Sanctions screening

Use case · Sanctions screening

One name on the list. Fifty ways to write it.

A single sanctions entry carries dozens of aliases, recorded from intelligence in whatever script it arrived in. Your customer types exactly one of them.

10–50aliases carried by a single sanctions list entry, across scripts and romanizations.OFAC SDN structure

$968MOFAC's largest single settlement, for screening failures at scale.Binance, 2023

95%+of screening alerts are false positives an analyst still clears by hand.Industry benchmark

Where the check runs

Twice: at onboarding, and every night after.

Every new customer is screened against OFAC SDN, UN, EU, and UK lists at account opening. Then the lists change, and the entire book is re-screened against the delta. Both checks come down to the same primitive: is this name that name.

Today

Rules against a moving list.

The list says Мухаммад Хусейни. The application says Mohammed al-Husseini. Exact match sees nothing.
Transliteration tables cover the spellings someone thought to add.
Loosen fuzziness to catch variants and alert volume explodes.
Every vendor scores differently, so thresholds never transfer.

With Onoma

A calibrated score per pair.

Cross-script pairs are first-class: Latin, Cyrillic, Arabic, Greek resolve in one space.
0.91 means the same thing on every pair, so auto-clear and auto-escalate bands hold.
Evidence ships with every score. Your auditors can read why.
Batch endpoint re-screens the book overnight, 500 pairs a call.

In the flow

Nightly re-screen, one call at a time.

Batch endpoint. Screen the delta of every list update against your entire book.
Alias-aware. Score the applicant against every AKA on the entry, take the max.
Evidence output. Attach the score breakdown to the case file.
Deterministic. Same pair, same score, every run. Regulators like that.

$ curl https://onoma.kineticpartners.org/api/v1/match \
  -H "Authorization: Bearer sk_live_..." \
  -d '{ "pairs": [
    { "a": "Mohammed al-Husseini", "b": "Мухаммад Хусейни" },
    { "a": "Mohammed al-Husseini", "b": "محمد الحسيني" },
    { "a": "Mohammed al-Husseini", "b": "Muhammad Hussaini" }
  ]}'

{ "results": [
  { "score": 0.92, "verdict": "match", ... },
  { "score": 0.90, "verdict": "match", ... },
  { "score": 0.95, "verdict": "match", ... } ] }

Twice: at onboarding, and every night after.

Rules against a moving list.

A calibrated score per pair.

Nightly re-screen, one call at a time.

Screen against every alias, not the one you happened to store.