Check a name pair
Compare the name on record (Aadhaar / PAN / customer input) against the name on the bank account.
Matching rules
How the score behind every decision above is computed — for anyone reviewing a REVIEW-queue item or deciding whether a threshold needs tuning.
Score bands
| Band | Score | Meaning |
|---|---|---|
| APPROVE | ≥ 80 | Structural variation only — word order, salutations, father's-name suffixes, spacing, or one dropped middle name with at least two full words still matching exactly. |
| REVIEW | 60–79 | Plausible but not provable from the strings alone — initials vs. an expanded name, phonetic/spelling variation, single-char typos. Routed to a human. |
| REJECT | < 60 | Different first name, different surname, or too little overlap to say anything useful. |
Reason codes
| Reason | Example pair | Score | Decision |
|---|---|---|---|
EXACT_TOKENS | ASHWIN CK / CK ASHWIN | 100 | APPROVE |
SPACE_VARIATION | ASHWIN CK / ASHWINCK | 100 | APPROVE |
MIDDLE_NAME_DROPPED | ASHWIN CHANDRAN KRISHNAN / ASHWIN KRISHNAN | 85 | APPROVE |
TWO_WORDS_DROPPED | ASHWIN CHANDRAN KRISHNAN MENON / ASHWIN KRISHNAN | 78 | REVIEW |
MANY_WORDS_DROPPED | ASHWIN CHANDRAN KRISHNAN MENON PILLAI / ASHWIN KRISHNAN | 70 | REVIEW |
INITIALS_VS_EXPANDED | S K MISHRA / SATISH KUMAR MISHRA | 69 | REVIEW |
PHONETIC_VARIATION | SREEJITH / SRIJITH | 69 | REVIEW |
SPELLING_TYPO | SURESH KUMAR / SURESJ KUMAR | 69 | REVIEW |
PARTIAL_MATCH | MOHAMMED / MUHAMMAD | 69 | REVIEW |
INSUFFICIENT_OVERLAP | ASHWIN / ASHWIN KRISHNAN | 34 | REJECT |
NO_MATCH | ASHWIN CK / PRIYA MENON | 9 | REJECT |
EMPTY_NAME | either input normalizes to no words | 0 | REJECT |
The strict-spelling policy
Strict spelling is on by default, meaning no purely phonetic or typo match can reach APPROVE, no matter how close the two names look. This is deliberate: for money movement, "Rajeev" and "Rajiv" — or "Mohammed" and "Muhammad" — may genuinely be two different people, so a spelling difference inside a word is treated as evidence to escalate to a human, never as evidence to auto-approve.
The honest cost: every spelling/phonetic variant is customer friction (a human has to confirm a payment that probably is fine). This trade-off protects against auto-releasing a payment to a different person who merely has a similar-sounding name.
Limitations
- Latin script only. Names in Devanagari, Tamil, or other scripts are not transliterated or compared.
- Single-word full names carry little signal. A one-word name matched against a multi-word name (e.g. ASHWIN vs. ASHWIN KRISHNAN) is scored as insufficient evidence and lands well below approval, by design.
- Initials cannot be safely expanded. The matcher can only check that an initial could stand for a given word, never confirm that it does — initials-heavy names (e.g. S K MISHRA) land in REVIEW rather than APPROVE, even when almost certainly correct.
- The reference sample set is small. 43 hand-labelled rows, heavily weighted toward one name across many variation types — a useful regression check, not proof the thresholds generalize to production volumes.