The problem
A B2B parts and equipment distributor — around thirty staff, roughly A$5M turnover, based in South-East Queensland — was losing work to the clock. Trade customers rang or emailed through a request: a list of part numbers, some described in the customer's own words, a rough quantity, sometimes a photo of a worn component and a "what've you got that fits this?"
Two estimators handled the lot. Each enquiry meant reading the request, matching loose descriptions to catalogue part numbers, looking up the customer's pricing tier on the rate-card, checking for contract rates and volume breaks, and keying a quote into the system. A straightforward enquiry took twenty minutes; a messy one with substitutions took the better part of an hour. Quotes went out one to two days after the enquiry landed.
The owner's read was simple and, we think, correct: in trade supply the first credible quote usually wins. A one-to-two-day turnaround meant a competitor had often quoted first. Hiring a third estimator would clear the queue but wouldn't fix the speed — the same cycle would re-form a rung higher.
What we built
Six weeks: two weeks of scoping (we sat with both estimators and read through ninety past enquiries and the quotes they became), three weeks of build, one week of pilot on live enquiries with both estimators reviewing every draft. The deployed flow:
- Watched the shared sales inbox and the phone-enquiry log, and pulled each new enquiry in
- Extracted the requested items — part numbers where given, and free-text descriptions matched against the firm's catalogue with a confidence score per line
- Looked up the customer's account to apply the correct pricing tier, contract rates and volume breaks straight off the firm's rate-card
- Drafted a quote in the firm's standard format, with each line showing the source description, the matched part number, and the price it was drawn from
- Routed the draft into the estimator's queue, flagging any low-confidence match or unpriced item for a human to resolve
How it works
The core of the build was rate-card fidelity, not clever pricing. The AI never sets a price. It reads the enquiry, proposes a match to a catalogue item, and pulls the price that the firm's own rate-card already dictates for that customer and quantity. Every price on the draft is traceable to a rate-card line — the estimator can see exactly where each number came from.
Loose matches are the hard part of parts supply, so we made uncertainty visible rather than hidden. A confident part-number match goes through clean. A vague description ("the big grey filter, the one for the older units") is matched to the most likely catalogue item with a confidence score and one or two alternatives, and flagged for the estimator to confirm. Nothing is silently guessed.
The estimator reviews every quote before it leaves. They confirm the flagged matches, add substitutions or stock notes the system can't know, adjust for anything unusual, and send. The point was never to remove the estimator — it was to hand them a mostly-finished, rate-card-accurate draft instead of a blank screen.
The results
These are estimated outcomes from the pilot and the firm's own before-and-after figures for a business of this size — realistic for the volume, not headline claims.
| Before | After (pilot) | |
|---|---|---|
| Quote turnaround | 1–2 days | ~10 minutes (incl. estimator review) |
| Estimator time per quote | 20–60 min | ~5 min review |
| Estimator time saved | — | ~22 hours per month (est.) |
| Win-rate on quoted work | baseline | +15–20% (est., faster first quote) |
| Rate-card pricing accuracy on draft | — | high; every line traceable to a rate-card entry |
| Cost per quote drafted (model + infra) | — | ~A$0.40 |
The ~22 hours a month is roughly a full working week of estimator time returned across the two of them — time that went back into phone relationships and harder quotes. At a loaded labour rate of around A$60/hour that is on the order of A$1,300 a month, against a running cost measured in cents per quote. The win-rate lift, if it holds, matters more: the firm wins more because a credible quote reaches the customer first, not because the quote is priced any differently.
How the team owns it
We trained both estimators and the sales lead over the pilot week, and handed over plain documentation for how matches are made, how confidence flags work, and how to correct a mismatch so the system improves. No quote leaves without an estimator reviewing it — that is a rule in the workflow, not a convention. The firm can update the catalogue and rate-card themselves; pricing stays where it belongs, in the firm's own rate-card, not in the model.
If you run a quoting desk where the first good quote tends to win, this is a pattern worth a look — a readiness audit is usually the right first step, and our AI automation work picks up from there.