The Proposal That Ate March: AI for a Custom Machine Builder's Quoting Bottleneck

Reading time: 10 minutes. What you'll see: the same disease we measured at a Kokomo job shop, wearing a heavier uniform — because when your product is a custom machine, every quote is an engineering project you may never get paid for.

The company

Merlin Automation Systems builds custom assembly and test cells in Cambridge, Ontario. Twenty-four employees, about $7M in revenue, a bay of machines in various stages of build, and a customer list across auto parts, medical device assembly, and food packaging. Their machines work. Their commissioning record is clean. Their sales process is where the money leaks.

Every serious inquiry demands a concept proposal: station-by-station layout, roughed bill of materials, cycle time estimates, integration assumptions, a risk register, and a price. The OEM customer expects that station-by-station structure, and producing it takes Merlin's two senior applications engineers away from billable build work. An industry analysis of integrator estimating published by Mavlon describes line quotes as running through four distinct kinds of work and consuming roughly six weeks, with a hole in the tooling market exactly where those weeks go. Merlin's experience matches: a full proposal is weeks of elapsed time, and the engineers writing it are the same people the current builds cannot spare.

The measurement

Four proposals a month at $7,700 of engineering each is roughly $370,000 a year of senior engineering time spent on documents, of which seventy percent, by the win rate, produces no revenue. That unpaid engineering is priced into every machine Merlin sells, and it is also the ceiling on growth: sales cannot pursue a fifth opportunity in a month because the engineers physically cannot write a fifth proposal.

The math, including the row where this fails

The mechanism for improvement is retrieval, not replacement. Merlin has built dozens of cells. Somewhere in its project history is a station arrangement, a costed BOM, and a real cycle time close to almost anything a new RFQ asks for. Today that history lives in project folders and two engineers' memories. The build makes it searchable: the system reads the customer's specification documents, pulls the closest past projects with what they actually cost, and produces a first-pass station skeleton, draft BOM, and risk checklist for the engineer to correct and price. The engineer's judgment stays; the archaeology goes.

The negative row deserves its sentence: if a shop's engineers have no billable work waiting for the freed hours, and sales has no pipeline to feed a fifth proposal, this project buys capacity the business cannot use yet. We have advised builders in that position to fix pipeline first and automate second, and our published method says exactly where that call gets made.

What actually gets built

Nothing exotic. Inquiry documents land where they always did. A workflow extracts the specification into structured form, searches Merlin's own project history for the nearest neighbors, and assembles the draft: stations in sequence, components with the prices they actually carried, cycle assumptions flagged with their source project, and a risk register seeded from what went wrong before. The senior engineer opens a document that is eighty percent familiar and spends their hours on the twenty percent that is genuinely new, which is the part the customer is actually buying.

For a Canadian machine builder there is a funding dimension worth naming plainly. Work of this shape, building a retrieval and drafting system on your own engineering history, is technical development, and technical development is what NRC IRAP advisors fund at Canadian manufacturers every year. Our guide to what is actually open in Canadian AI funding covers the current programs, including the dead one the internet still recommends.

What this costs and how long it takes

The engagement structure: a free 30-minute scan of the proposal workflow, a $2,500 audit producing the measured baseline and a fixed-price scope with the fee credited toward the build, and an automation build from $7,500, live inside 75 days or the money back. Against the base scenario the project pays back inside seven months on engineering time alone, before any new-capacity revenue. Against the negative row, it is a capacity decision, and we will say so before you spend.

Want this measurement run on your proposal process? Start with the free 30-minute scan, or see how a full audit runs, number by number.