Why Your Production Units Don't Feel Like Your Prototype — And How to Spec Button Feel to Your Factory

Your prototype buttons felt right and your production units don't. Here's what changes between T1 and mass production — and how to write a spec that holds.

IDEA TO DESIGN

Engineering Team

8/23/202610 min read

collage features refined tactile button details
collage features refined tactile button details

The complaint we hear most often after first production

A founder sends us a message that reads something like this: the hand-built prototype felt great, the first tooling samples felt stiff, and now units from the same production batch don't even feel the same as each other.

Nothing in the design changed. The CAD is the same, the switch part number is the same, the enclosure looks identical. And yet the product feels cheaper than it did three months ago.

This is not bad luck and it is not a mystery. Button feel is not a property of your design file — it is an output of your tolerance chain, your tooling, and your assembly process. It is the single most common place where a product that reviewed well in prototype loses its perceived quality in production, and it is almost always caught too late, because "feel" rarely appears anywhere in the spec that gets sent to the factory.

This article covers what actually shifts between a hand-built prototype and unit #500, how to diagnose which of those things is happening to you, and what to write into your documentation so your contract manufacturer can hit a feel target instead of guessing at one.

What "good feel" actually means, in engineering terms

Designers describe button feel with words: crisp, mushy, sticky, dead. Those words don't transfer to a factory. What transfers is a force-travel curve — a plot of how much force the button resists as it moves down through its stroke.

Four numbers describe that curve:

The snap ratio is the number that maps most directly to the subjective word. Industry practice across silicone keypad suppliers converges on 40–60%: below roughly 40%, the drop in force is too small for the fingertip to register as a distinct event, so the key feels dead even though it works electrically. Push the ratio much higher and the web fatigues faster.

Two things follow from this that matter more than the numbers themselves:

  1. Feel is measurable. A force-travel tester with a probe pressing at controlled speed produces a curve you can compare against a target, batch to batch. "Feels wrong" becomes "F1 came in at 240 gf against a 160 gf target."

  2. Everything in the mechanical stack moves that curve. Which is why it drifts.

Five things that change between your prototype and your production units

1. The tolerance stack collapses your travel budget

Your prototype was probably CNC-machined or 3D-printed, assembled once, by hand, by someone who cared. The gap between the button plunger and the top of the dome was whatever that one unit happened to have — and if it felt good, that was luck plus a caring hand.

In production, that gap is the sum of five or six independent tolerances. Each is within spec. Stacked in the wrong direction, they aren't.

Here is a representative stack for a side button pressing a metal dome on a PCB — the geometry in most handheld products:

Now compare those numbers to the dome's own actuation travel — around 0.25 mm. Your accumulated variation is larger than the entire distance the switch travels. Note also that the single largest contributor is not the moulded parts everyone worries about; it is the PCB, where a standard ±10% thickness tolerance on 1.6 mm FR4 alone eats two thirds of the dome's stroke.

Fall the wrong way and you get idle travel — the button moves and nothing happens, then the click arrives late. Users read this as sloppy. Fall the other way and you get preload or over-press — the dome sits partly compressed at rest, so the click is shallow, hard, or absent, and in the worst case the key is permanently near-triggered.

This is a well-known enough problem that Apple patented a fix: shims of individually selected thickness under each button assembly, specifically to compensate for varying tolerance stack-up so that every button in the product line feels the same. That is the level of effort a company spends when it decides feel is a shipped feature.

The practical version for a startup: run the stack-up before tooling, specify tighter PCB thickness if the budget allows, and nominate one dimension in the chain as your adjustment variable — usually plunger length or a spacer, because both are cheap to change after T1.

2. Coating thickness quietly adds force

This one surprises nearly everyone. On silicone keypads, the protective or colour coating sprayed on the key surface adds stiffness. More coats — which is what happens when a colour needs opacity, or when a laser-etched legend needs a base layer — means higher actuation force. The keypad supplier hasn't changed anything about your web geometry, and your buttons still got harder.

In practice, going from a bare or single-coat prototype part to a fully finished production part with two or three coats plus PU topcoat commonly adds somewhere in the region of 10–25% to F1. If your parts made that transition between prototype and tooling samples, that alone can account for the complaint. Specify the coating stack as part of the feel spec, not as a separate cosmetic line item.

3. Material hardness and process variation between batches

Silicone durometer, web thickness, and cure conditions all move within their own tolerances. Durometer is typically supplied at ±5 Shore A unless you specify tighter — the difference between a 50 and a 55 Shore A web is clearly perceptible at the fingertip. Compression-moulded parts from different shots, different cavities of the same tool, or different material lots will not produce identical curves.

