Quantifying engagement and retention in banking apps

Two customers funded accounts on the same Tuesday. Ninety days later, one has her salary landing on the 28th, three direct debits leaving from it, and a card she taps eleven times a week. The other still holds the £50 he moved across to clear the funding step, and opens the app only when a fraud alert wakes it up. Both count as funded accounts in the acquisition report. One of them is a banking relationship.

This lesson starts at day 31, after the application steps the funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → lesson maps. The question here is behavioural: which app actions tell you a relationship is live, how to compute them, and which features (gamified, or plain utility) move those specific numbers rather than just moving activity.

Active-user rates: who is actually here

The base engagement metric is the active-user rate: the share of your customer base that uses the product in a given window.

Two standard windows:

• MAU (monthly active users): unique customers who performed a meaningful action in the last 30 days.
• DAU (daily active users): same, but a single day.

"Meaningful action" is a decision you make, not a given. Logging in to check a balance counts as active. Mature teams still separate passive activity (a balance glance) from transactional activity (a payment, transfer, card action, product application), because the second predicts revenue and the first predicts very little.

The trap sits in the event list. If any account movement counts, a customer who mentally left two years ago but never cancelled a 4.99 subscription debit stays active in your dashboard forever. Interest postings, round-ups on a dead card and standing orders do the same work. Require a human-initiated event: a login, a manual payment, a card authorisation, a settings change. Run both definitions in parallel for a month and look at the gap. That gap is made of customers you have already lost, sitting inside the number you report upward.

The stickiness ratio

Divide DAU by MAU to get the stickiness ratio: the share of monthly users who show up on any given day.

DAU = 400,000
MAU = 1,400,000
Stickiness = 400,000 / 1,400,000 = 0.286 = 28.6%

28.6% means the average active user opens the app roughly 8 to 9 days a month. Banking apps run structurally high here, because balance checks and card management are habitual. Which published figures you may fairly compare that against, and how to adjust for market and product mix first, is the benchmarks lesson's business; the reading you own is your own trend across three quarters.

Stickiness also hides shape. Plot DAU by day of month. If it triples between the 25th and the 2nd and flatlines mid-month, you have a payday app rather than a habitual one, and the monthly average conceals it. The fix is a reason to open on the 12th: a spend tracker, a goal with visible progress, a bill forecast.

Primary-account share: the metric that pays the bills

Primary-account share is the percentage of your customers for whom you are the main financial institution. A primary customer routes salary to you, pays bills from you, taps your card daily. A secondary customer parks a little money and forgets you, and the deposit and interchange gap between the two is what the LTVLTVLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition → lesson prices.

How to define "primary"

No regulator defines it for you, so pick observable signals and hold them still:

• salary-type inflow present (the strongest single signal)
• debit card transactions above a threshold (5-plus a month is a common floor)
• recurring direct debits or bill payments set up

Working rule used by many teams: primary if a recurring salary-type credit lands and the customer makes at least a handful of card transactions that month.

Total active customers = 1,400,000
Customers with salary inflow + 5 or more card txns = 588,000
Primary-account share = 588,000 / 1,400,000 = 42%

Two things that rule gets wrong. Gig, self-employed and pension income arrives as irregular credits from several payers, so a strict "same payer, same date, three months running" test files a whole segment as secondary permanently. If primary share then drives who receives deepening offers, that segment never gets one and the rule becomes self-fulfilling. And in households one partner often receives the salary while the other spends: the second account looks like a side pocket and is nothing of the kind.

Card top-of-wallet has a cousin of that problem: you see your own authorisations, never your competitor's. The usable proxies are the share of days in the month with at least one authorisation, and the presence of low-ticket habitual merchants (transit, groceries, coffee). Four transactions of £200 is not top-of-wallet. Forty transactions of £6 is.

What gamified and utility features actually move

Kakao Bank, which passed 20 million customers in South Korea, runs a 26-week savings challenge: the customer commits to a deposit that rises each week, and the app shows the streak filling in with character rewards. It builds a savings habit and it manufactures app opens in weeks when nobody had a reason to log in. The honest reading is narrower than the internal excitement: small incremental deposits move balances slowly, so a team measured on DAU will call it a triumph while the deposit line barely shifts. Judge a gamified feature against the number it was aimed at, not the number it happens to lift.

