The pitch was flawless. A ministry team and I sat through a vendor demo for a national analytics platform — a wall of live dashboards, a satellite map shading itself in real time, an assistant answering questions in plain English. Heads nodded.
Then a deputy director at the back asked the question that mattered: "Does it work when the line drops?" The room waited. The sales engineer reached for the word "offline", hesitated, and moved on. The platform had been built in a city with fibre and standby power, for a country where neither is a given. It was a beautiful product for somewhere else.
That gap — between the tool and the place it is meant to serve — is the most expensive mistake a government can make with data. And it is almost always avoidable.
The locale swap is not a strategy
The default move, when a Western data platform meets an African ministry, is to translate it: swap the language pack, change the currency symbol, keep everything else. The reality however is that the hard constraints are not on the label — they are in the assumptions underneath. Always-on connectivity. Abundant clean records. Cheap power. One official language. A workforce that lives in the cloud.
Change those assumptions and the imported product does not bend; it breaks. A dashboard that needs a steady connection is useless to a district officer on an intermittent one. A targeting model that assumes a clean civil register cannot run where the register is half-empty. The constraint was never the country's backwardness — it was the product's narrow definition of normal.
Leapfrog, don't translate
The better move has a name and a long track record on this continent. Leapfrogging — skipping a technology generation rather than importing it wholesale — is how Africa reached mobile phones without first wiring landlines, and mobile money without first building bank branches. The same logic applies to decision products. A government does not need to repeat the West's thirty-year march from on-premise warehouses to business intelligence to the cloud, inheriting every assumption baked in along the way. It can design for the constraints it actually has, and skip the rest.
This idea is older than the smartphone. The economist E.F. Schumacher called it appropriate technology in 1973: the right tool is the one that fits the context, not the most advanced one money can buy. A decision product designed for intermittent connectivity and thin data is not a watered-down version of the real thing. It is the real thing, built for the place it serves — and it is usually more robust, because it assumes less.
The constraint is not a deficit to apologise for — it is the design brief.
A test you can run before you buy
The lever is to judge any analytics method against the constraint, not against the demo. Before a ministry commissions or buys a decision product, put it through four questions:
- Does it degrade gracefully? When the line drops, does it fail, or keep working and sync later? Offline-first is not a feature; it is the baseline.
- Does it run on the data you have? Not the clean register you wish you held — the proxy and partial data you actually hold. Methods that fuse mobile, satellite, and administrative signals beat methods that wait for perfect inputs.
- Is it multilingual where it matters? Not just translated labels, but a system that handles the languages your stakeholders and your data actually use.
- Does it run where the decision lives? In the district, the clinic, the border post — not only in a capital-city control room no one downstream can reach.
A method that survives all four was designed for your reality. One that fails any of them was designed for someone else's, and no amount of localisation fixes it after the fact.
The proof is already on the continent
That constraint-fit design delivers is not a hope; it is measured. Mobile money is the clearest case. By skipping bank branches entirely and running on the phones people already owned, M-Pesa did not just digitise payments — it changed livelihoods at national scale.
By skipping bank branches and running on phones people already owned, M-Pesa moved an estimated 194,000 Kenyan households — about 2% of the population — out of extreme poverty, with the largest gains for female-headed homes. The leapfrog beat the imported model.
Source: Suri & Jack, Science (2016) — summary via Georgetown University
The lesson for the state is direct. Togo's Novissi programme — the subject of our last Insight — targeted emergency cash by fusing mobile and satellite data precisely because no clean income register existed to drop a Western targeting model into. The constraint forced a better design. Both cases share a spine: insight built for the conditions on the ground, wired to a decision, producing an outcome you can measure.
None of this requires waiting for infrastructure to catch up. The states that win with data will be the ones that treat their constraints as a design brief, not an apology.
Tensō builds decision products for EMEA realities — intermittent connectivity, multilingual stakeholders, nascent data infrastructure — not Western tools with a locale swap. If your ministry or institution is weighing a data investment, that is the conversation we have.