A multi-cavity tool with an imbalance across cavities is the classic source of "some units feel different" — the units that feel different came out of cavity 3. Ask for cavity identification marks on moulded parts. It costs almost nothing at tooling stage and turns an unsolvable complaint into a traceable one.

4. Mold rework changes more than the dimension you asked to change

When T1 samples feel wrong and you request a change — reduce the button height, thin the web, loosen the aperture — the moldmaker adjusts steel. Steel comes off easily and goes back on expensively. Each adjustment also shifts things nobody wrote on the change request: draft, wall thickness at the transition, how the part ejects, how much the part warps as it cools.

The result is that iterations three and four often solve the reported problem and introduce a new one. The fix is not to iterate faster; it is to decide the geometry with a stack-up analysis and a soft-tool or printed test article before steel is cut.

5. Plastic-integrated buttons rely on the plastic itself as the spring

A button moulded as part of the housing — a living hinge, a cantilever tab, a thin-wall flexure — uses the elasticity of the plastic as its return spring. This is cheap, seals well, and is extremely sensitive to wall thickness variation. It also stacks two forces in series: the force to deform the plastic plus the force to trigger the switch. That sets a floor on how light the press can be, and it drifts with resin lot, gate location, and cooling.

If your target is a light, precise press, an integrated plastic flexure is the wrong architecture. Say so at DFM, not at T2.

Second-order effects your users will feel and your test plan won't catch

Sweat, dust, and glove operation. A key that tests perfectly on a clean bench in an air-conditioned room behaves differently under a film of sweat and dust. High-gloss surfaces feel tacky; the finger drags instead of landing. A matte, higher-abrasion-resistance surface treatment usually reads as more precise in field use even though it measures identically.

For gloved operation the mechanics change completely. Flat, near-zero-travel membrane surfaces give a gloved fingertip no edge to locate and no motion to confirm, so operators drift off the key and either double-tap or start looking at their hands instead of their work. Raised keys with real travel and generous spacing are what restore blind operation. There is also a sizing consequence that gets overlooked: research on protective-glove dimensional allowances found that a 95th-percentile index fingertip already exceeds the outline of typical control-panel buttons, and with a glove on it overlaps adjacent buttons — a direct route to mis-presses that no amount of force tuning will fix.

Wear over the product's life. Silicone webs soften with cycling; actuation force drops and the snap weakens over hundreds of thousands of actuations. Coated key edges abrade at the contact area first, which is exactly where the fingertip judges texture. Coating optimisation slows this down; it doesn't eliminate it. If your product is used many times per day for years, specify cycle-life testing at a number that reflects reality, and accept that end-of-life feel will not equal day-one feel.

Blind reachability. The reason a key is findable without looking is height difference, spacing, edge radius, and shape coding — not the legend. A primary function key raised 0.2–0.3 mm above its neighbours, or given a different shape, is locatable in the dark; the fingertip resolves height differences far finer than that. ISO 9241-410 treats travel, resistance, and feedback for physical input devices as design criteria rather than preferences, which is a useful thing to point at when a stakeholder wants to flatten everything for aesthetics.

What to actually write in your spec

Most feel problems reach us as a design that specifies a switch part number and nothing else. The switch is maybe a third of the feel. Here is the section we wish arrived with every project.

Feel targets

  • Actuation force F1, with tolerance — e.g. 160 gf ± 25 gf

  • Snap ratio target and acceptable range — e.g. 50%, range 40–60%

  • Total travel and contact travel, with tolerance — e.g. 1.6 mm ± 0.2

  • Return force minimum (a key must reliably come back up under its own web)

  • Maximum allowed variation between keys on the same unit — e.g. ≤ 15% of F1. This is the number that governs whether the product feels coherent, and almost nobody specifies it

Verification

  • Force-travel curve measured on a test rig at a stated probe speed (commonly 10 mm/min), not by hand

  • Full curve on every key of the first article; sampled per batch thereafter — a workable default is 3 units per production lot, every feel-critical key

  • Cavity ID recorded with each measurement on moulded parts

  • Cycle-life test: number of actuations and the post-test window that still counts as a pass — e.g. 1,000,000 cycles, F1 must remain within −20% of initial, snap ratio ≥ 35%

Materials and finish, treated as feel-critical

  • Silicone durometer with tolerance — e.g. 60 Shore A ± 3, tighter than the default ±5

  • Coating type, colour, and number of coats — flagged as force-affecting, so any change is a change request

  • Surface finish / texture spec on the keytop (mould texture standard, e.g. a VDI or MT number, or a defined gloss level)

