We tested it on their past jobs before building anything.
Signtech the Signmasters, a Christchurch signage manufacturer that works throughout New Zealand, asked whether AI could help with quoting. We answered with their own records first.
The question
Signtech’s estimators build each quote from the job’s brief and drawings. It’s skilled work, and it takes experience to do well.
The owners asked whether AI could draft quotes for the estimators to review, so they start from a draft instead of a blank page. Longer term, they want a hosted tool they can quote from on a job site.
What we did
- Checked all 4,335 of their job records before using any of them.
- Built a private workbench that compares what each past job was estimated to cost with what it actually cost, so every method gets scored the same way.
- Tested pricing rules, statistical calibrations and AI models from more than one provider, each on past jobs it hadn’t seen.
- Worked with Signtech’s owner and lead estimator to pull 150 past job sheets, some that came in close to estimate and some that didn’t, and tested against those.
What we found
Working from job sheets alone, AI-drafted quotes came close to the estimators’ accuracy, but didn’t beat it.
Used as a second check on an estimate, AI did better than chance at flagging jobs likely to run over.
Neither is good enough for AI to price jobs on its own, and we said so.
What phase one hands over
A written report, the dataset and analysis, a standard job-sheet format, acceptance criteria, an estimate of running costs and a scope for phase two.
What comes next
If Signtech signs off phase one, phase two is a hosted quoting tool built around the estimator. It drafts the quote, and the estimator checks it before it goes to a customer.
Why we work this way
A polished demo would have been quicker, but it wouldn’t have told anyone whether the idea works. Testing on a client’s own history first means everyone knows what a system can and can’t do before money goes into building it.