RFQ Friction Index: 25 Manufacturer Sites Audited (2026)

A buyer decides you can make the part, opens your site on a phone between meetings, and starts the RFQ process. Two minutes later there is no quote request. Not because the shop was wrong, and not because the price was too high. The path itself broke.

Factory Investigator built the RFQ Friction Index to find out where that break actually happens. We inspected 30 manufacturer websites and fully scored 25 of them against 12 public, observable checks covering the whole quote path: how many clicks to reach it, what the form demands, whether a drawing can be attached, whether it works on a phone, and whether the buyer is ever told when a quote will arrive.

The result was not what the “your website is broken” pitch predicts. Almost every site handled the beginning of that path well. The friction sits at the end, on the last screen, in three checks that a redesign brief rarely covers.

What the RFQ Friction Index measures

The RFQ Friction Index is a 12-check score of the public quote path on a manufacturer’s website, from homepage to submitted request. Each check is a pass or a fail against a fixed condition, points are added on failure, and the lower the total the cleaner the path.

The weights are a judgement call, fixed before any site was scored and not derived from outcome data. Broadly, checks that can stop a quote outright carry 10, checks that cost a buyer effort carry 8, and checks that cost a buyer confidence carry 6. Two sit awkwardly in that scheme: a forced login stops a quote outright but carries 8, and a page that never loads is not a confidence problem but carries 6. A perfect path scores zero; a site failing everything would score 94. Where a check cannot apply at all, because the site has no form, it is scored not-applicable and the site’s maximum is reduced rather than its score inflated.

Why the checks are public-only

Every check is observable without submitting anything. We filled in no forms, solved no CAPTCHAs, created no accounts and bypassed nothing that was blocked. That constraint costs coverage, and it is the reason five of the 30 sampled sites still cannot be fully scored. It also means the whole method is repeatable by anyone, including by the manufacturers being measured.

In this sample, most sites are not broken. They are leaking.

Of the 25 fully scored sites, 15 landed in the Low friction band and 10 in Moderate. Not one scored High. The median was 20, the range ran from 6 to 36, and the arithmetic mean was 19.6 against a possible 94.

That split is closer than it looks. Five of the 25 sites scored exactly 20, sitting precisely on the Low and Moderate boundary. The bands were fixed before any site was scored, so this is not a moved goalpost. But a threshold one point lower would have produced 10 Low and 15 Moderate, and the majority would have flipped. Read the split as a distribution, not a verdict.

Not one of the 23 sites with a form passed mobile usability, response expectation and a nearby trust signal together.

Two caveats before either of those results carries weight. The empty High band is partly a property of the rubric: once a site passes the four structural checks, which 21 of the 25 did, the remaining eight are worth 56 points and a site would have to fail every one of the eight to reach the High threshold at all. And with each of the three checks above failing on roughly half the sample, treating all three as independent predicts about two sites passing and gives roughly a one-in-seven chance of seeing none. So read the blockquote as a description of how common these failures are, not as evidence that they cluster.

What the distribution does show is a consistent shape. These sites were not outdated or hostile. They were unfinished: solid at the parts designed once and rarely revisited, weak at the parts that only surface when a buyer arrives on a phone with a job to price.

PCB and EMS came out cleanest at a median of 12, which is consistent with a vertical whose buyers routinely upload a bill of materials and Gerber files and whose sites are built around that transaction. With eight scoreable rows in that group, treat the gap as a signal worth testing, not a settled ranking.

Where the points actually came from

Counting failures by frequency and counting them by cost give different pictures, and the cost view is the useful one. Across all 25 sites, 490 friction points were recorded. Three checks produced 274 of them.

• C7 mobile usability — 112 points, 22.9% of all friction recorded
• C8 response expectation — 90 points, 18.4%
• C11 trust proximity — 72 points, 14.7%
• C5 file upload and C9 human alternative — tied at 48 points each, 9.8% apiece

Three checks hold 56% of the friction in this sample and five hold 76%, none of which requires rebuilding a website. That is the practical takeaway for a shop owner with a small budget and no appetite for a redesign.

Mobile is where the quote path fails first

Mobile usability failed on 14 of the 23 sites that have a form at all, or 60.9%, the most common failure in the index. The check is not subjective: at a 390×844 viewport the form must show no material clipping or overlap, editable text must be at least 16px, and the primary action must be at least 40px tall.

