Entity Resolution + AI

The same fan, recognized everywhere — even when the data disagrees.

Matching records by rules alone leaves money on the table: the same fan, spelled two ways across two systems, often scores just below the bar to merge automatically. athvin pairs a statistical matching engine with an AI adjudicator that reviews exactly those borderline cases — merging the ones it can confidently confirm, and flagging the rest for a human. Here is how it works.

Statistical matching

First, the math decides what it can.

Every candidate pair gets a match-confidence score. High scores merge automatically and the lowest are dropped. In between sits the gray zone — pairs too risky to auto-merge, yet too similar to ignore.

AI adjudication

Then the AI does the extra leg.

Each gray-zone pair goes to Claude, which weighs the records and the model's own evidence and returns a verdict with a confidence score — auto-merging the clear matches, routing the genuinely ambiguous ones to a human, and rejecting the rest.

Inside one decision

How a borderline pair gets judged.

The model shows Claude the two records and a field-by-field breakdown of what supports or argues against a match. Watch a confident merge and a careful rejection.

Borderline pair · same fan?
Ticketing
NameRob Brown
Date of birth1985-03-22
Phone(555) 123-4567
RecordSeason-ticket holder
Donations
NameRobert Brown
Date of birth1985-03-22
Phone(555) 123-4567
Record$5,000 donor
Model evidence · match weight (bits)
Phone
+14.2exact
Date of birth
+12.0exact
Surname
+8.5exact
First name
+0.6Rob → Robert
Email
-2.1differs
Statistical match: 72% — below the 95% auto-merge threshold → not merged
Claude adjudication

“Rob” is a common short form of “Robert.” An identical date of birth and an identical phone number are powerful corroboration; different email providers usually mean two accounts for one person, not two people.

Same person — merge · 94% confident
Date of birthPhoneNickname
The impact

More records unified — every call on the record.

Statistical matching does the heavy lifting; the AI recovers the merges it would otherwise leave on the table. Figures below are illustrative of a single resolution run.

0
auto-merged by statistics
+0
recovered by AI in the gray zone
0
flagged for human review

Nothing merges silently. Every AI decision is written to an immutable audit trail — the two records, the model's evidence, the verdict, the confidence, and whether it was applied — so your team can review any call at any time.

PairDecisionConfidenceStatus
ATH-2847 ↔ TKT-5519Merge94%applied
ATH-1180 ↔ CRM-3304No match96%logged
ATH-0942 ↔ DON-7781Review78%queued for human
See it on your data

Find the fans you didn't know you already had.

We'll connect a sample of your systems and show you the duplicate fans athvin resolves — including the borderline merges your current tools miss.

Request a DemoSee the Full Pipeline