AI quoting speeds up the manufacturing RFQ. The price is still only as good as your cost data.

There’s a manufacturing-quoting AI wave on right now, and the pitch is the same from every vendor: stop losing deals to slow quotes. In April, at Hannover Messe, ServiceNow shipped a Configuration AI Agent for CPQ to general availability — a tool that lets a sales rep configure a complex manufactured product by describing it in plain language, managing bills of material across thousands of items. Paperless Parts, Tacton, and aPriori all sell some version of “respond to the RFQ faster with AI.” And on June 15, Podium Automation raised an $18 million Series A, led by Construct Capital with Andreessen Horowitz in the round, to compress custom control-panel builds from the industry’s usual three-plus months to under four weeks.

The common promise is speed: AI reads the drawing, AI drafts the quote, you respond before your competitor does. The speed is real. But it’s solving the half of the problem that was never where the money was.

TL;DR: AI quoting tools collapse the readable part of a manufacturing RFQ — interpreting the drawing, configuring the BOM, drafting the quote. That part is mechanical and genuinely slow; by Paperless Parts’ figure it can run up to two hours per part by hand. What the AI can’t do is decide what the job actually costs you to make. That number comes from your material and machine costs, your labor assumptions, and the manufacturability rules your shop prices by — data that is usually stale, inconsistent, or stuck in one senior estimator’s head. Automate quoting on top of that and you quote fast and wrong. The bottleneck was never typing the quote. It was the cost data underneath it.

The 2026 manufacturing-quoting AI wave

This isn’t one launch. It’s a category filling in across the same few months.

Move	Date	What it sells
ServiceNow — Configuration AI Agent for CPQ, GA at Hannover Messe	April 2026	Configure complex products in plain language; manage BOMs across thousands of items
Paperless Parts, Tacton, aPriori — AI RFQ/quoting features	Through 2026	”Respond to the RFQ faster”; interpret drawings, automate the quote draft
Podium Automation — $18M Series A (Construct Capital, a16z)	June 15, 2026	Connect requirements, electrical design, and fabrication into one flow; panels in under 4 weeks vs 3+ months

The category is current and it’s funded. The reason is a real pain: MFG.com industry research found 86% of manufacturers have lost a deal to slow quoting, and a large share still take at least a full day to produce a single quote. aPriori’s analysis pins part of it on the handoffs — every pass between sales, engineering, and estimating adds 12 to 24 hours. So the vendor logic is clean: kill the delay, win the deal. It’s not wrong. It’s just incomplete.

What the tools actually compress

The slow, mechanical part of quoting is reading the job and writing it up. That’s the part AI is good at, and it’s a legitimate amount of work. Paperless Parts puts manual quoting at up to two hours per part. In aerospace, where drawings are dense, the bulk of quoting time goes to interpreting the print rather than pricing it, and the cycle runs five to six days when customers want one or two.

So when the AI reads the drawing, pulls the features, maps them to a configuration, and assembles a draft quote in minutes, it is removing real hours. That’s the intake step — and removing a step is exactly where automation earns its keep. If the work stopped there, the wave would be straightforwardly good.

It doesn’t stop there. The quote isn’t the document. The quote is the price.

Where the quote actually breaks

A quote is a number, and the number is a function of data the AI doesn’t generate: what your material costs this week, what your machine time is worth, how many labor hours this geometry really takes on your floor, what your scrap rate does at that tolerance, and which features your shop can make cheaply versus the ones that quietly blow up a job. The AI reads the customer’s drawing. It does not know any of that. It looks it up — in whatever cost model and rule set you’ve given it.

And that’s the layer that’s broken in most shops. Material costs lag the market. Machine rates were set two years ago. The manufacturability rules — the ones that say this finish on that alloy doubles the cycle time, that customer’s tolerance callout means hand-deburring, we lose money on anything under 50 units — those mostly aren’t written anywhere. They live in the head of the estimator who’s quoted for fifteen years and “just knows.” When that estimator quotes, the judgment is applied silently. When the AI quotes, it isn’t, because it was never captured.

This is the same failure mode as autonomous field-service dispatch running on field data nobody maintains: the optimizer is solved, the data substrate isn’t, and the automation faithfully amplifies whatever the substrate already is. A junior estimator quoting from a stale cost sheet at least quotes slowly enough for someone to catch it. The AI commits to the same bad number in minutes and sends it out the door with a confident margin printed next to it.

The speed trap

Here’s the tension the vendor pitch papers over. Yes, MFG.com says 86% of manufacturers have lost deals to slow quoting. But the cost of winning a deal on a wrong price doesn’t show up in that statistic — it shows up six months later, when the job runs 30% over and the margin you quoted was never real. Speed loses you the deal once. A wrong cost model loses you money on every deal you win.

The tools that promise three-to-six-times-faster quoting — one shop cutting quote time from 30 minutes to 5, an aerospace manufacturer dropping from 2.5 hours to 25 minutes — are quoting that fast off the cost data you already have. If that data is good, you’ve removed a step and kept your margin. If it’s stale, you’ve built a machine for losing money at scale, very quickly, with great UX. Same tool. Different cost contract underneath. Speed is neutral; it amplifies whatever it’s pointed at.

The working version: write the cost-and-rules contract first

The order that works is the unglamorous one.

Get the cost data current and owned. Material costs, machine rates, labor standards — these need an owner and an update cadence, not a spreadsheet someone last touched before the last price spike. This is the boring half of the project and it’s the half that decides whether any of the rest matters.

Capture the manufacturability rules that live in someone’s head. Sit with the estimator who “just knows” and write the rules down: which features, tolerances, finishes, and quantities change the price, and by how much. That’s the part no vendor can do for you, because it’s your shop’s hard-won knowledge, not their product. It’s the same shape as the data contracts that decide whether any integration works — a written, owned statement of what the numbers mean. This is the discovery work I do before anything gets automated.

Then turn on the AI quoting. With a current cost model and explicit rules behind it, an AI tool that reads the drawing and drafts the quote is genuinely valuable — it removes the two hours of intake and keeps the margin intact. Bought in that order, the tool compresses a step. Bought first, it just automates whatever your cost data already is.

The operator read

The manufacturing-quoting AI wave is solving a real problem — the intake step was slow, and the new tools make it fast. That’s worth having. But the deal isn’t won or lost on how fast you produce the quote. It’s won or lost on whether the number is right, and the number comes from cost data and tribal rules the AI doesn’t own and can’t invent.

Quote faster on a good cost model and you’ve removed a step. Quote faster on a bad one and you’ve just found a more efficient way to underprice the hard jobs. The tool doesn’t decide which of those you get. Your cost data does. That’s the part worth fixing before you turn the quoting agent on.