A manufacturer's website now qualifies inbound RFQs before a human ever replies

Representative examplesEvery case study in this library is an illustrative composite of the kind of engagement we deliver — written to show our method and standards, not to name clients.

Where they started

Custom manufacturing wins or loses work at the quote stage: the firm machines and fabricates bespoke parts for industrial buyers, and every quote begins with an RFQ that may or may not contain enough truth to estimate. The estimating team is three engineers whose time is the factory's scarcest asset, the website is a decade-old brochure that the owner refuses to replace before a larger rebuild, and the buyers who submit quote requests range from professional procurement teams to a purchasing officer forwarding a photograph of a drawing.

What it was costing

The quote-request form collected everything and nothing: 'need parts, urgent' with no drawings, no materials, no quantities. Engineers spent hours chasing details by email, and half the requests were for work the firm doesn't do.

What they could see

• Quote requests arrived as one-liners — 'need parts, urgent' — with no material, quantity, or drawing attached.
• Engineers spent hours emailing questions that every quote would eventually need answered anyway.
• Requests for work the firm doesn't do — volumes, materials, processes outside capability — occupied estimator time before anyone noticed.
• Quotes went out late, and the firm suspected buyers awarded jobs to whoever answered first.

The constraints we worked inside

• Engineering judgment is the product — the assistant gathers facts, engineers decide.
• Buyers attach files of wildly varying quality; the flow had to handle drawings gracefully.
• The firm's site was old; the assistant had to work within it, not demand a redesign.

What had been tried before

The quote form's mandatory fields were expanded to force structured detail.

Buyers who wanted a fast answer typed 'n/a' into everything or abandoned the form; the ones who needed help most submitted least.

One estimator was assigned to pre-screen every incoming request by email.

Screening consumed the estimator's quoting hours, and his quote output dropped enough that the shop noticed before the sales pipeline did.

What we proposed

We proposed a conversational assistant embedded in the existing site that mirrors the estimator's first-call checklist — process, material, quantity, tolerances, deadline — one question at a time, and handles attached drawings gracefully instead of demanding perfect files. Out-of-capability requests are told so directly and pointed elsewhere, because goodwill from an honest no beats a dead lead. Qualified requests arrive in the CRM as a structured brief with files organized, so the estimator's first reply can be a number rather than a questionnaire. Engineering judgment stays where it was; the assistant only gathers facts.

Just as important is what we ruled out, and why:

• A gated buyer portal with login — Buyers won't register an account to send an RFQ; gating intake would have filtered out exactly the urgent, under-specified requests worth winning.
• Replacing the old website — The owner has a rebuild planned for next year; the assistant had to live inside the current site, not hold it hostage.
• A rigid dropdown-only intake form — Tolerances, materials, and finishing requirements don't fit dropdowns; forcing them would recreate the 'n/a' problem with extra steps.

How the work ran

Delivered by the systems pod — automation specialist + engineer over 7 weeks, with working increments reviewed with the client every week.

The stack, and the reasoning

LLM assistant platform

The questions are free-form and the answers messier still; a conversational layer handles 'it's like this part but flange-mounted' better than any form.

Structured intake flow

Under the conversation sits the estimator's checklist — process, material, quantity, tolerances, deadline — so the brief that emerges is complete, not merely chatty.

File handling

Buyers attach drawings of wildly varying quality; the flow organizes and previews what arrives instead of rejecting the imperfect formats engineers actually receive.

CRM routing

Qualified briefs land in the CRM the estimator already checks, flagged by capability fit, so out-of-capability requests exit politely and early.

What went wrong

How we worked together

Cadence

A Thursday 45-minute working session with the lead estimator — reviewing real transcripts, tuning questions — with written notes after each for the two who couldn't attend.

Client side

The lead estimator owned the qualification checklist; the sales manager decided which polite rejections were allowed and how they were worded.

Decisions

Transcripts settled arguments: when the team disagreed whether a question was worth asking, we pulled a real one and read it aloud.

They provided

A set of past RFQs with their outcomes for tuning, estimator time every Thursday, and the capability matrix the engineers carry in their heads.

What changed

The headline: rfq-to-quote conversion (requests that became quotable briefs), first 90 days — +47%, read from CRM pipeline records. A second check: engineer time per qualified rfq, on average at −3.2 hrs.

The engineers read briefs instead of conducting interrogations; the first reply to a qualified RFQ can now contain a number, which changed how buyers negotiate. The polite rejections quietly improved the firm's standing, and the shop floor noticed — planners say the quotes arriving now describe work the machines can actually run. Estimators trust the intake because they wrote the checklist it follows, and the lead estimator has started editing the questions himself, which is the outcome we wanted more than any number.

The result was read from CRM pipeline records against the pre-engagement baseline over the stated window, with a guardrail check on engineer time per qualified rfq, on average. Where platform-reported numbers and business outcomes differ, this record says which layer it is quoting.

What they own now

• The assistant embedded in the firm's existing website, with the qualification flow and its question logic.
• The accepted-formats runbook for file handling, including the conversion fallbacks.
• The capability routing matrix in the CRM, editable by the sales manager, who also holds the assistant workspace credentials.
• A transcript archive with search, so estimators can revisit how buyers phrase things.
• A monthly scheduled check comparing qualified briefs against quote outcomes.