Ask a job shop owner why quotes take four days to go out and the answer is almost always “we need another estimator.” Sometimes that’s true. More often, the estimator is fast — and the queue is the problem. RFQs get worked in the order they landed, interrupted by whoever calls loudest, with no distinction between a perfect-fit repeat job and a tire-kicker’s blanket RFQ sent to nine shops.
Triage — deciding what to quote, in what order, with how much care — is a separate skill from estimating. Most shops fuse the two jobs and pay for it in response time on exactly the RFQs they most want to win.
Why first-in-first-out fails
A queue worked in arrival order treats every RFQ as equal. They aren’t, on either axis that matters:
- Fit. An RFQ that matches your machines, materials, and typical tolerances can be quoted quickly and won profitably. One that’s marginal — an unfamiliar alloy, a tolerance at the edge of your capability, a quantity far outside your sweet spot — takes longer to price and converts worse.
- Intent. A named engineer with one drawing and a target date is buying. A purchasing portal blast with 40 line items and no contact name is comparison shopping. Both deserve an answer; they don’t deserve the same hour.
When the queue is blind to fit and intent, the marginal 40-line blast can absorb a full day while the perfect-fit job from a known buyer waits behind it. The buyer with the real job, meanwhile, is forming an impression: this shop is slow. Buyers who send the same package to several shops tend to anchor on the first credible response — not because they’re lazy, but because the first responder sets the reference point the others get compared against, and because responsiveness before the order is the best available signal of responsiveness after it.
The triage grid
Here’s a scoring pass that takes two minutes per RFQ and changes the whole week. When an RFQ lands, score it on two axes before anyone opens the CAD file:
Fit (1–3):
- 3 — In your wheelhouse: familiar material, processes you run daily, quantities in your range, a part family you’ve made before.
- 2 — Workable: one stretch element (new material, tighter tolerance, an outside process to manage).
- 1 — Marginal: multiple stretch elements, or quantities/specs that don’t match how you make money.
Value signal (1–3):
- 3 — Named contact, real company, drawing package, target date, plausible quantity. Or: an existing customer.
- 2 — Real but vague — a contact exists, details are thin, one clarifying email needed.
- 1 — Anonymous portal blast, no contact, signs it went to many shops.
Multiply the two numbers and the queue sorts itself:
| Score | Queue | Handling |
|---|---|---|
| 6–9 | Fast lane | Quote today or tomorrow. Full attention, sharp price, a phone call if anything’s unclear. |
| 3–4 | Standard | Quote within your normal window. One clarifying email is allowed before estimating hours get spent. |
| 1–2 | Decline or template | Send a polite no-quote, or a budgetary number from a template. No estimating hours. |
The bottom row is where the recovered capacity comes from. Most shops quote everything that arrives, which means their best estimating hours subsidize RFQs they’d be lucky to lose. A clean “we’re not the right shop for this one” costs two minutes, preserves the relationship, and frees an hour for a fast-lane job. Declining well is a quoting skill.
One refinement worth adding after a few weeks: a customer multiplier. An RFQ from an account that reorders regularly outranks its raw score — protecting repeat business is cheaper than winning new work, which is the same logic that drives reorder capture on the other side of the pipeline.
What “parsed and logged” buys you
The grid only works if every RFQ actually enters the queue. In most shops, RFQs arrive by email to two or three different people, and the queue lives in inboxes. That produces the classic failure: the RFQ nobody opened until Thursday because the estimator was at a machine and the backup was at a trade show.
The fix is one intake point where every RFQ gets logged with the fields triage needs: customer, contact, material, process, quantity, due date, drawing count. Pulling those fields out of emails and PDFs is exactly the kind of reading machines now do well — this is where AI earns its keep in a quoting workflow, extracting specs and dates so a human spends their two minutes on the scoring decision rather than the data entry. With intake automated, the triage grid stops depending on discipline and starts being how the queue simply works. That’s the architecture of an RFQ triage system: one intake, parsed fields, scored queue, and a response-time clock the owner can see.
Measuring whether it’s working
Three numbers, checked weekly, tell you if triage is paying off:
- First-response time on fast-lane RFQs. The one to brag about. Measure from RFQ arrival to quote sent, in hours.
- Estimating hours per quoted dollar. Triage should push this down: fewer hours on marginal RFQs, more quoted value per estimating week.
- Win rate by lane. If fast-lane win rate isn’t meaningfully higher than the old blended rate, the fit scoring needs recalibrating — the grid is a hypothesis you tune, not a commandment.
Notice what’s not on the list: total quotes sent. Quoting more is not the goal; quoting the right jobs faster is. A shop that declines 20% of inbound RFQs and answers its fast lane in four hours will beat a shop that quotes everything in four days — on win rate, on margin, and on the estimator’s sanity.
Where to start on Monday
You don’t need software to pilot this. Print the grid, put it next to the estimator’s monitor, and for one week score every inbound RFQ before anyone prices anything. Track first-response time on the 6–9 scores. At Friday’s end, look at three things: what got declined, what got quoted same-day that previously would have waited, and how the week’s estimating hours were distributed.
Most shops find the pilot pays for itself before any automation enters the picture — and the pilot data tells you exactly what the automated version needs to do. When the volume or the inbox chaos outgrows the printed grid, that’s the moment to wire intake, parsing, and the scored queue into the CRM you already run.