Take the total number of people typing on a glass touchscreen instead of physical keys, multiply by the extra seconds each mistyped word costs, and Rory Sutherland arrives at a claim most people would find faintly ridiculous: that losing the physical keyboard was a real, measurable productivity disaster, hidden behind a device everyone agreed to call an obvious improvement.1 [CONTESTED] [SPECULATIVE] — this is one person's strongly-held opinion, argued vividly, not an established or measured fact, and it should be read that way throughout this page.
Sutherland's position is that the Canadians at RIM — the company behind the BlackBerry, with its physical keyboard — were right, and Steve Jobs, who bet everything on the glass touchscreen, was wrong, at least on this one specific design question.2 The argument isn't nostalgia for a discontinued product. It's a claim that something genuinely functional was traded away for something that looked and demoed better, and that the trade was never honestly evaluated on its own terms because the device that replaced the keyboard won on so many other dimensions that nobody stopped to audit this one loss specifically.
The claim's structure matches a pattern this batch documents elsewhere under different names: a genuinely inferior functional choice can still win in the market if it arrives packaged with enough novelty, visual appeal, and momentum on other fronts that the specific loss goes unaudited. Sutherland's phrase for this is blunt — "a case of display and novelty triumphing over utility."3 The predictive keyboard software that eventually shipped to compensate — larger touch targets for likely-next letters — is, on this account, an elaborate patch for a problem the physical keyboard never had, and the patch itself fails predictably on anything outside its training assumptions: foreign names, unusual words, anything the predictive model wasn't built to expect types badly, repeatedly, for exactly the population least well served by autocorrect tuned to common English patterns.4
Whether or not the specific productivity-loss estimate holds up, the claim is a useful test case for a more general and more defensible question: how does a market decide a design trade-off has been made correctly, when the winning option succeeded for reasons (aesthetics, brand momentum, ecosystem lock-in) that have nothing to do with the specific functional dimension in question? The physical-keyboard case is sharp precisely because the losing option (BlackBerry) had an obvious, easily articulated functional advantage on typing speed specifically, and lost anyway, decisively, for reasons unrelated to that advantage.
You're on a product team facing a design decision where the more visually striking, novel option has an acknowledged functional weakness relative to a more conventional alternative, and the team's instinct is to trust that the market will sort out the trade-off correctly if the product succeeds overall. You raise the keyboard case as a caution: market success on the whole does not validate every specific design choice inside a successful product. A product can win decisively while still carrying a real, unaudited functional regression that simply wasn't decisive to its overall success — and "the market chose this" is not the same claim as "the market correctly evaluated this specific trade-off."
Later, a colleague dismisses a user complaint about a specific feature regression by pointing to the product's overall growth numbers. You push back gently: growth numbers aggregate every design decision at once; they cannot tell you whether any single decision, evaluated in isolation, was actually the right one. The keyboard case shows a real functional loss can ride along inside an otherwise successful product for over a decade, entirely unaudited, precisely because nobody's growth metric was granular enough to isolate it.
The predictive-typing failure mode Sutherland cites — names and foreign words requiring repeated correction on an English-optimized keyboard — is a real, checkable phenomenon, and lends some concrete support to the broader claim, since it identifies a specific population (non-Anglo names) for whom the productivity cost is plausibly larger and more persistent than for the average user.
The unresolved tension, stated plainly: this claim is [SPECULATIVE] and largely unsupported by anything beyond personal conviction and a vivid rhetorical delivery in the source. No study, survey, or measured typing-speed comparison is cited. A skeptical reading treats this as an entertaining, possibly-correct-in-part opinion from someone with genuine design and behavioral-economics expertise, not a settled finding — and this page should never be cited elsewhere in the vault as though the productivity-loss claim were established.
This sits in genuine, unresolved tension with the same source's own account of the Walkman trick (see The Walkman Trick), where deliberately removing a familiar affordance (a display, in that case) is praised as smart design strategy. Sutherland does not reconcile why removing the keyboard reads to him as a regression while removing the display from smart glasses reads as clever legibility engineering — both are cases of a company choosing reduced physical/functional capability in service of a different design goal (aesthetics and unified touch-interface in one case, adoption legibility in the other). The most charitable reconciliation available, though the source doesn't offer it: the keyboard case involved removing a capability that had no adequate substitute (predictive typing measurably underperforms physical keys by his account), while the Walkman-trick cases removed capability that genuinely wasn't needed yet for the core use case. Whether that distinction actually holds, or whether Sutherland is simply more nostalgic about one device than the other, is not resolvable from the source material alone.
Behavioral-Mechanics — The Walkman Trick. Both pages describe a company removing a familiar hardware affordance from a product, but this page's case is framed by the same source as a loss and that page's cases are framed as strategy — and holding them together exposes an unstated evaluative double standard the source never resolves (see Author Tensions above). The insight the pairing produces: "was this reduction good or bad" may not be answerable from engineering or market-success evidence alone — it depends on whether an adequate substitute exists for the removed capability, a question that requires domain-specific evaluation (does predictive typing genuinely match physical-key speed? does voice-and-camera genuinely deliver what a display would have?) rather than a general rule about removal being good or bad.
Business — Misaligned Optimization Metrics: The Eurostar Case. Both pages document a company optimizing for a legible, easily-measured metric (journey-time reduction; unified glass-surface aesthetics and thinness) while a real cost lands on a dimension nobody was measuring (a productive train journey; typing accuracy and speed). The insight the pairing produces: the specific failure mode isn't unique to trains or phones — it's a general risk of any optimization process where the metric being tracked (minutes saved, millimeters of thickness) is easier to measure than the thing actually being traded away (a table to work on, a keyboard to type on), and the trade only becomes visible to a customer who experiences the loss directly rather than to the team looking at the tracked metric.
Sharpest implication: a product can be an unambiguous market success and still contain a real, unaudited functional regression that simply wasn't decisive to that success — success at the aggregate level tells you nothing about whether any specific design trade-off inside the winning product was actually correct.
Generative questions: