A pan-African remittance business, 31,167 transacting customers, $9.54m in annual fees, growing at 561% year on year and preparing to raise capital. Keystone IQ was commissioned to establish what that growth was actually built on. This is what the Revenue Quality Architecture™ found in a fintech customer base, and what it changed about the investment conversation.

The engagement: a fintech customer base, read from the transactions up

The business moves money across African corridors for consumers, principally Ghana, Kenya and Nigeria. By the time it came to market it was growing at a rate few businesses ever see: fees up 561% year on year, transaction value up 507%, customers up 408%.

Whether it was growing was never the question. What a raise required was evidence of something harder to see. What was the growth built on, would the revenue hold, and could the existing customer book carry the next stage of the story?

That question is not answered by a growth rate. A growth rate is a net position. It tells you the direction of travel, not which customers produced it, whether they will still be there next year, or how much of the number had to be bought rather than earned. Blended churn describes the average customer. Average revenue per customer describes a customer who often does not exist.

Remittance also carries dimensions most customer bases do not. Corridors behave differently from one another. Payout rails fail at different rates. Compliance tiers gate whether a registered customer can transact at all. And every customer sits at the centre of a beneficiary network, the people they actually send to, which turns out to be the dimension that decides how durable the relationship is.

We worked from an anonymised transaction extract at individual customer level. No names, no contact details, no PII. Three numbers would have shaped how the business read itself going into that raise, and all three were misleading.

Download full case study here

What the analysis found

Rebuilt from observed behaviour, the base resolved into five sender groups rather than one population. Network Senders, frequency multiplied by reach: 23.6% of customers, 45.4% of fees, 27.8 sends a year across nine beneficiaries, mostly family support and tuition fees. Power Senders, frequency multiplied by size: 15.0% of customers, 23.4% of fees, 23 sends a year at $94,692 in annual transaction value, but to just 2.5 beneficiaries. Below them, Regular Senders on 22.6% of fees, then Sporadic and Light Senders on 8.6% between them.

The spread is starker per customer. A Network Sender generated $590 a year in fees, a Power Sender $477, a Light Sender $48.

Blended, annual churn read at 49.3% and the unit economics looked acceptable. Underneath the blend, three of the five groups destroyed value at a standard acquisition cost.

The dependency picture. Fee revenue grew 561%. New customers drove 79.5% of it.

The established base sustained only 53% of current fees without continued acquisition. With no new customers at all, the business sustained $5.04m against the $9.54m run rate, a fall of 47.2%. Acquisition was not optional headroom on top of a self-sustaining book. It was load-bearing.

The exposure picture. The group with the best headline retention was the most structurally exposed.

Power Senders had the narrowest beneficiary network in the book. Network Senders had nine reasons to keep transacting. Power Senders had roughly two. They also paid out almost exclusively to bank accounts, the rail Nigeria failed on hardest, where 45.7% of attempts failed by volume against 32% to 34% in Ghana and Kenya. For Power Senders, only 35% of attempted value converted. They carried 65% of Nigeria’s attempted value, which meant $17.29m failing against $18.46m converting, a 48.4% failure rate by value. Nigeria’s established-base revenue still grew 130% year on year regardless.

Demand was not the constraint. Execution was, sitting directly on the most valuable group in the base.

The dormant picture. 150,072 registered customers had never transacted, which reads as reactivation upside. Only 7,273 had ever attempted a payment.

The rest split five ways between incomplete verification and no demonstrated intent, none of it addressable through a reactivation campaign. The tried-but-failed group was the only one where a customer had wanted to transact and been blocked. It was 4.8% of the headline.

What the architecture also surfaced: $4.38m to $4.63m already in the base

The same analysis that showed where the risk sat also sized what was recoverable from customers already on the book, with no acquisition uplift assumed. Eight levers, sequenced and sized.

The largest, worth $3.36m, required no new capability. It required not losing what was already working. 6,882 new Network and Power Senders joined during the year, and those retained past 90 days compounded at +54.2% in transaction value the following year. Alongside it sat the acquisition mix: 27.5% of new customers landed in Network, Power or Regular Sender behaviour and carried 58.3% of new-cohort revenue. That landing rate was the assumption the entire floor rested on, and where a growth team is measured on customer count, nothing protects it.

