Consumer · PlanIt · London transit · one case, one stealable principle

People broke it on purpose — just to watch the train look sad.

PlanIt could already predict which carriages would be empty. Real footfall data from the London Underground and National Rail, working as advertised. And still the hard part was untouched, because nobody chooses a slower train to be efficient — they choose it to breathe. This is the case where the thing everyone treats as a defect, the error state, turned out to be the most loved screen in the product.

6 monthsJul–Dec 2021 · designer and strategist · remote, distributed team
0 complaintsusability complaints to customer service in the launch window — no tutorial shipped
1,000 / 250monthly visitors and active users — the traction goal, honestly small

Consumer · PlanIt · London · 2021 · Designer & strategist

The principle this case proves

When the reason people won’t act is emotional, the interface is where you pay it — not the model.

The stakes: a correct prediction that changed nobody’s behaviour.

A London commute is not measured only in minutes. It is measured in elevated heart rates, tense shoulders, and the dread of being pressed into a full carriage. By 2021 the app landscape answered one question with real precision — what is the fastest route? — and ignored the one people actually asked at the platform edge: will I be able to breathe when I get on?

PlanIt arrived with the capability to answer it. Near-real-time footfall from the Underground and National Rail could say which windows would be quiet. On paper the problem was solved. In practice nothing moved, because the product was asking an anxious person to accept a slower journey on the word of a number. That is a trade only a calm person makes. The people who most needed it — elderly travellers, expecting mothers told to avoid stress, anyone with a stroller, luggage or a mobility need, anyone still wary after a pandemic — were the least likely to be persuaded by a grid of data.

The test: stop designing for a user, start designing for the frightened.

The first move was to narrow the audience deliberately, against instinct. Not commuters in general — the specific people for whom a crowded carriage is not an annoyance but a risk. That reframing decided everything downstream: a product for them could not look like the category. Transit apps speak in tech blue, dense grids and sharp edges, the visual language of a corporate dashboard. To someone already tense, that interface does not read as capable. It reads as one more thing making demands.

So the identity went the other way on purpose: warm primaries, soft accents, generous space, and Prompt for its rounded, legible forms — readable for an eighty-year-old without tipping into hospital signage. The brief was a strange one to write down and the right one: the app should behave less like an instrument and more like a sedative.

The PlanIt planner screen: a warm, uncluttered mobile web form — Using: London Underground, From, To, Day and Time — under the line ‘Let us help you find the quietest time to travel around London.’Ask
The same planner with a missing station, showing the validation message ‘No worries, it happens! Please add in the station.’Get it wrong
The whole product in one screen — and, beside it, the tone under pressure. Most forms scold: required field. This one says “No worries, it happens!” Tonality is not decoration when your user is already tense; it is the product.

The mechanism: one number, a British voice, and a train with a face.

A single score, not a data stream. Crowd data collapsed into Quietness, one to ten, so a route could be judged at a glance. Colour reinforced it — green comfortable, amber moderate, red busy — but softened away from neon, because a warning that induces panic defeats a product built to lower it. Large touch targets and tuned contrast were treated as first-class constraints, not an accessibility pass at the end: the same interface had to work for an eighteen-year-old student and an eighty-year-old grandparent.

A voice, instead of a readout. “Footfall density: high” tells an anxious person to brace. The copy went colloquial and distinctly British, trading metrics for pictures they already hold: “Quietness: 3/10 – very busy. Imagine Platform 9¾ on the first day at Hogwarts.” And where the old app would issue an instruction, this one made an offer: “A quieter time is coming up — bring your book, a seat should be waiting for you.”

