77%
of mobile banking users surveyed struggled with their apps, mostly with navigation and jargon.
User-Centric Design of UI for Mobile Banking Apps, 2026
01 · Context and research
Banking apps have grown too busy
Neobanks now bundle payments, budgeting, subscriptions, savings and investing into one app. Each addition makes sense alone. Together they create three points of friction.
26%
of UK adults signed up to a subscription by accident within a year.
Citizens Advice
39%
of payments got the wrong category from a typical engine, in a test of 5 million.
MX, vendor-run benchmark of categorisation engines
What five banking apps do today
I reviewed Revolut, Wise, Monzo, N26 and Bleap against the three problems. The full benchmark is on the research board in FigJam.
Feature overload
- Wise and N26 lead with a few core tasks. Revolut and Bleap put everything upfront.
- Personalisation changes what people see, rarely what they can do first.
Subscription creep
- Recurring payments are detected, then shown passively as a list.
- Support arrives after the charge. No app helps people decide before it.
Categorisation
- Common merchants are filed well. Food delivery and marketplaces often aren't.
- Monzo makes correcting easiest, but few apps show that a correction will be remembered.
02 · Problem areas
Three problems, and what I designed for each
Too many features bury simple tasks
The few things people open the app for get buried under everything else the bank wants to sell.
- Hick's LawMore choices, slower decisions.
- Tesler's LawComplexity stays. I decide where it lives.
Solution
- FocusOne insight on Home, the rest kept in Insights
- RankQuick actions ordered by the last 30 days of use
- ExplainAn "Adapted to you" label shows why
- ControlPin, reset or switch it off
Show my reasoning
Hick's Law
Decisions take longer as choices grow in number and complexity.
Current pattern
More featuresMore decisionsHigher cognitive effortSlower task completion
Design implication
Surface the relevant action at the right moment and let the rest wait.
Tesler's Law
Some complexity can't be removed, only moved.
Current pattern
More financial productsMore informationMore complexityUser overwhelm
Design implication
Finance stays complex. My job is to decide where that complexity lives: in front when needed, a step away otherwise.
Hypothesis
Ranking actions by behaviour could make everyday tasks faster without hiding a thing.
Subscriptions get noticed too late
Apps announce a renewal, then leave people to work out alone whether it's still worth paying for.
- Doherty ThresholdHelp must arrive while there's time to act.
- Peak-End RuleA surprise charge is what people remember.
Solution
- DetectRecurring payments found in card history
- FilterOnly unused, duplicate or pricier services flagged
- Nudge3 days before the charge, with the reason
- DecideKeep, remind or cancel, all equal
- CloseEnds on the saving, with a 24-hour undo
Show my reasoning
Doherty Threshold
A system has to respond in time to keep people in flow.
Current pattern
Subscription renewsUser notices too lateUnwanted spendingReactive management
Design implication
Help has to arrive before the charge, while there is still time to act.
Peak-End Rule
People judge an experience by its most intense moment and its end.
Current pattern
Unexpected renewalNegative financial surpriseReduced trustPoor product perception
Design implication
Prevent the surprise, and design the end of the flow as carefully as the start.
Where the burden sits today
Apps say "Your subscription renews in 3 days" and stop. Do I still use it? Is there a cheaper option? People do every step alone:
Recognise the issueEvaluate relevanceDecideAct
Hypothesis
A well-timed, reasoned prompt before renewal could prevent most unwanted charges.
Wrong categories make spending insights unreliable
Apps rarely say why a payment landed in a category, so one wrong label undermines the whole chart.
- Jakob's LawCategories should work like in other apps.
- Postel's LawAccept messy names, return clear categories.
Solution
- AskWhen unsure, Strom asks instead of guessing
- ExplainEvery payment shows its reason and confidence
- LearnOne correction becomes a rule, past payments included
- ProveInsights shows accuracy and how corrections helped
Show my reasoning
Jakob's Law
People expect products to work like the ones they already know.
Current pattern
Transaction occursIncorrect categoryDoubtReduced trust
Design implication
Categories should improve with use, so each merchant only needs one fix.
Postel's Law
Be flexible in what you accept, clear in what you return.
Current pattern
Ambiguous inputUnexplained categoryUser uncertainty
Design implication
Accept messy merchants like Uber Eats or Amazon, and return a category with its reason and a way to fix it.
Where the burden sits today
Apps rarely say why a category was chosen or whether a correction will stick. People keep the data clean on their own:
Detect errorCorrect categoryRepeatLose confidence
Hypothesis
Explaining each category and learning from corrections could make the spending chart trustworthy again.
03 · Persona and jobs to be done
Who I designed for
A proto-persona built from the research above, to be checked in interviews.
Maya, 31
Account manager in Amsterdam · one neobank as her main account
"I only look at my spending when something already feels wrong, and then I'm not sure the chart is right."
