From RFQ to priced quote in one pass

Ask any estimator where the day goes and you'll rarely hear "calculating prices." The math is the easy part. The hard part is everything before the math: a PDF here, an email thread there, a spreadsheet with three tabs and a note that says "same as the Henderson job but taller."

That first mile — turning a human RFQ into a clean, structured spec — is where quotes stall. So that's the part we handed to an AI skill.

The shape of the problem

A rules engine like DriveWorks is brilliant once it has structured inputs. Give it material, grade, thickness, quantity, and a few options, and it will price consistently every time. The trouble is that customers don't send you that. They send prose.

The skill's only job is translation: plain-English RFQ in, structured spec out. No pricing, no opinions — just "here is what they're asking for, mapped to the fields your configurator already understands."

What "good" looks like

We hold the skill to three rules:

1. Never invent a value. If thickness isn't stated, it comes back as unknown, not a guess. A blank field is a question for a human; a wrong field is a quote you lose money on.
2. Show its work. Every extracted value points back to the sentence it came from. An estimator should be able to audit a spec in ten seconds.
3. Fail loud. Ambiguity gets flagged, not smoothed over. "Customer says 'heavy gauge' — confirm which" beats silently picking one.

Where it plugs in

The skill sits in front of the configurator, not inside it. RFQ arrives, skill drafts the spec, an estimator glances at the flagged fields, and the clean spec drops into DriveWorks for pricing. The engine does what it's always done well; the skill just gets it fed faster.

The result isn't "AI quotes the job." It's "the estimator stops retyping." On a good week that's the difference between forty quotes and sixty — with the same team.

The uncomfortable part

The skill is only as honest as the guardrails around it. Left unchecked, a language model will happily fill every blank with something plausible. That's exactly the failure mode that costs you margin. The engineering here isn't the model — it's the discipline of making it say "I don't know."

That's a theme you'll see across everything in the marketplace: the AI is the easy 80%. The last 20% — knowing when not to answer — is the whole job.