The failures clustered into three shapes. Editable fields rendered at 13px or 14px, which forces an iOS zoom the moment a buyer taps into them. Submit buttons came in just under the threshold, clustering tightly at 38px across four sites, with one site rendering a 12px control. And several full-page mobile captures re-rendered or clipped repeatedly, so a complete, stable form never resolved.

That last category is worth dwelling on, because it is invisible from a desk. On three sites the desktop form rendered cleanly but the mobile capture never stabilised into a complete, usable form inside the observation window. We cannot separate the site from the observation there, and that is close to the point: if a scripted mobile load cannot reliably reach the form, a buyer on a phone may not either.

The 40px threshold is a usability choice, not an accessibility one, and it is worth being precise about that. The W3C’s WCAG 2.2 Target Size (Minimum) criterion sets the floor at 24 by 24 CSS pixels, which a 38px button clears comfortably; the enhanced criterion asks for 44 by 44, which it does not. We set the bar at 40px because a thumb on a moving train is not the population the accessibility minimum was calibrated for.

Most sites never tell the buyer when the quote arrives

An explicit quote-response duration failed on 15 of 25 sites, 60.0%, second only to mobile usability. The check asks for a number: 24 hours, two business days, same day. “Soon”, “promptly”, “we will be in touch shortly” and published production lead times all failed, because none of them tells a buyer when to stop waiting and start calling someone else.

Ten of 25 sites stated a real duration. It costs a copy change, and it is one of the two most commonly skipped.

Speed matters on the other side of that form too. The InsideSales.com and MIT Lead Response Management study, which analysed over fifteen thousand web leads and more than one hundred thousand call attempts, found the odds of qualifying a lead drop by a factor of 21 between a five-minute and a thirty-minute response. Treat that multiple as an order of magnitude, not a coefficient: it is a 2007 vendor study run on the vendor’s own CRM data across six unnamed companies, presented at a marketing summit and never peer-reviewed.

Stating a turnaround does two things at once. It gives the buyer a reason not to send the same drawing to three other shops this afternoon, and it commits your own team to a number.

The trust signal is missing exactly where it is needed

A qualifying third-party trust signal near the form failed on 12 of 25 sites, 48.0%. The check accepts any of three things within sight of the form: a certification or accreditation, a named customer or testimonial, or a third-party seal. Self-description does not count, and neither does a certification logo parked in the footer of a different page.

We did not audit what certifications these firms actually hold, only what was visible beside the form, so read the next point as advice rather than a finding. If you hold a certification, an accreditation or a named customer you are allowed to quote, it is almost certainly sitting on a quality page rather than beside your form. More than one million ISO 9001 certificates have been issued across 189 countries, per ISO’s own figures for the standard, and AS9100, ITAR registration and IATF 16949 turn up routinely in these verticals. For most shops this is a placement problem, not a credentials problem.

The adjacent failure follows the same pattern. A phone number or email beside the form failed on 8 of 25 sites, 32.0%. Header and footer contact details did not count, deliberately: a buyer who is stuck on a form field wants an escape hatch in the form block, not a scroll away.

What passed, and why it matters more than it sounds

Two checks recorded zero failures across all 25 sites. Every site put its quote path within two clicks of the homepage, and no site forced a login before an enquiry. Capability evidence came close: only one site failed to make technical capability and some fit or assurance findable before the form. Three of the 25 had no dedicated quote form at all, only a generic contact box or no form whatsoever.

That capability result deserves more credit than it usually gets. Making a buyer prove your capability before they can ask for a quote is the classic complaint about industrial websites, and 24 of 25 sites in this sample handled it. If your own site does this well, the work worth doing next is on the specific technical detail a buyer checks on a capability page, not on whether one exists.

The middle of the rubric was quiet. Seven or fewer required fields failed on 5 of the 23 sites with a form, 21.7%. A required vanity field appeared on just 2 of those 23. The five-second load failed on 4 of 25. File upload is the exception: it failed on 6 of 25, because neither of the two sites without a form can accept a drawing either.

