How a fintech approved its rebrand without anyone being asked whether they liked it

The rebrand had been proposed twice before I was involved, and declined twice. I think both refusals were correct. Each proposal arrived as a wall of competitor screenshots, a moodboard and the sentence “we look dated”. Each time the finance director asked what the change would return, and nobody in the room had a number.

A rebrand argued from taste is assumed evidence at 0.30, however senior the person arguing it. A board that declines to spend six figures on a 0.30 is doing its job.

The company is a fintech: business accounts, transfers and cards for small firms. It is identified by industry only and the figures are rounded. In August the same board approved the same rebrand in one meeting, with the finance director as its sponsor. Nobody in that meeting was asked whether they liked the new identity until the decision was already made.

What the baseline said

The baseline ran in June. The composite came back at 3.1 with 0.58 confidence, which the score bands call adequate: nothing urgent, several things worth money. The composite is computed from behaviour alone, and in this case it was hiding two problems that pointed in opposite directions.

Fig. 01

Four surfaces, read twice

Filled bars are behaviour, outlined bars are perception. Two surfaces work and are believed not to. One fails and is believed to be fine. Averaged into a single figure, the three cancel into a number that asks for nothing.

On the Task surface, transfers completed 99.4% of the time in a median of nine seconds, and users rated the experience of sending money at 2.0. Belief had the same shape, 3.4 against 1.9. In the say-do matrix that is phantom friction: the product works and is believed not to. Commitment ran the other way. Only 59% of approved accounts were ever funded, and the people who never funded reported nothing wrong.

Phantom friction has a prescribed treatment, and the framework words it as “change what the product says, not what it does”. At company scale that means an identity, a vocabulary and a set of confirmation states, which is a rebrand. For the first time, the proposal came with a diagnosis, and the diagnosis could have come back different. Had Task been loud friction, I would have had to recommend against the project I had been brought in to support.

Distrust leaves marks in behaviour

Perception is reported evidence, weighted 0.70, and a board that has declined a rebrand twice will not reverse on survey answers. So the research phase went looking for what distrust does when nobody is asking about it.

It found three things. After 38% of completed transfers, the sender went back to the balance or the status screen two or more times within ten minutes. The money had already arrived. Of all support contacts, 22%, about 2,900 a month, were some version of “did my payment go through?” about payments that had. And at the funding step, 61% of the sessions that abandoned never touched a field, after a median of 48 seconds on the screen. Nobody abandons an untouched form because of the form.

The other 39% did hit something mechanical: a manually typed IBAN with a 27% error rate on the first attempt. That part is silent friction, and it gets the silent treatment, which is removal. It matters that the two were separated. A rebrand sold as the fix for the whole funding gap would have been oversold by about two fifths.

Pricing the brand they already had

The finance director's question had been fair both times, so it was answered first, using the arithmetic from pricing friction in your own currency.

Fig. 02

The current brand, priced per year

The outline is the raw cost and the fill is what survives the evidence weight. Contacts were counted, so they enter whole. The funding figure is real, but blaming it on belief rests on dwell time without input, which is an analytics proxy, so it enters at half.

Reassurance contacts were simple: 2,900 a month at a loaded €6.40 is €223,000 a year, counted in the support system, observed at 1.00. The funding gap took more care. About 1,150 accounts are approved each month and 41% never fund. Treating a conservative third of those as recoverable, at €310 of first-year value each, gives €584,000. But the claim that belief is what stops them rests on a proxy, so it carries 0.50 and enters at €292,000.

The number that went to the board was €515,000 a year. The raw €807,000 stayed on the slide with the discount shown next to it. I had expected the larger figure to be the persuasive one. What the finance director said afterwards was that a number arriving already attacked was the first one in three pitches they had not felt the need to attack themselves.

What the measurement said not to do

Both declined proposals had included a redesign of the whole app, around 140 screens. The baseline removed most of that. The transfer flow had no behavioural headroom: at 99.4% completion, the best a rebuild can do is hold level, and the likely result is regressions and every existing customer relearning something that worked. Access was in flow and needed protecting, so every colour in the new palette had to pass the contrast gate before anyone discussed whether it was attractive.

The scope that went to the board was the identity, the vocabulary, the landing pages, and 23 screens: the ones where the product talks about the state of someone's money. It cost about 40% less than the version that had been declined. A proposal that tells the buyer what to leave out is rare enough that it reads as evidence of its own.

