Most enterprise networks degrade politely. A page loads slowly, a call drops, someone reconnects. Financial infrastructure does not have that middle state, because a great deal of what it does is tied to a clock. The cost of an outage is not the duration — it is whether the duration overlapped something that had to happen.

Somalia is an unusually sharp case. Mobile money is not a convenience layer here; for a large share of the population it is the payments system. That makes the connectivity underneath it systemically important, and it means an institution’s network design is a settlement-risk question rather than an IT one.

Three failures, in order of cost

The window

Reconciliation, settlement and reporting run to deadlines. A circuit that is up 99% of the time is excellent for email and useless for a process that has to complete between two fixed times, because the outage does not have to be long — it has to be badly placed. An institution that has priced its connectivity on availability alone has priced the wrong variable.

The branch that stays dark

Branches, agents and ATMs are usually connected on a single path, because that is what was quoted and what the building supported. On a single path there is no recovery plan short of physical attendance. In a country where attendance can take a day, a routine fibre cut becomes a site closed for a day — with the queue outside it.

The records nobody can locate

The slowest-moving and most expensive. Customer records held in a jurisdiction the institution cannot point to, under a default chosen years ago by someone configuring a platform. Since March 2023 there has been a Data Protection Act and an authority behind it, which turns a latent problem into a question somebody may eventually ask.

What a design that survives these looks like

Sites on protected paths, not spurs. A metro ring means a break in one direction reroutes rather than isolates. This is a topology decision made when the service is ordered, and it is difficult to retrofit.

At least one path that is independent in kind. Two fibres in the same trench are one fibre. For sites that genuinely cannot drop — a head office, a switch, a data centre link — the second path should be satellite, because it shares no duct, no route and no landing station with the first.

In-country hosting for what has an obligation attached. Not everything. The subset with a residency requirement, plus a restore point that is inside Somalia and in a different building from production.

International capacity with verifiable diversity. Correspondent banking and cloud platforms are reached over submarine systems. Ask which systems, and where the two paths actually diverge.

The question that reframes the budget

Connectivity is usually procured against a monthly cost and an availability percentage. Neither number describes the risk. The more useful framing is: what is the cost of this site being unavailable for four hours during the business day, and how does that compare with the annual cost of a second path?

For a back-office location the honest answer is often that a single path is fine and should be priced as such. For a branch handling agent float, or a link carrying settlement traffic, the arithmetic usually inverts the moment anyone does it — and it tends not to get done because connectivity sits in a different budget line from the loss it prevents.

What we would ask in a first meeting

Which processes have a deadline. Which records have an obligation attached. How many sites, and what each one does when it is offline. That is enough to design against, and it produces a network that is expensive in exactly the places where expense is justified and cheap everywhere else — which is the opposite of what a flat service catalogue produces.