Manufacturing Capabilities Site & Quote Funnel

A $480M contract manufacturer running eleven plants across three countries, machining precision components for aerospace, medical device, and semiconductor OEMs. It had grown by acquisition and greenfield expansion, and every plant had kept its own website. Two of them had not been meaningfully touched since 2014. They competed with each other in search results, and occasionally quoted against each other for the same part.

The commercial problem sat upstream of all of it. A procurement engineer at an OEM builds a supplier shortlist by searching for a process and a specification, long before anyone picks up a phone. These plants ranked for their own names — which nobody searches — and for very little else. They were losing at the stage of the sale where nobody knew a sale was happening.

And when somebody did make it through, the site asked for name, email, company, and a free-text message box. An estimator would receive "need a quote on a bracket," and spend the next three days by email working toward a number. Only 22% of inbound RFQs arrived complete enough to quote without a follow-up.

I went to the floor first, and not for the website. I wanted to be able to write about the work without embarrassing anyone who does it. Watching a five-axis cell hold a quarter-thousandth on a titanium part, and then watching the same part go through CMM inspection, is the reason the finished site leads with process capability instead of adjectives. You cannot fake that register, and this audience can hear the difference instantly.

Then the RFQ inbox. I read 200 inbound requests and tagged each one for what was present and what an estimator had to chase. The gaps were remarkably consistent — material, annual volume, tolerance callouts, certification requirement, and a drawing. Five things, every time. The form asked for none of them.

Ten interviews with sales reps and estimators produced the sentence the project turned on. I had assumed estimating was a capacity problem. An estimator corrected me: "We're not slow. We're waiting." Median time from RFQ to quote was nine days, and six of those were email ping-pong to assemble information the requester had on their desk the whole time.

Six interviews with procurement engineers at OEM customers, arranged through the sales team's relationships, explained the search behaviour. They query process plus material plus industry — never "contract manufacturer." They open six tabs and eliminate on the absence of specifics. One of them told me he closes any supplier page that claims world-class quality without printing a tolerance number, and I have not stopped thinking about it since. To this buyer, vagueness does not read as marketing. It reads as incapability.

The content model came before any layout, and it is the decision the whole project rests on. Every capability is a structured record: process, materials, tolerance range, part envelope, volume band, industries served, certifications held, and which plants run it. One model solved three problems that usually get solved separately — it is how an engineer evaluates a supplier, it is the shape of the long-tail query they arrive on, and it is what makes the page legible to a machine that is summarising the category for someone who never clicks through.

Lo-fi treated the capability page as a spec sheet you can read in fifteen seconds. I ran an actual fifteen-second test on paper with two of the sales reps: can you tell me the process, the material range, the tightest tolerance, and the certifications? If not, the page had failed regardless of how it looked.

The RFQ is where I was wrong, and it was an instructive kind of wrong. My first sketch was a better long form — nine well-organised fields instead of four bad ones, with helpful labels. I put it in front of two procurement engineers and both said the same thing within a minute: "I'd just email you." A form is not the task. The task is starting a conversation with enough on the table to get a number back.

So the flow got rebuilt around the artifact instead of the fields. Step one is the drawing — upload first, before anything else, because a drawing carries most of the five missing answers and because it is already open on the engineer's screen. Asking for it first also signals that you understand how quoting works, which is a brand statement disguised as a form field. The remaining steps are short, every field states the reason an estimator needs it, and a save-and-return link exists because RFQs get handed to a colleague halfway through.

The story work ran in parallel. This was a company whose people were genuinely its differentiator and whose website was stock photography of gears. We put a photographer in three plants for a week and came back with named machinists, real parts, and actual inspection reports. The copy split into two registers deliberately: capability pages written in the flat, precise voice of the estimators, and the trust content around them written in the voice of the plant. Precision where you are being evaluated, warmth where you are being chosen.

Consolidation was the unglamorous half. Eleven domains folded into one, plant sites re-cast as facility records hanging off the capability model rather than competing microsites, and a 301 map that carefully preserved the handful of legacy pages that had actually earned links over a decade.

One domain, one capability model, eleven plants as facilities rather than as websites. The navigation is organised the way the buyer's search is: by process, by material, by industry — with the plants underneath as an answer to "where would you make it," which is the question that comes fourth, not first.

A capability page opens with a sixty-word plain answer and a specification table above the fold: tolerance range, part envelope, materials, volume band, certifications. The process narrative and the proof sit below it, and the bottom of the page lists the plants that run that process with an RFQ link already scoped to that capability — so the request arrives at the estimator with half of itself filled in.

The quote flow is three steps and about ninety seconds. Drawing first, then part context, then who you are — deliberately last, because asking a stranger to identify themselves before you have done anything for them is how the old form lost people. Completeness went from 22% to 68%, and the median quote turnaround went from nine days to three, without the estimating team growing by a single person.

The website was not the bottleneck. It was the visible part of one. The single highest-leverage change on this project was asking for a drawing in step one, and that is a business-process decision that happened to be delivered as a form field. The work I am proudest of here looks like a form and is actually six days of email removed from the front of a sale.

The failed long form taught me something I now apply everywhere: when a user says "I'd just email you," they are not rejecting your layout, they are telling you the channel you built is worse than the one they already have. The fix is never a better version of the same thing.

On writing for this audience — the temptation in B2B is always to write bigger. Here the persuasive move was to write smaller and more precisely. Confidence reads as a number. Every adjective we cut and replaced with a specification made the page more convincing, which is a rule I have not found an exception to since.

And two days on the shop floor bought more credibility with the client than the proposal did. Being able to say "the Swiss cell, not the VMC" in a review meeting is not a research method anyone writes down, but it changes what people are willing to tell you.