Job shops are automating estimating. Upload a STEP file, get feature recognition, cycle-time prediction and a price — in minutes rather than the two days it used to take an estimator with other work on.
For buyers this is mostly good news, and it changes two things you should measure.
What speeds up
Quote turnaround is the visible one. Where a manual estimate took 3–5 working days, automated quoting returns same-day. Your RFQ cycle compresses:
cycle = issue → quotes_in → compare → award
If quotes_in drops from 4 days to 1, and comparison and award are unchanged,
a two-week cycle becomes nine days. That's real, and it's the argument for
inviting the shops that have automated: not that they're cheaper, but that
they're answerable.
What gets noisier
The second effect is less discussed. An automated estimate is a model output, and model outputs fail differently from tired humans. A misread feature, an assumed tolerance, an unrecognised thread — these produce quotes that are confidently wrong rather than obviously wrong.
The signature is variance. If a supplier's quotes used to cluster tightly against your should-cost model and now scatter, that's worth knowing:
z = (x − μ_history) / σ_history
Same z-score as any other outlier check, applied per supplier rather than per part. A shop whose |z| used to sit under 1 and now regularly exceeds 2 has changed how it prices, and it's worth asking whether a human still reviews the number before it's sent.
The specific case to watch for is the low outlier. An estimator who misprices a job low usually catches it before quoting because they'll have to build it. A model has no such instinct, and the first time anyone notices is first article.
The check that still matters
Ask whether a person reviews the output. Not as a gotcha — most good shops do, and the ones that do will say so plainly. It's the same question you'd ask about any process change from a supplier, and the answer tells you how much weight the number carries.
Where the answer is no, adjust what you do with the quote rather than disqualifying the supplier. A fast unreviewed estimate is genuinely useful as a budgetary number and a poor basis for a PO.
The mirror image
Whatever your suppliers automate, you can too — and the same caution applies in reverse. An RFQ generated without someone reading it produces exactly the ambiguity that shows up later as a 77% spread, and the supplier will price the ambiguity in their favour, not yours.
How this gets solved
Both of the checks above are historical comparisons: this supplier's turnaround against their own past, this quote against the distribution of quotes you've had for that part family.
Neither is available if quotes arrive as email attachments and only the winner is filed. RFQTrace timestamps the invitation and the response — so turnaround is measured rather than remembered — and keeps every quote with its line items, so a supplier's variance is something you can see changing rather than something you notice two years late.