Form bloat, in other words, was a minor problem in this sample. For scale, Baymard Institute puts the average e-commerce checkout at 11.3 form fields across 5.1 steps, and argues most sites need only eight. That is a different measurement in a different context, so it is a reference point rather than a comparison. If you want to go deeper on which fields earn their place, that is the subject of our guide to which RFQ form fields to keep and which to cut. What the index says is that field count is not where most of these sites were losing quotes.

What we could not see

Five of the 30 sampled sites still cannot be fully scored, and they are excluded from every number in this report. That is a real limitation, not a footnote.

Two used embedded third-party quoting widgets whose required-field behaviour could not be established without submitting. One linked to a form that was blocked in the observation session. One had an embedded form context that would not load. One had a quoting calculator behind a URL we chose not to bypass.

A sixth and seventh site were in this group in the first version of this study and are now scored. Both have no online form anywhere in their public quote path, which is not the same thing as a form we failed to observe. Charging them for a missing form’s field count and mobile layout would bill one fact five times over, so those cells are scored not-applicable and their maximum is reduced instead. Scored the other way, one of them would have become the only High-friction site in the study, on no additional evidence. That is the strongest objection to the choice, so it is stated here rather than buried: the rule was picked because charging one fact five times is indefensible, not because of what it did to the distribution.

Those five are not a random slice. Sites with third-party quoting widgets and gated calculators are plausibly different from the rest of the sample in ways that could move the distribution in either direction. Anyone citing this work should say 25, and should say complete-case.

The index also measures nothing that happens after the send button. It cannot see whether a quote was returned, how fast, or whether it was any good. A site can score 6 on this index and still lose the job in the estimating queue.

The five fixes worth doing first

Read this as a priority order derived from where the points were, not as a redesign brief. Each of these is a change to an existing page. The first three are the ones that matter; the last two tie for fourth.

1. Fix the form at 390 pixels. Set editable fields to at least 16px so mobile browsers stop zooming, give the submit button at least 40px of height and preferably 44, and load the page cold on a phone to confirm a complete form actually renders. Worth 8 points and the largest single share of friction in the sample.
2. State a turnaround next to the submit button. A specific number, in hours or business days, that your team can hold. Worth 6 points and it is a copy change.
3. Move one trust signal into the form block. A certification, a named customer, or a third-party seal, positioned where the buyer is deciding. Worth 6 points.
4. Put a phone number and an email in the form block. Not the header, not the footer. Worth 6 points and it gives a stuck buyer somewhere to go.
5. Accept a drawing at the form. A file upload on or beside the form, not an invitation to email an attachment separately. Worth 8 points, and it is what turns an enquiry into something an estimator can price.

The first three fixes are worth 20 points between them, and all five are worth 34. The median site in this sample scored 20 in total. That arithmetic is the reason this index exists: the expensive-looking problem and the expensive fix are rarely the same thing. Where the RFQ path is only one part of a wider funnel question, the sequence we would work through is in our guide to turning manufacturer traffic into RFQs.

How to run this on your own site

Record unknowns honestly, and keep "cannot observe" separate from "does not exist". Fifteen of the 360 criterion cells here could not be observed, and the five sites carrying them are excluded outright. A further six cells describe form properties that are simply absent; those are not unknown, and scoring them as failures would have charged one fact five times.

What comes next for this index

This is Field Report 001, published at 25 complete cases. It began at 23. The two sites added since were never unobservable; they were miscoded, because a form that does not exist was recorded the same way as a form we could not reach. Separating those two states is the whole of the change, and no site was re-observed to make it.

Five open rows remain. Closing them needs another pass on the same instrument this study used, a rendered browser session at 390×844 rather than a static page fetch, and in four of the five cases it needs a currently blocked, embedded or gated form to become observable without submitting anything. Re-observing them with a different instrument and pooling the result with the other 25 would mix measurement methods inside one dataset, which is worse than leaving them open. After that the sample scales past 30, with second-reviewer adjudication on every failure.

One more disclosure belongs here. The 30-site sample was locked before scoring, but the decision to report only the complete cases was taken afterwards, and so was the no-form branch that moved two sites into the denominator. Both are post-hoc and you should weigh them as such.

Until then, treat every number here as what it is: 25 manufacturer websites, one date, one rubric, one reviewer, and a rubric published in full so you can run it yourself and argue with the result.

Frequently Asked Questions