The Rfq Problem

An Ionio briefing for the people who own quoting and new business. Machined parts. Weldments. Fabricated assemblies. Metal, cut to somebody else's print.

An RFQ lands at four in the afternoon and it is never one clean file. There is an eight page request with the commercial terms buried somewhere inside it. Fourteen sheets of dimensioned prints. A thirty six page quality document. An Excel line list where some of the part numbers do not match the print.

One person opens all of it. Usually the most experienced person in the building, who also runs the floor. He builds the routing in his head, saw to mill to deburr to anodize to inspection. He remembers a part close to this one from two years ago and roughly what it cost. Then he prices it. Two days later, sometimes a week later, a number goes back. Sometimes no number goes back at all.

You have lived this. A program part gets quoted once and you live with that number for five years while the price-down clause eats it every January. You lose a job by three percent and nobody ever calls to tell you by how much. An estimate sits half finished for six days because the plating shop has not written back. Rev C becomes Rev D and the takeoff starts again on a blank sheet. None of it appears in any report you have ever read.

So this is not a piece explaining what an RFQ is. You run them every day. It is about a mistake the whole industry has made in how it decided to automate quoting, and what that mistake has been costing you.

The mistake is treating quoting like a geometry problem.

For a decade, the software built for it has been CAD to cost. You upload a solid model, it reads the shape, it nests the part, it calculates machine run time from the features, and it hands you a price. Clever engineering, honestly built, and useless on most of what actually lands on your estimator's desk. A very good calculator.

But quotes do not go out late because the arithmetic was hard. They go out late because the answer already exists somewhere in your building and nobody can reach it.

That is a different problem. It is a memory problem. And no calculator solves a memory problem, however good the calculator is.

This piece is about that gap. Where it comes from, why it keeps costing you programs you never counted, and why two people in your building are the only thing holding it together. Then what changes when the price stops being assembled from memory under a deadline and starts coming from the jobs you have already run.

Start with why your estimator is slow, because everybody who sells you software assumes he is inefficient, and everybody who has ever done the job knows he is not.

In automotive, a program part gets quoted once. You live with that number for three to seven years, with an annual price-down clause written into the contract before you ever cut a chip. Underquote by five percent and you bleed for the life of the program, every year, on every release. Overquote and you lose it to a shop across town, and nobody ever calls to tell you by how much.

That is the whole shape of the decision. Asymmetric, irreversible, and made from a pile of paper under time pressure.

So the slowness is rational.

Every hour your estimator spends checking is an hour the quote sits on his desk not going out. The fast quote is the wrong one. The careful quote is late. He picks one and eats the cost of the other, and he has been eating it for so long that it no longer registers as a cost at all.

Now try to count what that is worth, and notice that you cannot. There is no line in any report called quotes we did not send. You track win rate on the quotes you submitted. Nobody tracks the RFQs that arrived and were never answered, because the estimator was on the floor for three days and the window closed.

It shows up as a flat revenue line nobody can explain. It shows up as a customer who quietly stops sending you packages, because you take a week to answer and the shop down the road takes a day. The most expensive thing in your quoting process is invisible in your own books.

Here is a question worth sitting with. Send the same print to five capable manufacturers, with the same quantity and the same delivery date, and ask each of them for a price.

You get five different numbers, and not because four of them are wrong. Different routings. Different work centers. Different cycle times for the same operation. Different scrap assumptions. Different views on whether the tight bore sitting near the process limit is going to bite.

Same part. Same print. Five defensible answers.

An estimate is not a calculation performed on a drawing. It is an interpretation of one, made by somebody who knows your building.

Somewhere in your operation a senior estimator looks at that print and makes a hundred judgment calls that appear nowhere on the paper. Whether this runs on the five axis cell or the old three axis with two setups. Whether the deburr is two minutes or five, given how this alloy behaves. Whether to carry eight percent drop on the billet or ten. Whether this customer always demands a first article and never says so in the RFQ. Whether you lost the last three jobs to this buyer on price and should sharpen your pencil, or won them all and should not.

None of that is on the drawing. All of it is in his head.

The industry has a name for the two ways to build a number. Engineered costing means you build a fully burdened rate for every work center, direct labour plus fixed and variable overhead per hour, then you estimate how long each operation takes from the features of the part, then you multiply time by rate down the routing and add material and margin. The rates are real, and they differ per work center, which is the point.

