Where the Weeks Go: AI and the Engineered-Part Quote

Two to four weeks. That is how long it takes most suppliers to turn an RFQ for an engineered part or a custom system into a quote, and it has been that long for as far back as anyone in the industry can remember. Drawings went from paper to 3D CAD, ERPs went to the cloud, customer portals track every RFQ in real time, and the quote still begins the way it did in 1995: an experienced engineer opens the first PDF and starts reading.

Ask that engineer where the weeks go and the honest answer is not "costing." Costing a part, once you know exactly what it is, takes an hour. The weeks go to three other things, and AI changes each of them differently. It is worth taking them one at a time, because a supplier who understands where the time goes will buy the right thing, and one who doesn't will buy a faster cost engine and wonder why nothing changed.

The first place the time goes: reading

An RFQ package for a powertrain component or an automation cell runs forty to sixty documents. Drawing sheets, often across multiple revisions. A specification that constrains things the drawings never mention. A variant matrix. A quantity-break workbook. Commercial terms. Somewhere on sheet 14, a note that changes the finish called out on sheet 3.

None of this is hard to read. It is hard to read completely, and reading it completely is the estimator's first week. Every table gets retyped. Every note gets checked against the sheet it modifies. A harness drawing with no BOM block gets its wire list transcribed by hand.

This is the part AI removes almost entirely, and it is the least controversial part. Current models read a production drawing, including scanned legacy prints, notes, GD&T frames, and the non-tabular content that table extractors miss. What matters is that every extracted item points back to the sheet and zone it came from. The reason is simple. Every line the system produces, the estimator either trusts or checks. If the line links to the exact spot on the drawing it came from, checking it is one click. If it doesn't, the only way to check it is to go find it on the drawing by hand, which is the same reading the tool was supposed to remove. Without that link, the week doesn't come back; it just moves.

The second place: resolving

Reading produces facts. The quote needs one coherent picture, and the facts do not agree with each other. Which revision governs. Which of the six variants carries the optional sensor boss the customer mentioned in the clarification thread, not in the matrix. Whether the workbook's quantity breaks or the cover letter's are the real ones. What sheet 14 actually does to sheet 3.

Today this resolution happens in the estimator's head, over days, with emails to the customer spaced a week apart because each answer raises the next question. It is the part of the job that most resembles judgment, and it is where senior estimators earn their reputation.

AI does not make these decisions. It finds the questions, all of them, on the first pass, and then does the thing that used to take the estimator days: it goes looking for the answer in what the supplier already has. The revision record in PLM says which sheet is current. The customer's email from the last program says what they meant by the sensor boss. The 2023 quote, wherever it is stored, says what the operation cost last time. Each open question comes back as an exception with a proposal attached: the sheet, the conflict, what has to be decided, the answer the system recommends, and the evidence it found for it.

Seven exceptions on a forty-eight-document package is typical. The estimator's day starts with those seven instead of with the folder. Three get confirmed because the proposal and the evidence are right; two go to the customer in one email instead of four spread across the fortnight; two need the estimating manager. When the answers come back, the BOM re-resolves and shows the forty lines that changed, not all three hundred.

The third place: remembering

The sensor boss operation was last costed in 2023. That quote lives on a colleague's drive, in a format the shop no longer uses, and the colleague is on vacation. The estimator costs it again from scratch, differently.

Every supplier carries years of quoting history and almost none of it is usable at the moment it is needed. It is in PDFs, in email, in spreadsheets named final-v3, and mostly in the heads of the two or three people who have been there longest. When one of them retires, the shop does not lose a headcount. It loses the reasoning behind five years of prices.

This is the slowest change AI makes and the one with the longest tail. A system that stops at the drawing cannot help here at all. What the problem needs is something connected to where the supplier's knowledge already lives. That means the PLM and CAD systems that hold revisions and geometry, the ERP that holds rates and routings, the CRM that holds the customer's history, and the shared drives and workspaces where the past quotes and the email threads actually are. When a new RFQ arrives, the system reads it against all of that, not in isolation. The sensor boss gets costed from the 2023 precedent on the first pass, with a flag that the rate is three years old, instead of from scratch on day eight.

Then it learns. Every exception an engineer resolves and every quote they approve becomes structured history: which drawing features drove cost, what margin was applied, which questions the customer got asked, which proposals were accepted and which were overridden. That context is tenant-isolated, meaning it belongs to the supplier and is never used to train anything shared. The next similar RFQ starts from precedent. It compounds, and it compounds only from the day you start, which is the honest version of "why now." A shop that begins in 2026 has three years of usable, connected history by 2029. A shop that waits has three more years of PDFs.

What AI actually changes

The estimator does not change in any of this. Their four hours of real judgment on a package stay four hours, and nothing leaves without their name on it. What changes is that those four hours happen on the first morning instead of the last one, and that the RFQs which used to be no-bid on timing get read the same afternoon they arrive. Over a few months, that second change is the one a CFO can see.