Most manufacturers we sit down with can tell us, close to the cent, what it costs to make a part. Very few can tell us what it costs to quote one. Quoting sits in an odd blind spot: the function that decides which revenue you get and which you never see is almost never instrumented the way the shop floor is. Nobody runs a cycle-time study on the estimating desk.
When we do go looking, the figure that keeps surfacing is that manual quoting consumes 15–30 hours per week. That is most of a full-time position, spread across exactly the people you would least like to spend that way: the estimator who knows your process capability by heart, the engineer who can spot the tolerance that will bite you in month three, the sales lead who holds the customer relationship.
What those 15–30 hours are really buying
Every hour spent manually building a quote is an hour not spent on design collaboration, customer problem-solving, or closing high-value opportunities. It is worth being specific about what actually consumes that time, because the answer is rarely thinking.
It is re-keying quantities, materials, tolerances and delivery dates out of a PDF or an email body into a spreadsheet. It is hunting through last year’s files for what you charged a similar customer for a similar part. It is chasing a material cost from a supplier who has not replied yet. It is remembering, on day nine, that nobody followed up.
Some of those hours are genuinely valuable. Deciding whether a job fits your machines, or whether the customer is worth having, is judgement, and it earns its time. The problem is that the judgement is buried inside the clerical work, and the clerical work sets the pace.
The costs that never appear on a timesheet
The larger costs are invisible, precisely because nobody logs hours against work that never happened.
RFQs get missed because the pipeline is backlogged. Deals are lost to competitors who simply answered faster. And pricing errors quietly eat margin in both directions: priced too high, you lose the order without learning why; priced too low, you win work that costs you money to run — the more expensive mistake, because it looks like success on the way in.
This is why we think speed-to-quote now determines who wins the order. When two suppliers are both credible, the buyer is frequently not choosing between two quotes at all. The first response arrives while the requirement is still soft, shapes how the buyer thinks about the spec, and becomes the number everything else is compared against. The second quote is not competing on price. It is competing against a decision that has largely already been made.
None of it shows up on a timesheet. It shows up as a flat quarter nobody can quite explain.
First, measure two numbers before you change anything
There is no proprietary framework here, only an order of operations that starts with measurement — because the quoting projects we have watched fail were tools bolted onto a process nobody had counted first.
If you measure nothing else this quarter, measure two things. The first is elapsed time from RFQ received to quote sent. Elapsed, not touch time — calendar hours, including the weekend it sat in an inbox and the two days it waited on an approval. Touch time flatters you; elapsed time is what the customer experienced.
The second is harder and more uncomfortable: how many RFQs never received a quote at all. It is hard to produce in most shops, because unanswered RFQs tend not to exist anywhere: never entered, never assigned, never closed — they simply stopped. If you have to reconstruct one month by hand from mailboxes, do that. It is worth the afternoon.
Those two numbers describe the health of a quoting function better than any activity report, because they measure outcomes rather than effort. A team can be extremely busy and still be slow.
Second, separate the rules from the judgement
Once you know your baseline, take one product family and pull a quote apart into two piles.
The first pile is rules-based work: extracting the requirement from an inbound document, checking it against your capability list, pulling standard material and process rates, applying your own markup logic, generating and logging the document, and chasing the follow-up. These steps have right answers, and they do not improve when a senior person performs them — they only get slower.
The second pile is judgement: whether you want this customer, whether that tolerance is achievable on your equipment, whether to price for a full shop or an empty one, and when to break your own rules. These steps have no right answer, only an informed one.
Doing this honestly is usually the most useful hour of the exercise, and uncomfortable in two directions. A great deal of what feels like irreplaceable judgement turns out to be an undocumented rule living in one person’s head — a single point of failure, not a skill. And some of what a spreadsheet already automates turns out to be judgement that got buried in a formula years ago and has been quietly mispricing a category ever since.
Third, automate the rules and leave the judgement alone
Only then does the technology question become answerable — and much smaller than “should we do AI in quoting.”
An agentic system can own the first pile end to end: read the inbound RFQ in whatever form it arrives, extract and structure the requirement, check it against capability, assemble a draft against your documented rules, route it for pricing approval, send it, and run the follow-up. The estimator stops assembling quotes and starts reviewing and pricing them. That is where days become hours — not because anyone works faster, but because the waiting is gone.
To be plain about what this does not do: it does not make your quotes correct. It makes them fast and consistent. If your pricing logic is wrong, automation will apply that wrong logic more reliably and at greater volume than any human would — which is why the second step is not optional, and why we keep a human approval gate on anything that leaves with a price on it. Start with one product family and one intake channel, hold the two baseline numbers alongside it, and expand only where they move.
Where this work actually lives
This is the work we run through Interactive Intel, Alton Worldwide’s agentic-AI practice — a small, Miami-based team that designs, builds and runs production AI agents, every engagement led by Paul personally and scoped around a measurable outcome. Quoting is a near-perfect case for AI workflow optimization: high volume, mostly rules-based, and sitting directly on the revenue line.
The entry point is deliberately small — a fixed-price AI Opportunity Scan: one workflow, two weeks, the payback math in writing before you commit to anything larger. And if the math does not work for your shop, we will say so. But whether you speak to us or not: if quoting sits between your pipeline and your revenue, it is worth an hour of honest scrutiny this month.
Related reading: AI for medical practices · AI consulting for small business