A slide says almost every payment succeeds and the room relaxes, while somewhere a merchant has just lost a Saturday because one operator, at one amount, failed all morning.
Averages make a good sales tool and a weak operating tool, because they hide the bad day.
The blend is not the shop
Most of the traffic can succeed while a slice of it is unusable: a rail that works at noon and fails on payday, an amount band that always times out, or a test ledger that looks fine while live money does not move.
If you only watch the blend, you will keep the partner who is cheap on the slide and expensive at the till.
I have sat in rooms where the number on the wall was healthy and the merchants' WhatsApp thread was not, and both were true at the same time. The blend averages many shops, many hours and many methods, but a merchant lives in one shop, at one hour, with one method.
Slice before you pick
Ask for the same number cut by rail, by operator, by amount and by environment. A partner who cannot show that does not know their own factory.
Then ask what happens on the slice that fails: who sees it, how fast the merchant hears about it, and whether the buyer gets a receipt or a shrug.
A partner who only reports the blend is asking you to underwrite their worst hour.
Live is not test
A sandbox that always returns success tells you nothing about the rail. Live Wave in Côte d'Ivoire promises nothing about another country, live MTN stays pending until the customer approves on their phone, cards have a minimum amount, and wallets only appear on supported devices.
If a partner shows you one success rate for all of that, they have already decided you will not look.
Ask what the number leaves out, such as timeouts, abandoned payments, test traffic or an operator they quietly switched off. If the list of exclusions is longer than the chart, the chart is a brochure.
The hour that matters
Payday, lunchtime, the evening after a TV spot and the Saturday market are the hours that matter most. They are the reason you hired a processor in the first place, so they cannot be waved away as a rounding error.
I care less about a monthly blend than about whether the method your customers actually use still works at half past six on the 30th. If it fails then, the month can still look fine while the shop suffers.
Amount bands lie in the same way. A rail that loves 2,000 F CFA and chokes at 50,000 F CFA will still post a handsome average if most tickets are small. Ask for the band your catalog actually sells.
What a receipt is for
When a slice fails, the buyer should not have to guess. They need a receipt, a pending state they can see or a clear decline, because a shrug in a chat window leaves the merchant doing the processor's job.
Ask who owns the conversation when money is stuck. If the answer is always you, the merchant, price that into the contract, because support is part of the rail.
What we watch
We watch the slices because the merchant lives in them. One bad operator is not a rounding error if that is the method their customers use.
We will not sell you a blended success rate as proof that money moved. When a day goes badly we should say so, and then fix the slice.
None of this means we never fail; it is about how we talk about failure. The dashboard shows the payment, the method and the status, and webhooks fire for the event that actually happened, even when another event would have made the slide prettier.
What this is not
This is not a published uptime number. We have no sourced figure to put in the first sentence, so I will not invent one, and incidents go on status.lomi.africa because a blog post is the wrong place to launder a percentage.
Nor is it advice to pick the most expensive partner. It is advice to refuse a single blend as the whole story.
How the blend gets built
A blended rate is a weighted average. If most of the volume sits on the rail that works, the number stays pretty while the thin rail your customers actually use is failing, and that small slice can be the whole weekend for one shop.
That is why I ask for the denominator, since success can be measured against attempts, against completed checkouts or against money that reached the merchant's balance. Those are three different factories, and mixing them is how a slide stays calm.
I also ask who is in the sample: all merchants, only those above a volume threshold, or only those who did not complain. If the sample is made of people for whom the factory already works, the number will flatter it.
Product last
On lomi. you can see the method on each payment and whether it is pending or completed, and a Wave sale in Côte d'Ivoire shows as Wave instead of disappearing into a generic mobile money bucket. That will not save a bad hour by itself, but it stops us from talking about that hour as if it were part of a blend.
What to ask on Monday
Before you renew a contract, ask for last month cut by rail and by environment, ask for the worst hour instead of the best day, and ask who gets paged when a slice fails and how the merchant hears about it.
If the answer is a single percentage, you already know enough.
A healthy average can still be one shop having a terrible day.
lomi. is payment infrastructure for francophone West Africa. Accept payments, send payouts, and automate financial workflows with ease. To learn more about what drives our product roadmap, read our blog, visit our GitHub, connect your AI agent to our services, or just create an account.