Pemberton & Rowe
A document intake assistant for a Chicago bookkeeping practice, extracting figures from client paperwork with low-confidence cases routed to a person.
AI Strategy & Development →The challenge
Month-end meant junior staff keying figures out of statements, invoices, and receipts that arrived in a different shape from every client and often a different shape month to month. The formats came from the clients' own banks and suppliers, so there was no intake process Pemberton & Rowe could standardise upstream.
Our approach
- 01
Built an extraction step with an explicit confidence threshold: clean, high-confidence reads populate directly, and anything below the line goes to a person with the source document alongside it
- 02
Set the threshold deliberately conservative at launch and moved it only after three months of live data showed where the assistant was actually reliable
- 03
Reconciled every extracted total against the source document automatically, so a silent extraction error surfaces as a mismatch rather than a plausible wrong number
Have a similar brief?
We can walk you through exactly how the Pemberton & Rowe engagement was scoped and priced.