In UK banking, the app is almost never the problem

Twice a year the Competition and Markets Authority requires the seventeen largest current account providers in Great Britain to publish independent service quality scores. The August 2026 results came out on 17 August. Almost nobody outside compliance reads them.

They are worth reading, and not for the league table the coverage always reports. The useful part shows up only when you subtract one column from another.

In all seventeen banks, customers rate the mobile and online banking service higher than they rate the bank. Every single one, in both waves published this year, on a survey with fixed methodology and roughly a thousand respondents per provider.

The size of that difference runs from 1 point to 18.

That range is the finding. It tells you something specific about where a bank's problems actually live, and it contradicts the brief most product teams are working from.

The survey exists because of the CMA's 2016 retail banking market investigation. Ipsos asks current account customers how likely they are to recommend four things separately: the provider overall, its online and mobile banking, its overdraft service, and its branches. Published figures show the share answering "extremely likely" or "very likely".

The seventeen are Bank of Scotland, Barclays, Chase, first direct, Halifax, HSBC UK, Lloyds, Metro Bank, Monzo, Nationwide, NatWest, Royal Bank of Scotland, Santander, Starling Bank, The Co-operative Bank, TSB and Virgin Money. Between them they hold the overwhelming majority of personal current accounts in Great Britain.

Banks cannot opt out, cannot select the sample, and cannot adjust the result. For a UK audience this matters more than it might seem. You can put this number in front of a board without having to defend where it came from.

The four separate dimensions are what make the data useful. Most banks measure app satisfaction. Very few subtract it from anything.

The median gap across the seventeen is 11 points. Nationwide holds the narrowest at 1. Royal Bank of Scotland holds the widest at 18, scoring 48 overall against 66 for its own app.

There is an obvious objection to make here, and it is worth making. Part of the gap is designed into the questions: the overall score absorbs pricing, fees, queue times and hold music, while the app score covers the digital channel alone. Some positive gap would appear even if every bank were identically competent.

That artefact is the same size for everyone. The seventeen-point spread between Nationwide and RBS is not. What is worth explaining is why one bank's app outperforms the rest of its service by a single point while another's outperforms by eighteen.

The two scores are not measuring the same thing. The overall score is a customer experience score, covering everything the bank does to you, of which the app is one part. The app score is a user experience score, covering the interface alone. Subtract one from the other and what remains is everything in customer experience that is not the app.

For most banks the largest piece of that remainder sits offline: counters, ATMs, call queues, and whoever picks up. It is also the only piece the survey measures separately, which makes it the one part of customer experience that can be tested against the gap directly.

It tests well. The four providers Ipsos records as operating no UK branch network average a 5.8 point gap. The thirteen with branches average 11.4. Across the branch operators, the correlation between branch service score and gap size is −0.74.

Where none of that exists, customer experience does not disappear. It concentrates into a much smaller set of moments: whether support resolves things when they break, and whether you can actually reach your own cash without paying for it when the bank owns no machines. Fewer surfaces, and correspondingly less for the app to outperform.

There is a second reason, and it is structural rather than perceptual. Digital service holds its quality because it works as a system. Build it once and every customer gets the same thing. Human service is produced fresh each time, by a professional who is also, some days, having a bad Tuesday. Skill does not remove that. A bank with 500 branches runs 500 instances of its service standard, and the variance that produces has nothing to do with how good anyone is at their job. It is the cost of delivering a standard through people rather than running it as a system.

Nationwide is the clearest confirmation. It has the strongest branch scores in the country by a wide margin, 83 against 71 for the next provider, and the smallest gap in the table at 1 point. Its digital channel does not outperform the rest of its service because the rest of its service is good.

Royal Bank of Scotland sits at the other end of the same line. Its branch score is 51, the lowest published, and its gap is the widest. The app is not the outlier in that picture. Everything around it is.

Branch score and overall score correlate with each other at +0.87 across the thirteen branch operators, so on thirteen observations this is the most plausible mechanism rather than a proven cause.

One case breaks the easy reading. The Co-operative Bank scores 55 overall and shows a gap of only 5. Not because its service is coherent, but because its app scores 60, joint lowest in the table. A small gap means either everything is good or everything is weak. The gap on its own is not a quality score, and reading it as one will send you in the wrong direction.

Here is the uncomfortable version for anyone in a branch-operating bank with a redesign budget in front of them.

Your app is already the strongest-scoring part of your service. If your gap sits near the median, your customers rate your digital channel roughly 11 points above how they rate you. Another redesign will widen a gap that is already the widest thing about you. It will not move the number your executive committee cares about.

