Quotes in one morning, priced from actual job history

The shop's estimating queue ran through two senior estimators with forty years of combined experience — and a four-day average turnaround. Every day of quote latency cost bids: buyers increasingly award to the first credible number, and the shop was arriving last.

The knowledge to quote faster already existed. A decade of job costing, routings, and actuals sat in the ERP. None of it reached the quoting desk in usable form.

We spent the two-week scoping phase at the estimating desk, timing real quotes and tracing where the hours went. Most of the time wasn't judgment — it was hunting: finding the comparable past job, its routing, and what it actually cost against estimate.

The scope that came out of it was deliberately narrow: don't replace the estimators' judgment; eliminate the hunting. Acceptance criteria were written and signed before the build started — quote-ready drafts for 80% of repeat-family work, in under an hour, priced within 5% of what the estimators would produce by hand.

The system reads an incoming RFQ, matches it against ten years of job history by geometry, material, and tolerance callouts, and assembles a draft quote: suggested routing, cycle-time estimates from actuals rather than standards, and margin flags where past jobs of that family ran over.

Estimators review every draft. The interface shows why each comparable was selected and where the numbers came from, so a thirty-year estimator can overrule it in seconds — and the system logs the correction for the next quote.

Quotes that took four days go out in one morning. Bid volume is up 22% with the same two estimators, and win-rate tracking against the baseline we recorded during scoping shows quality held while speed tripled.

The system runs under managed operations: measured monthly against the signed acceptance criteria, retrained as new job families appear, and maintained without an internal IT hire.