- Opens the app
- 3 to 4 times a day
- Mostly uses
- Balance, send, bills
- Subscriptions
- 6 · €87/month
- Spending tab
- Doesn't trust it
Goals
- Get routine tasks done fast
- Drop unused subscriptions
- See spending at a glance
Frustrations
- A home screen full of offers she never uses
- Notices renewals too late
- Fixed the same Deliveroo category three times
What builds trust
- Knowing why things appear
- Undo for anything automated
- Corrections that stick
Main jobs to be done
J1
When I open the app for a specific task, I want the action I need right there, so I don't scan past features I never use.
J2
When a subscription is about to renew, I want to hear in time to decide, so I can keep or cancel before the money leaves.
J3
When I correct a category once, I want it remembered for that merchant, so I can trust my spending without double-checking.
04 · Design challenge
How might AI take work off people in their banking app, without taking control away from them?
My goal was never to add AI to a banking app. It was to find the few moments where AI takes work off the user, and to design only those.
What Strom does, and where it stops
I set these rules before designing any screen. Each capability comes with a limit it can't cross.
Strom doesRanks actions and insights by behaviour
Shows why, and lets people pin, reset or switch it off
Strom doesFlags subscriptions worth a second look
Never cancels or moves money on its own
Strom doesLabels every payment with a reason
Says "Not sure" instead of hiding uncertainty
Strom doesTurns corrections into rules
Every rule can be seen, edited and undone
With AI off, it's still a complete bank. Every AI feature has its own switch, and Home falls back to a fixed layout.
05 · Information architecture
Familiar tabs, AI underneath
The same five tabs people know from other banks. AI works underneath them instead of getting its own tab.
In scopeAI-driven surfaceOut of scope
Home
- Balance and account
- Strom insight slot
- Adaptive quick actions
- Explanation sheet
- Recent transactions
Insights
- Spending breakdown
- Categorisation accuracy
- Review queue
- Category → transactions
- Transaction detail
- Correction sheet
- Learning confirmation
Subscriptions
- Summary and potential saving
- Flagged for review
- Renewing this week
- Active and cancelled
- Renewal decision
- Cancellation outcome
- Pre-renewal nudge
Cards
- Card controls
- Freeze and limits
- Transfers and payments
More
- Profile and settings
- Strom intelligence controls
- Learned rules
- Investments and savings
- Onboarding and KYC
✦ Strom intelligence layer
Behaviour rankingOrders actions and picks the one insight for HomeHome
Recurring detectionFinds subscriptions and flags unused, duplicate or pricier onesSubscriptions
Confidence scoringFiles confident payments, queues ambiguous onesInsights
Learning from correctionsTurns edits into rules users can inspect and editInsights · More
ExplanationWrites the "why" behind every AI decisionAll AI surfaces
06 · User flows
Every flow feeds back into the next
In every flow, the AI does the routine work, hands the user the decisions that matter, and learns from what they choose.
ScreenAI layerSystem actionDecision
07 · Wireframes
Wireframes, with the reasoning behind them
Greyscale on purpose, to get structure and AI behaviour right before visual style.
08 · High-fidelity design
Twelve final screens
Strom means stream. Deep teal carries the brand, mint and a sparkle mark everything the AI produces. Click a screen to see it large.
Amber marks time-sensitive renewals, violet marks decisions waiting for the user. Manrope throughout, chosen for its clear numerals.
09 · Prototype
Try the three journeys
On screen
10 · Design decisions
The trade-offs I made on purpose
DecisionWhat it costHow I kept the cost small
01Ask instead of guessingA wrong label hurts trust more.
More questions at first.
One question per merchant. After three skips, Strom makes a marked guess.
02Always show when home adaptsA home that rearranges itself without warning feels unpredictable.
Less of a "magic" feel.
A visible label, pinning, one‑tap reset and an off switch.
03Keeping is a good outcome tooNudges that guilt-trip people get ignored.
Fewer cancellations.
Keep, remind and cancel weigh the same. Keeping mutes the flag for 90 days.
04Confidence in plain wordsA score like "0.82" means little to people.
Less technical detail.
"High confidence" or "Not sure" on each payment. Accuracy trend in Insights.
05Works fine without AISome people don't want AI in their bank.
Every AI feature needs a fallback.
Every AI feature has its own switch, and Home falls back to a fixed layout.
11 · Reflection
Limits and next steps
No primary research yet
The persona and hypotheses come from secondary sources and a benchmark. They point in a direction but haven't been validated.
No working AI behind it
Confidence levels, detection accuracy and data access are assumed. Real engineering limits would change some interactions.
Narrow on purpose
Subscriptions alone could fill a product. I went deep on three problems rather than covering the whole bank.
What I would test first
Nothing here has been measured yet. This is what I would test in a pilot, against the current app.
Faster everyday tasks
How I would measure it
- Time to first tap on key tasks
- Task success in usability tests
Action before the charge
How I would measure it
- Flagged renewals handled before payment
- Fewer refund requests
A chart people trust
How I would measure it
- Corrections per 100 payments over time
- Trust rating for Insights