Strategy, identity and interface, each with a gate

The strategy work came from the expectation gap. An intent question at sign-up showed 64% of new accounts arriving for some version of “business payments I can rely on”, and 81% of funded accounts were small firms. The brand voice had been written for consumers. It was playful, and it threw confetti when a payment succeeded. The identity was making a promise to people who were not the customers, and the customers were reading the playfulness as a lack of seriousness about their money.

Each of the three strands was tied to a finding, a treatment and a number that would count as failure, all written before design began.

Strand	Finding	Treatment	Gate
Strategy	Expectation gap: a consumer promise made to business users	Reposition the promise around dependable business payments, on the landing pages and in the first session	Entry intent matched at exit: 61% to 72% or better
Identity	Belief in phantom friction, 3.4 against 1.9	New identity and one vocabulary across product, help and marketing	Belief perception to 2.9 or better, behaviour not falling
Interface	Task in phantom friction, 4.1 against 2.0	Confirmation states, status language and receipts on 23 screens. No flow rebuilt.	Re-checks under 20%. Reassurance contacts under 12%. Completion not below 99.1%.

The third row is the one designers find hardest to accept. The interface work was mostly words and states: what the screen says in the four seconds after someone sends €8,000 to a supplier. The previous success state was an animation and the word “Done”. The new one states the amount, the recipient, the time the money left, when it will arrive, and where to find the receipt.

Three weeks on a fifth of the traffic

From late July, a fifth of new accounts and a fifth of transfer sessions got the new states, the vocabulary, the repositioned landing page and a funding step carrying the new identity, with account linking in place of the typed IBAN. Seven treatments in all, each with its gate on file.

Fig. 03

Six of the seven gates, as read

The seventh gate, Belief perception, is on a five-point scale and is reported in the text. Completion was never meant to rise. Its gate was a floor, there to catch the regression a working flow risks whenever someone touches it.

Five passed. Re-checks after a transfer fell from 38% to 17%. The reassurance share of support contacts in the pilot cohort halved. Funding rose from 59% to 68%. Belief perception went from 1.9 to 3.1, and intent matched at exit reached 74%. Transfer completion held at 99.3%, as predicted for a flow nobody rebuilt.

Two failed. A redesigned card-order screen left card activation at 71% against a gate of 78%. Renaming the statement export moved export questions to support from an index of 100 to 96, against a gate of 70. Both went to the board in the same table as the passes and at the same type size. One director said afterwards that the two failures were why they believed the five.

The meeting

The August meeting took forty minutes, and the order of the material mattered. First the diagnosis, then the price, then what had been cut, then the gates, then the pilot read against the gates. The identity itself came last and took eleven minutes.

Nobody was asked whether they liked it. They were asked whether the gates were the right gates, which is a question a board is qualified to answer. The finance director asked for one to be tightened for the full rollout, the funded rate from 66% to 70%, a number the pilot had not reached. It went into the plan at the stricter figure. Someone who has tightened your acceptance criterion has joined the project.

There is a version of this story in which the design was so good that it won the room. I do not think that is what happened. Two competent identities had already lost in that room. What changed was that this one could be wrong in ways the board could check, and had already been wrong twice in front of them.

What this does not show

Three weeks on a fifth of the traffic is a pilot. Rollout runs through the autumn, and under deploy decay every reading above loses confidence on the day it ships, at 0.8 per release touching the surface. The rescore follows the rollout.

The move in Belief perception is reported evidence and may include novelty. And I cannot separate the identity from the account-linking change at the funding step. They shipped together, so the nine points belong to both. That was a mistake in how I designed the pilot, and it is the reason the funding figure stays at 0.50 in the ledger until the two are run in separate arms.

Designers tend to lose rebrand arguments because they bring the weakest class of evidence to the most expensive decision a brand makes. The work here was ordinary identity, strategy and interface work. It was approved because it arrived with a diagnosis that could have gone the other way, a price that had been discounted by the person presenting it, and gates that two of seven treatments went on to fail.

Framework this draws on

Phantom friction and its treatment in the say-do matrix, the evidence classes and the priority formula in Layers 4 and 5, the expectation gap for the strategy strand, and phase 04 of the methodology, where acceptance criteria are numeric and written before the work begins.

Find your gap