Remembered costing means you look at what a similar part actually cost you last time, and you adjust. No formula. Pattern matching against experience. On bespoke, low volume, messy work, estimators use the second one. Every time.

The industry half knows this about itself. The standard costing references in adjacent trades are built as learning curves, where the first unit takes longest and each repeat runs faster, and the serious manufacturers throw out the book times and substitute times pulled from their own operations. That is pricing from history, done badly, in a spreadsheet, by hand.

So see the estimate for what it is. It is the externalisation of one expert's understanding of how your operation makes this part and what it costs you when you do. Not a document. A judgment, written as a number.

Two things follow from that, and both are strategic problems rather than paperwork problems.

The first is fragility. When those people are on the floor, quoting stops. You cannot clone them and you cannot hire them, because a costing engineer who understands five axis work or powder metallurgy is not sitting on a job board. When they retire, the interpretation retires with them, and what is left behind is a folder of old quotes, which is the output of the judgment and not the judgment itself.

The second is inconsistency inside your own building. If interpretation lives in people, two estimators will price the same part with different logic. Your customer sees the variance across a program even when they cannot name it. You are not one supplier to them. You are a couple of individuals who share a logo.

The real asset in your quoting process is not the quotes. It is the interpretation layer underneath them. It is scarce, it is walking out of the door, and it has never been written down. Hold that thought, because it is the reason the technology conversation is about to matter.

Geometry pricing tools read the solid model and compute machine time from the shape. On clean CAD, a standard process and repeat volume, they work, and they work well. That is a real business and we do not compete with it.

They fall down in two places, and both places are where your work lives. The first is simple. Half the packages that reach an estimating desk have no usable CAD at all. They have a PDF of a print with the dimensions sitting inside the drawing views. A geometry engine has nothing to eat.

The second place is the deeper one. Geometry knows the shape of the part. It does not know that your five axis cell runs slow on the second shift. It does not know that this customer always demands a first article and never writes it into the RFQ. It does not know that you quoted this buyer twice, won once, and that he is sensitive on price. It does not know what you charged for a manifold in 2024, and it certainly does not know what that job actually cost you when it closed.

Your past jobs know all of it.

They price the part. You price your history.

That is the whole difference, and it is not a small one. A model can tell you how long a pocket takes to cut. It cannot tell you what you charged for the last part that had one, or what you gave up to win it. That number was never in the model. It is in a job folder, and it is the only number that has ever been proven right or wrong by your own bank account.

Before we go further, the three fixes everybody tries first, and why each one fails in exactly the same way.

Hire another estimator. Costing engineers who understand your process are scarce and expensive, and then it is months of knowledge transfer before they are useful. At the end of it the expertise still lives in a few heads, and now you have one more head to lose.

Build it internally. One manufacturer we know put ten people on this for eighteen months and spent about a crore and a half. It never worked. To build it you have to become a software company first, and that is a culture, not a project.

Buy an off the shelf quoting tool. Generic CPQ software does not understand how your parts are costed. It knows nothing about tooling amortisation over program volume, or press tonnage, or five axis setup strategy. Estimators try it, hit the wall in the second week, and quietly slide back to Excel.

Notice that all three failures share one root. The pricing logic is specific to your operation and it is written down nowhere. Every fix that ignores that fact fails the same way, and it fails after you have already paid for it.

For thirty years the constraint was that a print is a picture. A dimensioned sheet built for a person to read, with the numbers living inside the views, the GD&T in frames, the notes stacked in the corner, and a revision cloud thrown over whatever changed. When the definition is a picture, extracting it needs a person. The whole fragile human chain follows from there.

That constraint is gone.

Document and vision understanding got good enough, recently and quickly, to read a dimension string out of a drawing view, to pull a feature control frame with its datums intact, to find the delivery term buried on page 148 of a spec, and to notice that the port depth on sheet 9 was never dimensioned at all. Not perfectly. Well enough that a person checks the extraction instead of performing it.

That is one half of it. The other half has been sitting in your building for years.

Every quote you have ever sent is a labelled example. The ones you won and the ones you lost. Every closed job is the answer key, because it tells you what the part actually cost when the chips were on the floor, and not what somebody guessed it would cost on a Tuesday afternoon. Your quote history is a training corpus for your own judgment, and it has been quietly accumulating in a folder nobody opens.