When a bank brings me a redesign, I ask two things before anything else. What is this redesign for, and how will you know what it returned.

The first question usually gets an answer. Business goals, roughly stated, sometimes precisely. The second rarely does, and the reason is almost never that the person in front of me has not thought about it. The answer does not exist yet. Nobody captured a baseline, and the initiative was scoped before anyone asked what it was supposed to move.

That gap in the conversation tells you more than the survey does, and it arrives before you have opened any data. A redesign that cannot state its own return is still a budget looking for a decision.

There is a third question, narrower than the other two. What exactly do you want to improve, and where do you think the bottleneck sits. That one usually gets a confident answer, and the answer is usually about the interface.

This is the situation we see most often. A bank arrives certain that the app is the problem, having benchmarked it against a competitor and found it wanting. The audit finds the app is fine. What is broken sits behind it: pricing that works for a month and then stops, a support function that cannot resolve at first contact, a back office that cannot execute what the interface promises.

The pattern underneath is almost always the same. A large share of what the bank wants to build cannot be built, and the obstacle is rarely design or engineering capacity. The business process behind the feature does not support it, and an interface can only expose what the organisation is already able to do.

That makes for an awkward report. The bank expects something shaped like "build this and it will work". What arrives is a set of unglamorous numbers showing where money leaks and where retention falls. We say this early now, because a diagnosis nobody asked for is more useful than a redesign nobody needs.

The habit that produces this is benchmarking. You open a competitor's app, list what it does that yours does not, and the list becomes the brief. It feels like diagnosis because it produces a document. It is not, because it never asks whether those features are missing for a reason no amount of design work will reach.

You do not need us to check which situation you are in. Five things, all in data you already hold, and between them they give you the baseline that second question needs.

Compare your customer base to your monthly active app users. Not registrations, actives. This single ratio separates two situations that look identical from the outside. If most of your customers have never registered, or registered once and stopped, they are telling you the app offers them nothing they need often enough to open it. That problem sits in the app. If they arrive in volume and leave within weeks, the app is doing its job and something around it is not: what was promised, what was delivered, what happened when they asked for help. Two banks can hold the same app store rating and have opposite problems. This ratio is what tells them apart.

Measure the gap, not the app. Ask the same customers both questions in one survey: would you recommend the bank, would you recommend the app. Subtract. If you operate branches and your gap is well above 11, the app is not where your problem is. If you have no branch network, there is less around the app for it to outperform, so a wide gap is harder to account for. The four UK providers in that position sit between 3 and 7 points, which is four observations rather than a benchmark.

Follow a promotional cohort to month three. Take customers acquired on cashback, referral rewards or a welcome bonus. Count how many are still active ninety days later. A cliff at the end of the promotional window means the marketing offer is working and the product is not. The interface is part of that product, so the job is to make the two work together rather than to keep buying the same customers twice.

Categorise your support contacts. Volume alone tells you little. Take your top ten contact reasons and map each to the screen the customer was on when they gave up. If most contacts are about things the app should have answered, your problem is genuinely in the app. If most are about pricing, terms, exceptions and decisions, it sits somewhere the interface cannot reach.

App store reviews work as a supporting signal here, read for content rather than score. The rating tells you little, since it comes from a self-selected group. What people write in the one-star reviews usually maps onto the same top ten.

Take three things customers ask for that you have not built. For each, ask why not. There are three honest answers: design and engineering have not got to it, the back office cannot support it, or it does not earn the bank enough to be worth building. Only the first is a design problem. If two of three land on process or economics, the redesign budget will not move anything, and you now have the evidence to say so.

The clearest example I have of this began as an audit rather than a redesign.

Oschadbank, Ukraine's state savings bank, came to us in 2019. We audited the existing app, prioritised the customer groups that mattered and interviewed them. What those interviews turned up is the pattern this article has been describing. People were opening the app, taking their cash out immediately or moving it to a card at another bank, because what they needed from it was not there. The customer base was branch-led, and the app was giving them no reason to change that.

Alty designed the concept, the journeys and the interface. The app launched at the end of 2021. The bank now reports that more than 70% of all customer banking operations happen in the app, double the share from before the full-scale invasion.

That shift was decided at the level of what the bank does where. It held because the decision was made once, on evidence, rather than revisited every year.

A brief built from a competitor teardown will get delivered. Somebody will take it, price it, and build exactly what was asked for, on time. That is what the arrangement is for, and it is why the second redesign so reliably follows the first.

If you are looking at a gap you cannot explain, the useful first step is a diagnosis rather than a redesign. We do that work as a scoped review, and sometimes it ends with us saying the build is not the answer. We treat that as a result.

Explore how we work →