AML screening API for sanctions, PEP, and watchlists

You send a name and whatever supporting details you already hold. Naiza cleans the name, builds its spelling variants, and compares it with more than 1,100 global sources. Every candidate is scored on how closely the name matches and whether the date of birth, nationality, and entity type agree.

The response tells you what to do next. Most customers come back AUTO_CLEAR and continue without anyone looking at them. The few that resemble a listed party come back REVIEW with the matched record attached, so an analyst can compare them side by side.

• Inputs: legal name, entity type (individual or organization), date of birth, nationality, aliases, and document numbers when you have them.
• Outputs: a decision (AUTO_CLEAR, REVIEW, or CONFIRMED), the lists that were hit, and a score and classification for each candidate match.
• Every call returns a screening ID. Store it on the customer record so support and compliance can find the result later.

Screen at the moment you have a legal name and a stable customer ID. Date of birth and nationality are optional, but they remove most of the noise on common names.

curl -X POST https://api.naiza.ai/api/v1/aml/screen \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "inputName": "Mohammed Abdullah Ahmed",
    "externalId": "cust_204118",
    "entityType": "INDIVIDUAL",
    "inputDob": "1987-03-09",
    "inputNationality": "SA"
  }'

A REVIEW result carries the candidate record, the source list, and a score. Here the input was written as "Mohammed Abdullah Ahmed" and the list entry as "Muhammad Abdallah Ahmad". Exact matching would have missed it.

{
  "id": "scr_abc123",
  "externalId": "cust_204118",
  "decision": "REVIEW",
  "listsHit": ["OFAC_SDN"],
  "matches": [
    {
      "watchlistEntityId": "we_987",
      "score": 92,
      "classification": "STRONG",
      "matchedName": "Muhammad Abdallah Ahmad",
      "source": "OFAC_SDN"
    }
  ]
}

One request runs against every category below. You do not pick lists per call or maintain list files yourself.

Arabic names appear in several Latin spellings across passports, bank records, and sanctions entries, so one person can look like three different people. Naiza matches across Arabic script and the common Latin spellings, so you do not need to generate variants yourself.

• محمد is written Mohammed, Muhammad, Mohamad, or Mohamed.
• عبدالله is written Abdullah, Abdallah, or Abd Allah, with or without a space.
• The Al- prefix appears as Al-, Al, El, or not at all.
• Bin and Ibn, and the order of given name, father's name, and family name, change between documents.

Supporting fields do the rest. A strong name match with a different date of birth and nationality is a weak candidate. The same name with matching identifiers goes to the top of the review queue.

Screening works best as a step in the flow, the same way you treat a payment authorization. Call it from your server at these points:

• Onboarding, once KYC gives you a verified legal name.
• Before the first payout or withdrawal.
• When a customer adds a new beneficiary or counterparty.
• When a customer changes their legal name or nationality.
• Business onboarding, for the company and each beneficial owner.

Decide the product behavior for each result before you go live. Analysts should only see the cases that need a person.

A clean result at signup only reflects the lists on that day. Naiza re-screens your existing customers automatically when the lists update. New possible matches open cases and reach your systems through webhooks, so a designation published tonight surfaces against a customer you onboarded last year.

Screening also connects to Naiza transaction monitoring. A payment to a counterparty who later matches a list shows up in the same case history as the original screening.

You can call the API for weeks before letting it hold anyone. Store the result, keep the customer journey unchanged, and look at the numbers: how many REVIEW results a day, how many turn out to be real, and how long each one takes to clear. Switch to enforcing once the queue fits the team you have.