Journal

What Togo Teaches Me About Data That Serves a Purpose

One morning this week, I read that Togo was serving as an international benchmark. For a method of handling data. The World Bank's World Development Report 2026 cites the country as an example for Novissi, the programme rolled out during COVID-19. I read the paragraph twice. Then I looked for what an entrepreneur in Lomé could take from it by Monday morning.

Novissi did not win thanks to the technology

The programme had to identify vulnerable households, fast, without sending surveyors from door to door. The answer was not a sophisticated model. It was a cross-reference: mobile phone metadata on one side, satellite imagery on the other. Two sources that already existed. What was new was looking at them together.

The result, as the World Bank reports it: "a 421% reduction in exclusion errors compared with traditional geographic targeting". I am using the report's own words, because the figure is its own: what it describes is a gain in targeting. "Exclusion error" is the technical name for something very concrete. Someone who was entitled to the aid and did not receive it, because their street fell on the wrong side of a line drawn on a map.

Here is what I take from it. The performance did not come from the tool. It came from a question put to data that was lying dormant.

You already produce data, you just don't read it

I see that same data lying dormant with almost everyone I work with. A seamstress in the neighbourhood takes her orders on WhatsApp: months of conversations, with the dates, the styles requested, the deadlines met or missed. A mechanic writes his jobs down in a spiral notebook. A trader at the Grand Marché takes payment in mobile money and holds, without thinking about it, a time-stamped history of every sale. A practice has an appointment book that says exactly which month is overflowing.

Nobody looks. Not out of laziness: looking means stopping, and someone who is running does not stop.

The method, in one week

Take stock before you buy anything at all. Take a sheet of paper. Write down every trace your business already leaves: mobile money statements, WhatsApp thread, cash book, invoices, diary, follow-up messages. Do not judge their usefulness yet. List them.

Choose a single source. Not three. The one that is the most complete and the least painful to read. For the seamstress, it is WhatsApp. For the trader, it is the mobile money statement.

Frame a question that forces a decision. Not "how are my sales doing": too vague, it decides nothing. Rather: which three days of the week do I take in the least money? Which style costs me the most time compared with what it brings in? Which customer has to be chased twice before every payment?

Draw one decision from it, just one, and write down the date. Close on Tuesday afternoons. Turn down a type of order. Ask a particular customer for a deposit. One decision, written, dated.

Measure after a month. Even roughly. If the decision produced something, start again with the next source. If it produced nothing, the question was a bad one: change the question, not the tool.

This protocol fits on one sheet of paper. It is the exact order Novissi followed, on another scale. Existing data, precise question, decision, measurement.

What really gets in the way

The report tells no fairy tale. It points out that AI can become an engine of growth for developing countries, provided they invest quickly in infrastructure, skills and institutions. And it names the obstacles: the cost and stability of access to electricity, uneven rural connectivity. In sub-Saharan Africa, about one third of rural schools have no stable electricity, and more than two thirds have no reliable internet.

I know this constraint from the inside. This year, I lost my work machine, then the one I had bought to replace it. A work tool, for me, is not an accessory: it is the whole workshop. I am not looking for someone to blame, I am looking for a process that holds. That is why my way of working rests on very little: light files, work that restarts where it stopped, nothing that depends on a continuous connection. Design your processes for the electricity and the network you have, not for the ones you would like to have. A paper notebook kept every day is worth more than software you do not open every day.

The real signal is the laboratory

Togo is also developing a data analysis laboratory intended to evaluate public policy. It will get less attention than Novissi. In my eyes, it is at least as important. To evaluate is to accept measuring what did not work. The report judges that capacity essential for a responsible deployment of AI in public action. It is also what the small organisations I meet lack most. Launching, almost everyone knows how to do. Measuring, almost nobody.

I am proud to read my country's name in that report. Proud enough to want to make something of it at my own scale. Data I already have, a question, a decision, a measurement.

The data you are missing, you already have.