PlanIt results: a bar chart of quietness across the morning, teal for quiet and pink for busy, above a list of suggested windows — Quiet 08:00–08:15, Quiet 11:15–11:30, Busy 08:45–09:00 — and the line ‘Quietness: 3/10 — Very busy. Imagine Platform 9¾ on the first day at Hogwarts.’Suggested trains
  • The chart answers first. Teal and pink carry the verdict across the whole morning before a single word is read — you can pick a train without parsing anything.
  • Windows, not orders. Every row is an offer with a time attached — Quiet, 08:00–08:15 — so the traveller chooses, and the choice stays theirs.
  • The number gets a picture. “Quietness: 3/10 – Very busy. Imagine Platform 9¾ on the first day at Hogwarts.” Nobody has to be taught what that means.
The answer screen. Colour carries the judgement before a word is read, the list offers windows rather than issuing an instruction, and the explanation is a picture you already hold — not a density figure.

A mascot doing real work. The PlanIt train was not decoration. It was the emotional anchor: greeting the traveller, approving a clear route, and visibly feeling bad when something went wrong. A face is a shortcut to a judgement humans make in milliseconds — is this thing on my side? — and the product needed that answer before it could sell a slower journey.

The PlanIt train mascot, cheerfulgreeting you
The PlanIt train mascot, annoyedthis one’s busy
The PlanIt train mascot, wincingsorry — that broke
The PlanIt train mascot, pleased with a quiet routego now, it’s quiet
The same character across states. The range is the point: a system that can only look pleased is not trusted when things go wrong.

What nobody predicted: they broke it on purpose.

During testing and again in the wild, people deliberately triggered broken flows — entering journeys they knew would fail — to watch the little train be sad about it. The error state, the screen every product team treats as damage control, had become the thing users went looking for.

The PlanIt error screen: the train mascot with a downturned mouth above the line ‘Oh no, there seems to be a small problem! We are working on it, please try again later.’ and a Refresh button.The screen they went looking for
  • It owns it in the first person. “Oh no, there seems to be a small problem! We are working on it” — not an error occurred. Somebody is on the other side of this.
  • The face carries the apology. A downturned mouth does in a glance what a paragraph of sorry-for-the-inconvenience cannot, and it costs the reader nothing.
  • Nothing you had is taken away. The last good answer stays on screen underneath, so the failure is an interruption rather than a loss — and one button puts it right.
The most-loved screen in the product. It apologises in the first person, shows you a face that feels bad about it, and keeps the last good answer on screen underneath — so the failure costs you nothing you already had.
Why that is more than a charming anecdote

A failure state people seek out is a failure state that costs no trust. The moment the product could not deliver — ordinarily where confidence leaks away — became a small act of empathy that made the system more believable, not less. It is the same conviction that runs through the AI work on this site, arrived at from the opposite direction: what a system does when it cannot help is what decides whether anyone believes it when it can.

Falsifiable evidence — small numbers, honestly reported.

Usability: zero usability-related complaints reached customer service in the early launch window, and the product shipped with no tutorial — people from teens to eighty-somethings navigated it unaided. Engagement: the returning behaviour was driven by the voice and the mascot as much as the prediction; the emotional layer behaved as a retention feature rather than decoration. Business: the comfort niche hit its traction goal — 1,000 unique visitors and 250 active users a month. Small, and stated as small: PlanIt never tried to out-Google Google. It owned a territory the speed-optimised apps had written off.

The PlanIt desktop experience, showing the journey planner and quietness guidance in the same warm identity as the mobile web app.
The same calm on the desktop surface — one identity, two form factors, six months.

Where it breaks — and how it ended.

Warmth cannot rescue a weak signal. Every choice here rests on the footfall prediction being genuinely useful; a mascot on top of bad data is a lie with a face on it. The approach also assumes an emotional cost worth paying to remove — for a purely transactional task, this much personality is friction, not comfort. And charm is not a moat: the affection users had for the train would not have survived a competitor with better data and a warmer product.

The honest ending: PlanIt shipped fast, found a small loyal base, and was deprecated within months. planitnow.co.uk no longer resolves. I would rather show a product that reached real people and was then switched off than pretend everything I have touched is still running — the six-month build and the shutdown are the same skill, which is knowing what a bet is worth before you keep paying for it.

Send me the role ↗ Next case study →