A wholesale inquiry lands in your inbox at 6:12 PM on a Friday. Nobody sees it until Monday at 9. By Wednesday it has been matched to a spreadsheet. By Friday, the one person who knows the pricing finally gets to it.
The customer bought from someone else on Tuesday.
That is not a sales problem. It is a quoting problem, and it is the quietest revenue leak in small manufacturing and distribution. Nobody tracks the quotes that never went out. The CRM only knows about the ones someone finished.
Where the three days actually go
Ask a shop owner why quoting takes three days and they will say "the estimator is busy." That is true and it explains nothing. Here is where the time goes, from watching it happen in real shops:
Day one: inbox triage. The RFQ arrives as an email with three attachments, two of which are named "final_v2_REALLYFINAL.pdf." Someone has to open it, figure out what the customer actually wants, and decide whether it is even quotable. This takes an hour. It feels like five minutes.
Day one to two: document hunting. The drawings live in one folder, the bill of materials in another, and the pricing history in the estimator's head. Matching the request to past jobs means searching email threads from 2023. Every interruption resets the clock.
Day two to three: the pricing queue. The estimator is the only person who can price confidently, and they are also the person everyone needs for everything else. Your quote waits behind five other decisions. This is the bottleneck no software has ever fixed, because it is a people bottleneck wearing a paperwork disguise.
Day three: the actual quoting. The part everyone thinks takes three days takes about ninety minutes. Typing line items, checking margins, formatting the proposal, sending it. The skilled work is a fraction of the elapsed time. The rest is waiting, hunting, and retyping.
Read that list again. Three of the four stages are paperwork. One is judgment. Guess which one the customer is paying for.
What AI touches, and what stays human
This is the part most vendors get backwards, so let me draw the line clearly.
AI takes the paperwork. Reading the RFQ email and pulling out the requirements. Matching drawings and BOMs to your part library. Drafting the quote line items from your pricing history. Chasing down the missing spec with a polite follow-up email. Sending the quote the moment the estimator approves it, and following up if nobody replies in 48 hours.
Humans keep the judgment. Pricing authority stays with the estimator, full stop. So does the customer relationship. So does the call that says "we can do this, but not the way you drew it, and here is why." No customer ever chose a vendor because the quote arrived formatted beautifully. They chose the vendor they trusted, who also happened to respond fast.
The shops that get this right treat AI like the world's fastest apprentice. It does the fetching, the sorting, and the first draft. The estimator does the thinking. The shops that get it wrong try to remove the estimator, and they end up with fast quotes nobody trusts, which is worse than slow quotes everyone trusts.
This is not theoretical anymore. Uptool, an AI quoting startup, raised $6 million earlier this year specifically to pull quoting time from hours to minutes for small machine and fabrication shops. Their customers are not enterprises. They are shops like yours, drowning in the same inbox triage. The money is flowing here because the pain is real and measurable.
Under the hood, briefly
For the folks who want to know what the machine is actually doing, here is the pipeline in five stages.
Ingestion. Every RFQ email, attachment, and drawing gets pulled in automatically. CAD files, PDFs, BOMs, even photos of napkin sketches. Nothing waits for a human to open it.
Extraction. The system reads the documents and pulls structured data: dimensions, materials, quantities, tolerances, delivery requirements. Modern document parsing handles this well for standard formats. Handwritten notes and forty-year-old scanned drawings are where it earns its keep or shows its limits, and an honest vendor will tell you which of your documents fall where.
Matching. The extracted requirements get matched against your part library and pricing history, with a confidence score on each match. High confidence matches draft themselves. Low confidence matches get flagged for the estimator. This is the key design decision in the whole system: the machine never guesses on price. It drafts where it is sure and asks where it is not.
Draft and approval. The estimator gets a complete draft quote, not a blank screen. They adjust, approve, or reject. Every approval trains the matching for next time. This is the human-in-the-loop gate, and it is non-negotiable. Remove it and you have automation running blind. Keep it and you have backup that never gets tired.
Send and follow-up. The approved quote goes out branded and formatted, and the system watches for replies. No reply in 48 hours triggers a polite check-in. The estimator never thinks about follow-up again.
The whole pipeline writes back to whatever you already use. If your ERP is the system of record, the quotes land there. Nobody gets a new dashboard to ignore. For more on why that last point matters so much, read AI Implementation for Manufacturing. And if the cost of all this sounds scary, The $200 Stack will recalibrate your sense of what the tools themselves cost. Spoiler: the software is the cheap part. The setup is the investment.
The two-week pilot
Do not buy a quoting system. Run a pilot. Here is the shape that works:
Week one: measure. Pick one product line, the one with the most RFQs. Have your estimator log every quoting task for five days: what came in, how long each stage took, where it stalled. You cannot improve what you have not measured, and most shops discover their "three days" is really six hours of work spread across three days of waiting.
Week two: automate the paperwork. Connect the ingestion and extraction for that one product line only. The estimator still prices everything. Measure the same stages again.
The decision. Compare. If the paperwork stages collapsed and the estimator got hours back, you have your answer, with your own numbers, from your own shop. If they did not, you spent two weeks and learned exactly why, which is worth more than any vendor demo.
A pilot like this typically costs [CONFIRM] from us. The broader market runs from a few thousand for a guided setup to tens of thousands for a full build. Anyone quoting you six figures for a quoting pilot is selling you a different project than the one you asked for.
What to do Monday morning
Pull last month's RFQs and count the ones that went out more than 48 hours after they came in. Then ask your estimator what they were doing during those 48 hours. I already know the answer: everything except quoting.
Pick your slowest product line. Run the two-week pilot. One estimator, measured quote-lag before and after, no grand commitments.
Start with a free 30-minute diagnostic and we will map your quoting workflow together. You will leave knowing exactly where your three days go, whether you hire us or not.
— Greyson