An RFQ lands in the estimator's inbox as an email with a PDF drawing attached, sometimes a STEP file, sometimes a phone photo of a drawing that lives on someone else's shop floor. The answer to whether the job makes money already exists somewhere in your building — in the traveler from a similar part two years ago, in the price history, in the mill certs from the last run of that grade. The estimator's actual job is finding all of it before the customer's deadline, and most of it is not where anyone remembers leaving it.
Quote speed is a competitive number, and most shops lose on it
Ask any estimator where the quoting day actually goes, and the honest answer is hunting: the drawing of the part you ran in 2022 that was almost this part, the routing someone already worked out once, the material price that has moved since the spreadsheet was last saved. The work that changes the number — the takeoff, the routing call, the margin decision — is a small slice of the elapsed time. The rest is retrieval, and retrieval is what stretches a quote from the morning the RFQ arrives to the end of the week.
The buyer's side explains why that stretch is expensive. An RFQ goes to several shops at once. The first credible number frames the whole decision, and the shop that responds in a day reads as the shop that will hit the delivery date. The cheapest way to win more work is usually not a lower price. It is the same price, two days earlier — and those two days are buried in retrieval, not in engineering. You do not have to take that on faith: pull your own quote log, mark the turnaround on the last fifty quotes and which ones won, and look at whether the fast list and the winning list overlap. Most shops have never run that comparison, and it is the one number in this whole subject you can check without trusting anyone.
Quoting Rev C when the customer is on Rev D
A drawing arrives by email in March and gets saved to the server. In June a new RFQ references the same part number, and the estimator quotes from the drawing on file — Rev C. The customer released Rev D in April. It added a weld callout and tightened a tolerance from ±.010 to ±.005, which means a different fixture and an added inspection step. The quote is wrong the moment it goes out, and nobody finds out until first article, when the margin has already left the job.
Revision control gets described as a discipline problem. It is more honestly a retrieval problem. The current revision exists — in a customer portal, in an email thread, in a folder three levels deep — and the failure is that the person quoting retrieves the wrong one. A system that indexes every drawing you hold with its part number, revision letter, and date can answer the question that matters at quote time: what is the newest revision of this part in our possession, where did it come from, and is the one currently open on the estimator's screen superseded. That last check — a flag that says this file has a newer sibling — prevents the most expensive kind of confident mistake a shop makes.
The MTR binder, and why heat numbers matter
Every heat of steel arrives with a mill test report: the mill's certification of chemistry and mechanical properties for that specific heat. The shop's traceability obligation runs through the heat number. This column came from heat 5A1234, and heat 5A1234's cert states the yield, the tensile, the carbon content. On structural, pressure, and aerospace work, a customer will ask for the cert on a part shipped months ago, and an auditor will ask with no notice at all.
In most shops those certs live in a binder or a shared-drive folder full of scanned PDFs named by supplier and delivery date — not by heat number, which is the only key anyone ever searches by. Retrieval means opening scans one at a time until the right number appears. Extraction changes the filing, not the paper: a system reads each incoming MTR, pulls the heat number, grade, and reported properties into searchable fields, and stores the PDF against them. The audit request becomes a lookup instead of a twenty-minute dig, and the answer arrives with the source document attached.
WPS, PQR, and the question the inspector asks
Weld procedure specifications, and the procedure qualification records that stand behind them, have the same structure. When a CWI asks which WPS covers a joint, the answer is a chain: this joint configuration, this process, this base metal and thickness range, therefore this WPS, qualified by that PQR. The documents exist. The chain between them usually lives in one senior person's head. Indexing the WPS library by process, material group, and thickness range lets anyone in the shop ask the question in plain language and get the governing document plus the qualification record behind it. That is also succession insurance — the shop's welding knowledge stops depending on the memory of whoever has been there longest.
What a private system looks like in a fab shop
Everything above is one architecture wearing different documents. RFQs, drawings, travelers, MTRs, and weld procedures are ingested into an index that runs on infrastructure the shop controls. Extraction reads the structured parts — heat numbers off certs, revision letters and dates off title blocks, quantities and due dates off RFQ emails. Retrieval answers plain-language questions and shows the page it answered from, so the estimator can check the work in one glance. Nothing gets pasted into a public chatbot, which matters because a customer's drawing usually arrived under an NDA signed long before anyone thought about where employees paste things. This is a description of how such a system is built; whether it is worth building in your shop depends on part count, mix, and how much of the quoting day currently goes to looking for things.
What this does not fix
It will not do takeoffs, price a part, or replace estimating judgment — it retrieves what the shop already knows, and if the routing on the old traveler was wrong, retrieval returns the wrong routing faster. Badly scanned drawings defeat extraction; a title block that is unreadable to a person is unreadable to software. If an incoming revision never gets filed at all, no index can surface it. And a shop with thirty active part numbers and one estimator who has personally run every one of them may not need any of this yet. The pain scales with part count and with turnover in the front office, and it is fair to wait until it actually hurts.
Where to start
Pull the last twenty RFQs from the inbox and write down two dates for each: the day it arrived and the day the quote went out. The average of those gaps is your baseline turnaround, it costs an hour to compute, and it is the number any system — software or otherwise — has to beat before it deserves a dollar.