disagree materially. Documented false positive with rationale memo. Customer added to known-cleared list for this specific sanctioned-party pairing. Alert closed in under three minutes.

Scenario 2 — Partial match requires human judgement at JPMorgan London

A KYC analyst at JPMorgan London receives an 82% name match, same nationality, DOB within two years, but different middle name and no photograph available. The alert is against a Foreign PEP.

Workflow: Single identifier disagreement (middle name) is not sufficient on its own given the PEP severity. The analyst escalates to Senior Analyst; Senior Analyst requests additional customer data (spouse name, employer, residence history) to build the second disagreement. Additional data confirms the customer is a different individual. Documented clearance with senior sign-off.

Scenario 3 — Known-cleared list refresh catches a data change at BNY

A customer at BNY has been on the known-cleared list for two years against a particular PEP-database entry. During the periodic re-audit of known-cleared lists, the PEP-database entry is refreshed and the match score rises to 91% with a new address that now overlaps with the customer’s address history.

Workflow: Cleared status automatically suspended. File re-opened for full investigation. Additional data confirms it is still a different individual, but the audit trail now includes the refresh, the re-investigation, and the renewed clearance rationale. This is exactly why known-cleared lists need periodic re-audit.

Scenario 4 — Pressure-test clearance at eClerx

A senior analyst at eClerx Mumbai reviewing QA samples notices a pattern: one junior analyst is clearing PEP partial matches at a 97% rate, versus the team average of 85%. Is the analyst unusually efficient, or clearing too aggressively?

Workflow: Senior analyst re-reviews 20 randomly sampled cleared alerts. Finds 3 clearances that do not have documented rationale sufficient to support the decision. Coaching applied, outputs re-checked at higher frequency for 30 days, clearance rate normalises. This is QA working as designed — detecting productivity-driven over-clearance before it becomes a finding.