Architecture advice written for a European or North American environment assumes things that are not reliably true in much of West Africa: constant grid power, cheap low-latency bandwidth, and a support engineer two time zones away at most. Those assumptions are not small details. They change what the right design is.
The constraints are real, and they are engineering inputs
Power. Grid reliability varies by location and by season. A site without conditioned power and a tested failover path is a site with an availability risk that no cloud region can absorb. Where workloads run on-premises or at the edge, power design is part of the availability design.
Connectivity. Fibre routes can be cut, and international transit can add tens of milliseconds. Applications designed with chatty protocols between a local user base and a distant region will feel slow in a way that is architectural, not a matter of tuning. Caching, regional placement and asynchronous patterns are not optimisations here; they are requirements.
Latency. Round-trip time from Accra or Lagos to a European region is workable for many business applications and unsuitable for others. Measure before deciding, with the real user population, rather than reasoning from a map.
Talent and on-call. A support rotation staffed from another continent introduces response gaps that an SLA document does not fix. Coverage in the business's own time zone matters more than the size of the provider.
Residency and regulation are now architecture decisions
Data protection regimes across the region have matured, and the questions they raise are specific:
- Ghana operates a Data Protection Act with registration and processing obligations for controllers.
- Nigeria has a data protection framework with clear expectations on lawful processing, breach notification and cross-border transfer.
- South Africa applies POPIA, with conditions on processing and transfer of personal information outside the country.
- Sector rules in financial services and health add further constraints on where data may sit and who may access it.
The practical consequence is that residency cannot be an afterthought negotiated with legal after go-live. Workloads holding personal data need a documented placement decision, and that decision needs to survive an audit.
Where the data sits is a decision with legal, latency and cost consequences. Make it once, deliberately, and write it down.
What a resilient design looks like here
Design for degraded connectivity. Offline-tolerant clients, local caching, retry with backoff, and queues between systems that do not need to be synchronous. Assume the link will fail and make failure boring.
Place compute close to the user where it changes the experience. Not every workload, and not by default. The ones where latency directly affects revenue or safety.
Treat power as part of availability. Conditioned supply, tested failover, and monitoring that distinguishes a power event from an application fault.
Keep a warm recovery path in-region where regulation requires it. A backup in another jurisdiction that cannot be restored into a compliant environment is not a recovery plan.
Instrument the things that actually break. Link quality, DNS resolution time, certificate expiry, disk pressure and queue depth. Most incidents announce themselves through one of those five.
What to insist on in an SLA
An SLA expressed only as a monthly uptime percentage is close to meaningless in this context. Insist on:
- Defined response times per severity, with the clock starting at detection, not at ticket creation.
- Coverage in your time zone, with named escalation paths and out-of-hours contact that works.
- A recovery time objective you have seen demonstrated, not one written in an annex.
- Reporting cadence and content: what is measured, how it is presented, who reviews it with you.
- A change and maintenance window agreement that respects your business hours, not the provider's convenience.
Reliable infrastructure in West Africa is not a matter of importing a blueprint. It is a matter of designing honestly for the environment the business actually operates in, and then operating it with the discipline the design assumes.