How NPI entity resolution works

From spreadsheet rows to review-ready NPI matches.

Four steps turn messy provider and organization records into confidence-rated matches you can audit before they move downstream.

  1. 1

    Upload your file

    Drag in any CSV: claims roster, referral list, CRM export. The first row should be headers; everything else is up to you. Uploaded tables can hold up to 50,000 rows, with large files imported in chunks.

    No template required. Files with inconsistent, real-world formatting are fine.

    Drop your CSV here

    First row is headers · large files import in chunks

    provider_roster.csv184 rows · ready
  2. 2

    Map your columns

    Tell it which columns hold names, addresses, phones, specialties. Any combination works, even just an organization name and a state. Unmapped columns still travel with the row as context for the agent.

    Pick a type filter if you need one: Auto, Type 1 (individuals), or Type 2 (organizations). The filter is enforced inside the engine.

    • providerorganization_name
    • addr_1address
    • citycity
    • zippostal_code
    NPI type
    AutoType 1Type 2
  3. 3

    Run the investigation

    Each row runs through registry-first discovery with nickname and former-name expansion, web corroboration for what the registry can't see, and NPPES verification for candidate NPIs before high-confidence results are shown.

    Easy rows skip the agent entirely. Deterministic fast paths resolve them instantly at full confidence.

    resolution sequence

    NPPES

    registry discovery

    Web

    public corroboration

    Pages

    evidence review

    Verify

    check digit + registry

    Candidates have to survive verification and the name-identity gate. Rows return a confidence-rated match or a clear no-match with reasoning.

  4. 4

    Review and export

    Results stream into the grid as they finish. Click any row for the full story: chosen NPI, matched registry record, confidence, candidates considered, and source URLs. Rerun anything you want a second opinion on.

    One click exports everything: your original columns joined with the verdicts, ready for the claims system or the data warehouse.

    Cascade Heart & Vascular

    Bend, OR · Type 2

    High

    Chosen NPI

    1663984528

    CASCADE HEART AND VASCULAR LLC · street match

    Also considered

    1934762505 — different org, same plaza

    1772049387 — name gate failed

    2 sources citedExport CSV ›

Inside the engine

The name gate
comes first.

The matching rule that keeps wrong answers out: the chosen NPI's registered name must plausibly denote your entity, meaning the same person or the same organization. Only then do location signals come into play, selecting among candidates and setting the confidence grade.

A same-name record at a different address can still be a match, since organizations may hold multiple NPIs across locations and people move. A different name at the same address is treated as a co-tenant until the evidence says otherwise.

Corroboration ladder — after the name gate passes

  • Street match→ high confidence
  • ZIP match
  • City + state match
  • State only
  • Phone matchdigit-for-digit

If your file carries no address signal at all, confidence is capped at medium.

The deliverable

One file, fully accounted for.

Every completed row exports with its verdict, reasoning, and sources, so the person who reviews the file six months from now knows why each NPI was chosen.

Run your first batch free

resolved_provider_roster.csv

  • Resolved NPI
  • Status
  • Confidence
  • Matched name
  • Matched address
  • Full reasoning
  • All candidate NPIs
  • Source URLs
  • Error detail (if any)
  • …joined to every original column

Watch your own file resolve.

Five rows free, results in minutes. Curious what it costs at scale?

Start for freeSee pricing ›