We turned a 14,274-row spreadsheet into 413 customer cards — and found where the data was lying
The real face of merging customer data: one company under three spellings, a 32-character truncation in the source that both merges and splits companies, 1,344 suspicious rows that were only 21 names. From an insurance agency's ledger.
Normalising names got 14,274 rows down to 413 cards. Then: the agency's biggest client sat under three spellings (card grew from 361 to 1,164 rows), caught via shared plates and addresses rather than name similarity.
The real surprise: the source system truncated names at 32 characters, silently merging different companies and splitting the same one. Guards written for both directions.
1,344 suspicious rows collapsed to 21 names; one CAFE/KAFE pair deliberately left apart because lowering the threshold produced 629 false suggestions. Every import must reconcile with the file's own footer totals before it touches the ledger.