Commonwealth Bank of Australia took the utility route with Benefits Finder in the CommBank app, matching customers to government rebates and refunds they had not claimed. It earns opens, goodwill and main-financial-institution positioningpositioningThe mental space you want your brand to occupy in your target customer's mind relative to alternatives.View full definition →. It also has a ceiling: someone who opens the app to claim a rebate has not moved a salary. Utility features win the login. Salary-attached features win the relationship.

Monzo works the second lever. Get Paid Early, and Salary Sorter, which splits an incoming salary across Pots the moment it arrives, both hang off the one inflow that reclassifies a customer from side pocket to main account. Neobanks historically struggled with exactly that: plenty of users, salaries still at the incumbent. Treat published primary-share claims as company-reported.

For payments-system context behind these behaviours, the Federal Reserve's payments research hub is a free, credible starting point.

Attrition curves: dormant versus churned

In a subscription app, churn announces itself: the customer cancels. In banking almost nobody closes an account. They go quiet. So you build attrition curves that separate two states.

The two states

• Dormant: open account, perhaps a small balance, no meaningful activity for a defined period. Recoverable with the right campaign.
• Churned: salary redirected, balance drained, no logins. Recovery unlikely.

Treating dormant accounts as churned makes retention look worse than it is and starves win-back. Treating churned accounts as dormant flatters the number and hides the leak.

Defining the thresholds

Set explicit, time-based rules. One workable framework:

Illustrative, not regulatory. (Some jurisdictions define a "dormant account" in law for unclaimed-funds purposes, the UK dormant assets scheme among them; that is a compliance concept, not your marketing metric.)

Any behavioural rule needs a balance override. A customer holding a large term deposit who logs in once a year is silent, not lost, and classifying them dormant aims a win-back offer at one of your most profitable relationships. Activity defines engagement; value decides who matters.

Building the curve

A retention curve plots the share of a cohort still active over time. Take everyone who opened in January, then measure how many remain active at months 1, 3, 6 and 12.

Month 1:  10,000 active   (100%)
Month 3:   7,200 active   (72%)
Month 6:   5,800 active   (58%)
Month 12:  5,100 active   (51%)

Two readings. The early drop (28% between months 1 and 3) is activation failure: customers who never set up direct deposit or never made a first card purchase leave fastest. The flattening point is your durable core, so a plateau near 50% at month 12 is your real customer base whatever the download figure says.

Putting the three metrics together

The three answer different questions and only make sense read together:

1. Active-user rate tells you how many people show up.
2. Primary-account share tells you how many treat you as home.
3. Attrition curve tells you how long that lasts.

Strong MAU with weak primary share is the classic neobank shape: lots of users, few salaries. Strong primary share with a steep early curve is a good product with a bad activation sequence. One metric alone lets you tell yourself either story.

A quick diagnostic

• Is stickiness rising or falling over three months, and is the day-of-month curve flat or payday-shaped?
• Is primary-account share rising quarter over quarter? If not, you are buying side-pocket users.
• Does the retention curve flatten, or keep sliding past month 12? A curve that never flattens is a structural leak, not onboarding friction.

Why classification changes your budget

At-risk and dormant customers earn win-back spend, and the offer that works is usually the salary one ("set up your direct deposit and get X") because that inflow is the lever. Churned customers rarely justify it. Misclassify 100,000 churned customers as dormant and you fund reactivation offers almost nobody redeems, then conclude that win-back does not work. The classification was the failure, not the tactic.

There is a second-order cost too. A team measured on DAU discovers that the cheapest lever is push volume. Opting out is one tap and in practice permanent, so every over-notified cohort shrinks the audience you can still reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → when the dormant-stage campaign arrives next quarter. You borrow engagement now against reactivation later.

Key Takeaways

• Define "active" as human-initiated. Counting standing orders, interest postings and round-ups keeps long-departed customers alive in your MAU.
• Primary-account share (salary inflow plus regular card activity) predicts revenue better than downloads or MAU, but check the rule against gig income and household patterns before you let it drive targeting.
• Never equate dormant with churned, and override the behavioural rule when the balance is large: silence from a big depositor is not attrition.
• Read the curve's shape. A steep month 1 to 3 drop is activation failure; the plateau is your durable core.
• Judge a feature on the number it was aimed at. Gamified savings (Kakao Bank's 26-week challenge) buys habit and opens; utility (CBA's Benefits Finder) buys goodwill; salary-attached features (Monzo's Salary Sorter, Get Paid Early) buy the relationship.