How the Revenue Quality Architecture™ works

Step one: find the structure that exists in the data. We do not impose a segmentation. Customers are scored on observed behaviour and the boundaries emerge from where they actually cluster. Here, beneficiary depth separated two groups with similar value and opposite risk profiles. A segmentation by revenue tier would have put them in the same box.

Step two: track how customers move between those tiers. A static picture shows concentration. Movement shows what the base can support. Net movement was positive at +$534k, but the largest single loss line was silent churn, worth $302k: customers who simply stopped, with no failed attempt and no advance signal. By the time it appears in the numbers, the customer has gone.

Step three: apply operator judgement above the analytics. Analytics is increasingly commoditised. The data finds patterns in every dataset. Deciding which carry deal significance is a different task. The largest lever here was not a new initiative but a 90-day retention curve already working and an acquisition mix nobody was measuring. No model surfaces that as a priority.

Why this matters at entry, hold, and exit

The evidential bar has moved. Bain’s 2026 Global Private Equity Report puts the typical requirement at 10% to 12% annual EBITDA growth to deliver a 2.5x multiple on invested capital, against around 5% historically. In African fintech, the first half of 2026 saw $1.44bn raised across 146 disclosed deals, against 252 a year earlier for a near-identical total. Fewer companies, larger cheques, more evidence per deal.

Growth was already visible in this business. What the analysis added was what it was built on, how resilient it was, and what management needed to do before the next stage of capital.

These numbers rarely appear in standard reporting. They should.

Frequently asked questions

What is customer base due diligence?

Transaction-level analysis of a customer book, used to establish whether a reported revenue line is durable. It measures concentration, movement, retention by value group and acquisition quality. Financial due diligence verifies the numbers are accurate. Customer base due diligence establishes whether they will hold.

Is customer base due diligence the same as customer due diligence?

No. Customer due diligence is an anti-money laundering check on who a customer is. Customer base due diligence is a commercial check on whether a target’s revenue will hold.

Customer due diligence, usually shortened to CDD and closely tied to KYC, is a regulatory obligation. A bank or a payments business verifies identity, assesses money laundering risk and monitors activity because the law requires it. It runs continuously, and it sits with compliance.

Customer base due diligence runs on a deal. Keystone IQ rebuilds a target’s customer base from anonymised transaction data to show which customers the revenue depends on, how concentrated that dependency is, and whether it is durable enough to underwrite. It answers a pricing question, not a compliance one.

The clearest difference is what each one needs. CDD cannot work without personal identity data. Customer base due diligence requires none. No names, no contact details, nothing that identifies a person.

That last paragraph is doing three jobs at once. It’s the sharpest available contrast, since CDD is definitionally about identity and this is definitionally not. It reinforces “No PII required”, which already runs across the site. And it’s the bit most likely to get lifted, because it’s a clean binary rather than a description.

Does this work for fintech and payments businesses?

Yes, and the transaction data in payments is usually richer than in most sectors. Alongside frequency and value, a remittance or payments base carries corridor, payout rail, compliance tier and beneficiary network. In this engagement, beneficiary depth was the dimension that separated the group with the best retention from the group with the most structural security. They were not the same group.

What is a revenue floor?

What a business sustains from existing customers with no new acquisition at all. For the business above, $5.04m against a $9.54m run rate.

What data is needed, and is PII required?

An anonymised transaction extract covering 12 or more months of individual-level data. No names, no contact details, no PII. Most companies can produce it within 24 hours.

When should it be commissioned?

At entry, to establish what is being bought. During the hold, to run the value creation agenda against the same data. Ahead of exit, to build the equity story on observed movement. The Health Check delivers in five days, the Customer Base Diagnostic in two weeks.

Talk to us about your customer base

Know what it can sustain before you price it.

Days not weeks. Fixed scope. No PII required.

Read the full case study · Talk to us