What makes a new customer valuable?

Situation and task

The year's growth plan set a target for monthly active personal-banking customers well above where the year started. By the end of the first quarter, actives were behind plan and so were sign-ups (case study 02 sets out the gap and the levers). The base itself was in reasonable shape: the share of registered customers who were active was in line with what banks usually see. The problem was at the front door. A large share of new accounts never made a single transaction, and another large group was active for a while and then went dormant. Every one of those accounts had cost money to acquire.

The Head of Growth and the CMO asked me to find out what a valuable customer actually does in their early days, so that onboarding could push new users towards it rather than relying on reactivation campaigns to win them back later.

Five behavioural personas

I took the January 2025 onboarding cohort and followed every account for thirteen months. Each customer was classified by what they had done over their life on the platform. VAS here means value-added services: everyday payments such as airtime, mobile data and bills like electricity or TV.

Retention here is strict: active users in a month divided by everyone originally onboarded in that cohort. It measures how much of the acquisition spend is still alive, not just how busy the survivors are.

What the cohort showed

• Transfer-only customers collapse. They start about as active as everyone else and most are gone within three or four months, in every region. Moving money in and out is a utility, not a relationship.
• Early VAS customers hold and earn. Their retention eases off after the first couple of months, then settles just above half the cohort. They transact far more than anyone else, and their revenue per active user keeps climbing through the year. By the end, the gap over transfer-only customers is several-fold on retention, and their cumulative revenue is several times that of any non-VAS group.
• Late VAS is the most loyal, but slow to pay. Customers whose first everyday payment came after month one have the highest, flattest retention curve of all. Their revenue starts close to transfer-only levels and climbs steadily all year, but still finishes below early VAS. The first month is the cheapest point to start that climb.
• Card-only and hybrid customers stay, but stay small. They retain reasonably well and keep a modest level of activity, yet earn little on the payments that drive revenue. They are the obvious cross-sell pool.

One cohort could be a fluke, so I repeated the analysis on every monthly cohort through to January 2026, twelve more in all. The ordering of the personas, and the size of the early-VAS gap, barely moved from month to month. The newest cohorts opened a little higher, especially late VAS users, which I flagged as worth watching rather than a break in the pattern. That stability was what made the finding usable as a planning baseline rather than an interesting chart.

Everyday payments come on top, not instead

A fair worry about pushing everyday payments is cannibalisation: does an airtime purchase just replace a transfer the customer would have made anyway? Splitting each persona's monthly transactions by type answered that. Early VAS customers made about four times as many transactions a month as transfer-only customers, and more than twice as many transfers. The everyday payments came on top of the core activity, not instead of it. Once a customer buys airtime in the app, it tends to become the account they run their money through.

The revenue mix makes the same point from the other side. Mobile data was a small slice of early VAS customers' transactions but close to a quarter of their revenue, so a data purchase earns several times what a transfer does. Late VAS customers show the shift happening: they started out earning nothing from everyday payments, and by the end of the year airtime and data made up about half of their revenue.

The details that shaped the design

The second purchase is the habit

Looking at what customers did second, not just first, everyday payments jumped. Many customers who started with a transfer moved to an airtime purchase next; it was the most common path on the platform. Most top-tier users made their second everyday payment within a few days of the first. A habit forms in days, not weeks. That pointed at a reward for the second purchase, inside a short window.

Agent-referred and digital customers are different customers

Customers onboarded through the agent and referral network got a physical card as part of signing up, so card adoption looked high from day zero. That was a process step, not intent. Their first transfer came later (a trust lag), but once active they were more loyal. Digital customers churned harder early, yet the ones who stayed earned more per head. One onboarding journey for both would underserve both.

Region changes the product, not the principle

The early-VAS effect held nationally. What varied was which product led. Airtime was common everywhere, but mobile data adoption was far lower in the north than the south, so a data-first message would miss many northern users. The southern markets earned more per user but were more competitive and churned faster.

Verification tier is a revenue multiplier

Customers at the highest verification tier (KYC level) earned considerably more and retained better in nearly every persona. Higher tiers unlock higher limits and more products, so verification is not only a compliance step. It also enables value, which made it a lever worth rewarding.

Who is stuck at the lowest verification tier?

If each step up in verification tier carries more revenue per customer, the next question is where the customers stuck at the bottom actually are, because that is where upgrade nudges should go. I broke the customer base down by single year of age, onboarding channel and region. Three things stood out: the youngest customers are overwhelmingly at the lowest tier, the share at the bottom is smallest in the thirties and creeps back up after the mid-forties, and agent-referred customers are far more likely to stay at the lowest tier than digital ones at almost every age.

Balance-holders need a different offer

A minority of dormant accounts held meaningful balances without transacting. For them, a payments cashback is the wrong tool: their value is in deposits. I split them out as a separate savings opportunity.

How I did it

1. Define the outcome. Strict cohort retention (active this month over everyone onboarded) plus volume and revenue per active user, always read next to retention so a shrinking-but-busy cohort could not look healthy.
2. Classify behaviour. Lifetime transaction history rolled up into five mutually exclusive personas, with the 30-day line separating early from late VAS.
3. Replicate. The same queries run on each monthly cohort across a year, comparing persona curves and gaps rather than trusting one month.
4. Break down the mix. Each persona's monthly transactions and revenue split by type (transfer, card, airtime, data, bills), to test whether everyday payments replace core activity or add to it.
5. Sequence analysis. First-to-second transaction transitions and the days between them, to find the habit-forming step.
6. Cut by context. Region, verification tier, onboarding channel and balance behaviour, to find where one national answer would break; then verification tier by single year of age to see who is stuck at the bottom.

Caveats and what I'd do differently

• This is correlation. Early everyday payments might mark customers who were going to be good anyway. I said so at the time and designed controlled experiments to separate cause from selection, but not all of them ran. The strongest evidence is the before-and-after change once onboarding was rebuilt, which has caveats of its own (case study 02).
• Personas use lifetime behaviour. That makes them clear but slightly circular for prediction. The day-by-day activation view in case study 02 was my fix; I would build it first next time.
• Survivorship in per-user averages. Volume and revenue are averaged over customers still active, which flatters personas that shrink. I always showed them next to strict retention; next time I would also chart revenue per onboarded user as the headline.
• Card adoption in the agent channel is procedural. I excluded it as a behavioural signal for those customers; a cleaner event log would have let me separate issued from used cards.

Skills shown