Most fintech companies don’t have a reconciliation problem when they start. They have a spreadsheet, a few hundred transactions a day, and someone who eyeballs the numbers every morning. It works.
Then volume doubles. Then you add a second corridor. Then a third IMTO partner. And suddenly the person eyeballing the numbers is drowning — and nobody notices until there’s a $40,000 discrepancy sitting in a settlement file that should have been caught three weeks ago.
I’ve watched this happen across multiple companies and multiple African corridors. The pattern is always the same: reconciliation is treated as an accounting task until it becomes an operational crisis. By then, it’s expensive to fix.
Here’s what actually breaks and how to rebuild it.
The Three Stages of Reconciliation Collapse
Stage 1: The Single-Source Trap. Early on, teams reconcile against one source — usually their own system. They check if the transaction was recorded, maybe compare it to a bank statement. This feels fine until you’re working with IMTO partners who send their own settlement files, mobile money operators with their own transaction logs, and agents with their own float records. You now have three or four sources of truth, and nobody has defined which one wins when they disagree.
Stage 2: The Manual Matching Wall. As volume grows, someone starts spending four hours a day copying transaction references between spreadsheets. They’re matching by eye, using CTRL+F, and relying on memory for edge cases. At 500 transactions a day, this is painful but survivable. At 5,000, it’s a full-time job for multiple people. At 50,000, it’s physically impossible to do accurately.
Stage 3: The Exception Backlog. Unmatched transactions pile up. Nobody has a clear process for classifying them — is this a tolerance issue, a timing difference, a duplicate, or a genuine loss? Without classification, exceptions don’t get resolved. They just accumulate until someone senior asks why the numbers don’t add up, and the team scrambles to explain a backlog they stopped understanding weeks ago.
What a Rebuilt System Looks Like
The fix isn’t buying a tool. It’s building a framework. Tools help, but if you don’t have the logic right, you’ll just automate the wrong process faster.
Define your matching hierarchy. For every partner and every corridor, document which fields are used to match transactions, in what order. In our case, that might be: IMTO reference → amount + date → beneficiary phone number → manual review. The hierarchy is different for every partner because their reference formats are different. MoneyGram references don’t look like RIA references. Write it down.
Classify exceptions on creation, not on review. When a transaction doesn’t match, it should be immediately classified into one of four categories: tolerance (the amounts are close enough), adjustment (the partner has confirmed a correction), reversal (the transaction was cancelled), or dispute (we disagree with the partner’s record). Each category has a different resolution path. If you’re dumping all exceptions into one bucket called “unmatched,” you’re guaranteeing a backlog.
Set tolerance thresholds by corridor, not globally. A $1.00 tolerance makes sense for a USD corridor. It makes no sense for a corridor denominated in Guinean Franc, where exchange rate rounding can create $3-5 differences on every transaction. Your thresholds need to reflect the reality of the currency you’re operating in.
Automate the match, not the judgment. The matching itself — comparing references, amounts, dates across sources — should be automated. We use Google Apps Script for this because it’s free, it sits inside the same Sheets environment the team already works in, and it runs fast enough for our volumes. But the judgment calls — whether to approve a tolerance exception, whether to escalate a dispute — should stay with a human. Automate the boring part. Keep humans on the decision part.
Build a daily sign-off, not a monthly review. If the first time anyone looks at reconciliation is at month-end, you’re three weeks too late. We run a daily reconciliation sign-off: each channel lead confirms their corridor is matched, flags exceptions, and signs off before end of day. Problems get caught in 24 hours, not 30 days.
The Governance Layer Most Teams Skip
The technical system is half the job. The other half is governance — and this is where most fintech ops teams fall short.
You need clear ownership. Not “the finance team does reconciliation.” Specific people own specific corridors or channels. In our structure, we moved from geography-based ownership (one person owns Sierra Leone) to channel-based ownership (one person owns all IMTO reconciliation, another owns agent/cash, another owns payout partners). Channel-based is better because the reconciliation logic is more similar within a channel than within a country.
You need escalation timelines. An unresolved exception at Day 1 is a data point. At Day 7, it’s a problem. At Day 30, it’s a potential loss. Define when each category gets escalated and to whom.
You need an exceptions register that’s visible to leadership. Not a hidden tab in a spreadsheet. A log that tracks every exception, its classification, its age, and its resolution status. If leadership can’t see the exception backlog in 30 seconds, the governance isn’t working.
The Cost of Getting This Wrong
Every fintech that’s scaling in Africa will hit this wall. The question is whether you hit it at 1,000 transactions a day — when it’s cheap to fix — or at 50,000, when the backlog is six figures and the partner is asking questions you can’t answer.
Reconciliation isn’t glamorous. Nobody raises a Series A because they have great reconciliation. But plenty of companies have lost partner relationships, eaten settlement losses, and burned operational trust because they didn’t.
Build the framework before you need it. By the time you need it, it’s already too late to build it well.
Taha El Ghrib runs cross-border payment operations across 15+ African corridors. He writes about payment infrastructure, operational scale, and what actually breaks at TheOpsDesk.co.