Susu works because the collector shows up every day. It breaks for exactly the same reason: one person walks a line of stalls holding other people's savings, marks a card, and reconciles at night from memory.
Nothing about that is dishonest. It is just a system with no second copy. When a client says they paid ₵20 on Tuesday and the office ledger says ₵10, there is no third record to settle it — only two people who each believe themselves.
Here is what actually closes that gap, in the order it matters.
1. Make the record at the stall, not in the evening
The single highest-leverage change is moving the entry from the notebook-at-night to the doorstep. Not because evening entry is lazy, but because it is reconstruction, and reconstruction is where amounts drift, clients get skipped, and the day stops adding up.
An entry made while the client is standing there carries things an evening entry never can: the actual amount, the actual time, the collector's identity, and the channel the money arrived through. The client keeps a receipt. The office sees it within seconds. Now there are two matching records instead of one contested one.
The practical blocker is speed. A collector with a cash bag in one hand and a queue behind them is not going to type into a cramped form in the sun. That is why voice entry matters here more than in a branch: the collector says the amount, the app matches the client and shows it for confirmation, and the round keeps moving.
2. Keep savings and loans apart at the point of collection
This is the one most operations get wrong, and it is expensive.
On a typical round, the same collector takes a contribution from one stall and a loan repayment from the next. Those are opposite obligations:
| Contribution | Repayment | |
|---|---|---|
| Whose money is it | The client's | Yours |
| What it does to your books | Increases what you owe | Reduces what you're owed |
| What it is at year end | A liability | Income recognition against principal |
They arrive in the same bag, in the same denominations, on the same afternoon. If your system records them as one undifferentiated daily total, nobody can decompose the day afterwards — not you, not your auditor, not the client asking for their balance.
Record the meaning at the doorstep and the cash reconciles back into its two halves without anyone reconstructing anything.
3. Close per collector, every evening — not per branch, monthly
A shortfall discovered at month end points at nothing. A shortfall discovered the same evening points at one person, one round and one day.
So the close has to be:
- Per collector, not pooled across the branch
- Daily, not weekly or monthly
- Against an expected figure, not just a declared one
That means each round carries an expected collection derived from the client schedules, an actual collection, the collector's expenses, and what they remitted. The gap between expected and actual is a conversation you can have the same day, with the round still fresh in everyone's memory.
The point of a daily close is not catching thieves. It is that an honest collector who was short ₵40 can tell you why on Tuesday evening, and cannot on the 31st.
4. Make the round itself checkable
"I went, the shop was closed" is unfalsifiable on paper. With a planned round assigned to a named collector and tracked by GPS with a location history, it becomes checkable — and the behaviour changes before anyone checks.
This is not surveillance theatre. It is what makes an end-of-day cash settlement a reconciliation rather than an argument, because every contribution and expense carries the round it came from.
5. Gate the adjustments, not the collections
Here is the trade-off most systems get backwards. If you require approval for everything, your collectors stop using the app and go back to the card. If you require approval for nothing, your book is whatever the last person typed.
The split that works:
- A contribution posts immediately on confirmation. It happens eighty times a day, the amount is known, and the risk is mishearing — which the confirmation screen already handles.
- A balance adjusted by hand, a client signed up in the field, or a loan rescheduled after a slow market week waits for an approver. These happen rarely and change what someone owes or holds. The risk there is judgement, and no confirmation screen catches judgement.
That is what a maker-checker approval workflow is for: fast where volume lives, slow where consequence lives.
6. Assume the network is not there
Markets are where signal dies. If your app stops recording when the connection drops, collectors will build a paper workaround within a week — and the paper workaround becomes the real system.
Offline has to be the assumed condition of a round, not a degraded mode. Entries save on the device and sync when the signal returns, and the collector never notices the difference.
A note on what software cannot do
None of this makes an operation lawful. What you are permitted to collect, from whom, and under what registration is a question for your regulator and your lawyer — and in Ghana the tier you sit in decides it. Taking daily contributions from the public is a different activity from lending your own funds, and a susu enterprise does not become a microfinance company by growing.
Software's job is narrower and still worth doing: record the entry where it happens, attribute it to a person, keep the two obligations apart, and close the round the same day. Then the nature of your operation is visible in the records rather than asserted afterwards.
The card is not the problem because it is paper. It is the problem because it is the only copy, written by one party, checkable by nobody. Fix that and most susu disputes stop existing.
Want to see it on a real round? Bring one collector's client list and yesterday's card, and we will build that round and close it end to end on your own numbers.