  • Abrasion test requirement if the product sees outdoor or high-frequency use

Mechanical interface

  • Nominal gap or preload between plunger and dome, with the tolerance stack-up analysis attached. A common starting point for metal domes is a light preload around 0.10–0.15 mm rather than a nominal zero gap, so the worst-case stack still lands on the dome instead of leaving idle travel

  • Which dimension in the chain is designated as the adjustment variable

  • Keytop-to-housing flushness and gap tolerance — a wobbling or proud key reads as poor quality regardless of force. 0.15 mm all-round gap is a reasonable target on a moulded handheld

  • Anti-wobble features on tall or wide keys

Acceptance

  • One golden sample per feel-critical key, retained by both sides, signed off after T1 and re-verified at each production run

When to lock button geometry, and what it costs if you don't

The cost curve is steep, and it is worth being blunt about it with your team. Figures below are typical for a small-to-mid handheld enclosure tooled in Shenzhen:

The asymmetry between removing steel and adding it back has one clear implication: tool the button aperture and the plunger on the tight side of your range and open them up after T1. Starting loose and trying to recover feel later means welding, and welding a cosmetic surface is a decision you do not want to be making under launch pressure.

How we handle this on an NPI project

Button feel gets its own line in our DFM review rather than being folded into "enclosure." Concretely, before any steel is cut we check:

  • The full stack-up, in writing, from housing datum to dome apex — including PCB thickness tolerance, which is the item most often left out because it belongs to a different supplier

  • Which dimension absorbs the error. One dimension is nominated as the adjustment variable and toleranced deliberately loose on the safe side; everything else gets tightened

  • Coating and durometer as force inputs, not finish inputs. If the prototype was uncoated, we say out loud that the production part will be harder and by roughly how much

  • Cavity marking on every moulded part with a feel-critical feature

  • Architecture sanity check. If the target is a light, crisp press and the design uses an integrated plastic flexure, that conversation happens at DFM, when changing it is free

Then the sequence runs: printed or CNC test article to confirm the geometry → T1 samples measured on a force-travel rig, every key, cavity recorded → one adjustment round → T2 measured and compared → golden sample signed by both sides → sampled verification on every production lot against that same curve.

A typical correction cycle looks like this. T1 samples come in at F1 ≈ 235 gf against a 160 gf target, snap ratio around 28%, travel short of nominal — a button that reads as hard and dead. The curve shows the cause: too much force before the collapse and too little drop after it, which points at web geometry plus coating rather than at the dome. Thinning the web by roughly 0.05 mm and dropping from three colour coats to two brings T2 in near 170 gf with a snap ratio in the high forties. That is one iteration, about a week, and a few hundred dollars. The same problem discovered after 2,000 units have shipped is a recall conversation.

The mistake worth avoiding: treating feel as something to evaluate by hand at T1. By the time a sample is in your hand, the tool exists. Everything that makes feel reproducible — the stack-up, the adjustment variable, the coating decision, the switch architecture — is decided weeks earlier, on paper, for free.

FAQ

My prototype buttons felt perfect. Why should I test feel again after tooling? Because almost nothing that determines feel survives the transition unchanged. The prototype was one hand-assembled unit sitting at a random point in the tolerance range. Production is hundreds of units distributed across that range, made from different material, with a coating the prototype didn't have.

Can I just ask the factory for "better feel"? You can, and you will get a different result each time. Feel is only reproducible if it is expressed as a force-travel target with tolerances and a verification method. Otherwise you are asking one engineer's fingertip to match another's, across a language and a time zone.

Which switch type gives the most consistent feel in volume? Metal domes have the tightest and most repeatable force profiles of the common options, which is why they are the default when consistency matters. Tact switches are similarly consistent but need more Z-height. A silicone web providing its own tactile response gives the most design freedom and the most process variation. The right answer depends on your height budget, sealing requirement, and how tightly you need units to match each other.

How many mould iterations should I budget for a button-critical part? Plan for T1 plus two adjustment rounds. Most button-critical parts close out at T2 or T3 when the stack-up was done up front; without it, four or more is common and the later rounds start trading one problem for another. In calendar terms, budget three to four weeks after T1 before you consider the geometry frozen.

Does sealing (IP rating) affect feel? Yes. Sealing membranes and gaskets add stiffness in series with the key, raising actuation force and sometimes damping the snap. Specify the sealing requirement against real exposure rather than over-speccing, because feel is what you trade away for protection the product never needed.

What does force-travel testing cost? Very little relative to what it prevents. The measurement itself is a standard bench test at any keypad supplier or a competent CM, and adding it to first-article and lot-sampling routines is a documentation change more than a cost line.