Readable documents on one side. Your own accumulated judgment on the other. That pairing is new. It is the reason this is possible now and was not possible in 2019, and it is what everything below comes down to.

Be precise about where AI fits, because the wrong version of this is everywhere. It is not a chatbot bolted onto your ERP. It is not a portal that reminds you which RFQs are still open. It is not automation that quietly replaces your estimator and asks him to trust it.

The last one matters most. An estimator will not put his name on a number he cannot defend. He is the one sitting in the room when the job runs at a loss. He carries that personally, he will never hand it to a black box, and he should not. Any serious answer starts from that constraint instead of wishing it away.

So the model that works is not automation. It is approval-native.

The system drafts. The estimator reviews. He approves, edits, or rejects, at every stage. The machine carries the reading and the first pass at the number. The person keeps the judgment and the accountability. That single design choice is the line between a tool an estimating desk adopts and one they quietly stop opening.

Inside that constraint, here is what it does. It reads the whole package, all four animals, pulling every dimension and every GD&T callout off the prints, finding the commercial terms and the quality requirements that move the price, and reconciling the line list against the drawings. Then it builds the takeoff, every operation tied to a real work center in your building, with cycle times drawn from what your machines have actually done. Gaps get flagged rather than guessed.

Then it quotes the outside processing before it prices anything. That ordering is not cosmetic. Price first and the bought-out line is a placeholder, so when the real supplier number lands you re-price the whole quote. The supplier quote is an input to the price, not a step after it. So the anodize is quoted and awarded first, and when your estimator arrives at pricing, nothing is provisional. The number is whole on first view. That is how a careful estimator already works, and most software gets the sequence backwards.

Then it prices every line from named past jobs, and this is the part that earns trust or loses it.

Every priced line names the job it leaned on, the year, how close the match is, and what that job actually cost when it closed. The similarity percentage is the honest part, because it tells your estimator how far to trust the match. Click any line in the takeoff and the print highlights the exact place the number came from. Override a value and the cell shows it, with the original underneath and an undo. Nothing there is a trick. It is what you would demand before signing a number you did not build yourself.

And the gaps surface at your desk instead of in the customer's inbox. A port depth that was never dimensioned gets pinned on sheet 9 with a red flag, and your estimator either types the value or has the system draft the RFI. A bought-out line with no supplier quote cannot hide inside a total. The things that used to arrive as a lost job or a bad margin now arrive as a flag you clear on a Tuesday.

Here is what falls out of it, and it matters more than the cycle time. The interpretation layer finally gets captured. Every quote your estimator approves teaches the system how your operation thinks about routing, cycle time, scrap and risk. The judgment that used to live in one head becomes an asset that sharpens with every job and does not walk out of the door when he retires. Your second estimator inherits the reasoning instead of the templates.

And quoting stops depending on who is free. Turnaround stops being a function of whether your best person is on the floor this week. You answer more RFQs, you answer them faster, and you stop losing the packages you never counted, from customers who were never going to tell you why they stopped calling.

Take the block from our demo build. A hydraulic manifold in 6061-T6 aluminium, machined from billet, Revision D. A batch of 240 pieces against an annual usage of 1,536. PPAP Level 3 with a first article. Eight week delivery, Net 45, prices firm for thirty days.

Seven operations. The raw material comes off job 4102. The two five axis setups are priced off job 4471, a valve manifold sitting at ninety two percent similarity. Saw, deburr and final inspection carry standard times. The hard anodize goes outside, quoted by three platers already on the approved list, and awarded at $58.40 a piece, three dollars and sixty cents under the provisional estimate.

That saving lands in the price before the price is set, rather than turning up as a margin surprise after the job has already run.

The batch comes to $75,396. The annual comes to $482,534. The margin floor in this business is thirty percent, and this one clears it at thirty four.

The RFQ opened at 10:14 in the morning. The quote was ready at 12:23.

Be clear about what that sentence is and what it is not. That is our demo build, on a manufacturer we invented, with numbers designed rather than measured off a live customer. We are not going to dress a prototype up as a case study. The point of it was never the two hours.

The point is the shape. Every line on that quote traces back to a place on a print. Every price on it traces back to a job that actually ran, and to what that job actually cost when it closed. Your version of this is built on your parts, your work centers and your own history, and the first thing we do is show you that it reproduces the quotes you already sent.