Why most RPA projects fail in West Africa; and what to do instead.
Robotic Process Automation has generated more enterprise disappointment per dollar spent than almost any other technology category in West Africa over the past five years. This is not a technology problem. It is a deployment context problem; and it is almost entirely predictable in advance, which makes it all the more frustrating to watch unfold.
The RPA promise and the West African reality
The vendor pitch for RPA is compelling: software robots that work 24 hours a day without holidays, at a fraction of the cost of a human employee, executing rules-based tasks with perfect consistency. For a Nigerian enterprise managing repetitive back-office processes; account reconciliation, invoice processing, data migration; the ROI models look extraordinary on paper.
The models are usually right. The deployment context is usually catastrophic for them.
RPA bots are, fundamentally, interface recorders. They interact with software the way a human would; clicking buttons, reading screen values, entering data into fields. This architecture is highly sensitive to the stability of the interfaces they interact with. Change the font, move a field, update the underlying application, or introduce a network latency spike, and the bot breaks. A human handles these variations automatically. An RPA bot stops and waits for intervention.
In the US or Western European enterprise context, where legacy systems are often stable, networks are reliable, and change management is disciplined, this brittleness is manageable. In the West African enterprise context, it is a deployment-ending vulnerability.
The five failure modes; in order of frequency
Infrastructure instability
The most underestimated failure mode. RPA bots depend on the availability and consistency of the systems they automate. Nigerian enterprises; including Tier-1 banks and major telcos; experience application downtime, network interruptions, and interface changes at rates that RPA architectures built for Western markets cannot handle. A bot that achieves 95% uptime in Atlanta is failing 150+ hours per month in Lagos. The economics of the business case evaporate before the first quarter of operation is complete.
Process selection failure
RPA implementations succeed when they automate processes that are stable, high-volume, and rules-based. West African enterprises, under pressure to show AI transformation progress, frequently select the highest-visibility processes rather than the best-fit ones. HR onboarding workflows with conditional logic. Customer complaint routing with local language nuance. Procurement approval chains with relationship-based exceptions. These processes look automatable on a flow diagram. They are not automatable with RPA. The bot exceptions exceed the bot executions within six months.
Governance and maintenance neglect
The RPA vendor's business model is implementation. Post-deployment maintenance; bot monitoring, exception handling, system-change management, version control; is either charged at premium rates or not addressed in the original contract. Nigerian enterprise procurement teams, focused on implementation cost, often do not budget for the maintenance overhead that keeps a bot estate functional. The typical West African RPA estate starts degrading in month four and is partially abandoned by month twelve.
Change management underinvestment
RPA is often sold as a workforce supplement rather than a workforce change. In practice, it changes job roles, eliminates tasks that people perform, and requires new skills; bot monitoring, exception handling, process documentation; that most organisations do not have. The humans who were doing the automated work frequently become the people responsible for fixing the bots when they break. Without deliberate change management, this produces resentment, passive sabotage, and adoption failure that no technical fix can address.
Vendor mismatch
The global RPA market is dominated by vendors whose implementation frameworks, support infrastructure, and success stories are built on large, stable enterprise environments in North America and Europe. Their Nigerian implementation partners; frequently system integrators with limited RPA depth; are selling certifications, not field experience. The gap between what is promised in the sales process and what is delivered in the field is not fraud; it is a genuine capability deficit that the market has not yet corrected for.
What to do instead: an architecture for the West African context
The solution is not to abandon automation. The processes that RPA promised to automate are real, and the economic case for automating them is real. The failure is in the technology choice and the deployment approach, not in the ambition.
The architecture that performs in the West African context shares four properties that standard RPA does not:
1. API-first integration over UI scraping
Where the underlying system exposes an API; and most modern enterprise systems do; automate against the API, not the interface. API integrations are orders of magnitude more stable than UI automations. They survive interface updates, theme changes, and browser upgrades without breaking. For legacy systems without APIs, the correct investment is often a lightweight API wrapper around the legacy application, not an RPA bot that clicks through the legacy UI.
2. Agentic AI for processes that require judgment
The class of processes where RPA most frequently fails; those with conditional logic, natural language inputs, or relationship-based exceptions; is precisely the class where agentic AI performs well. An AI agent with tool access can read an unstructured supplier invoice, interpret the payment terms, cross-reference the purchase order, and route the exception to the right approver with a summary of the discrepancy. An RPA bot cannot. The two technologies serve different process archetypes. Using the right one for each process type eliminates most of the failure modes described above.
3. Infrastructure-aware deployment
Any automation system deployed in West Africa must be designed for the infrastructure it will actually run on, not the infrastructure the vendor's reference architecture assumes. This means: graceful degradation on network interruption, local state persistence to survive application downtime, and exception queues that allow human processing during bot unavailability. Designing for failure is not pessimism. It is the only professional standard for the environment.
4. Internal capability, not vendor dependency
The durable automation deployments I have seen in Nigerian enterprises are run by internal teams who understand what they have deployed, can modify it when processes change, and can debug it when it fails. This requires deliberate knowledge transfer from the implementation partner, documentation standards that survive staff turnover, and hiring or developing at least one person internally who owns the automation estate as a technical discipline.
A practical migration path
For organisations currently running a degraded RPA estate, the path forward is not a wholesale replacement. It is a triage:
- 01Identify the bots that are actually running reliably and leave them alone. RPA works for a specific class of problems and those bots are evidence of a good process-technology fit.
- 02For bots that are failing regularly, diagnose the failure mode before replacing them. API availability, process stability, and maintenance capacity are distinct problems with distinct solutions.
- 03For new automation initiatives, start with the process architecture before the technology selection. Map the process, classify the decision types within it, and then select the automation technology that fits the decision type.
- 04Build the internal maintenance capability in parallel with any new deployment. The half-life of an automation system without internal owners is approximately 18 months in the West African enterprise context.
The opportunity, correctly framed
The automation opportunity in West Africa is larger than any other region I am aware of, for a simple reason: the labour arbitrage that suppressed automation investment in Western markets for decades does not apply here in the same way. The economic case for intelligent automation in Nigerian enterprises is driven not by replacing expensive labour but by enabling scale without proportional headcount growth; the only viable model for an enterprise competing in a market where skilled talent is scarce and salary inflation is compounding.
The organisations that get this right will build operational leverage that their competitors cannot close without a fundamental rearchitecting of their own systems. That advantage is available now. But it requires building it correctly; which means understanding why the first generation of automation investments failed, and making different architectural decisions the second time around.