In the first quarter of 2026, 10.1% of people aged 15 to 64 in Togo said they had used a generative artificial intelligence tool. One adult in ten. In the same country, mobile penetration is above 70% of the population. I have been setting these two figures side by side for weeks, knowing that one does not subtract from the other: the first counts declared use among 15-64 year olds, the second counts mobile lines measured against the whole population. But the order of magnitude speaks for itself. The handset circulates widely. The use does not.
What is really moving forward
The pipes are being laid, and that has to be acknowledged. 4G is rolling out in the main cities. The country is connected to the ACE submarine cable. Yesterday the conversation was about signal; today it is about use. That is no small shift.
The obstacles that remain are cited regularly, and I do not play them down: the cost of equipment, still-irregular access to electricity, a shortfall in advanced digital skills. I paid for the first two out of my own pocket this year, so I know what they weigh. But those three factors do not, on their own, explain the distance between a widely available mobile network and a practice that is still rare. There is a space left in the middle. That space is not bought, it is learned.
Having the tool means nothing
I see a lot of people open an AI tool, type "hello", look at the screen, then close it. They have access. They have no task. That is the exact difference between owning a camera and knowing which photograph you came to take.
A seamstress in my neighbourhood takes her orders on WhatsApp. Dozens of messages a day, measurements buried in voice notes, appointments confirmed several times over. She has a smartphone, she has network, she uses no AI tool at all. Yet her chore is visible to the naked eye: turning messy messages into a clean order sheet. The day someone names that task for her, the tool becomes useful. Before that, it stays a gadget.
Monday morning, one task and one only
Here is what I have every person I work with do. It fits on a single sheet of paper.
List the week's chores. Not the projects, the chores. The things you redo identically: the same quote, the same follow-up message, the same meeting report, the same post caption, the same filing of receipts. Seven to ten lines is enough.
Hand over only one. The most repetitive, the least risky, the one nobody will fight to keep. One only, for two weeks. If you hand over five at once, you will never know which one worked.
Measure one thing. Time. How many minutes before, how many after. It is the only indicator that convinces a workshop owner, a roadside mechanic or a professional firm.
Write an instruction, do not ask a question
The most frequent mistake I come across: treating the tool like a search engine. "How do I write a quote?" produces nothing you can use.
An instruction is something else. It says who is speaking, to whom, in what format, under what constraints. "Here are my site notes. Write a quote in French, six lines maximum, amounts in CFA francs, sober tone, no promise on deadlines." At that point, what the tool hands back to me is usable on the first attempt, or close to it.
And I always give it material. Your old quotes, your notes, your way of saying things. A tool fed with your context works far better than a tool questioned with nothing behind it.
Two rules I never negotiate
Nothing goes out unread. Nothing the tool produces reaches a client without my having read it line by line. AI produces fast and gets things wrong with confidence. The misspelled name, the amount that slipped, the sentence that is too sure of itself: I am the one who signs, so I am the one who checks.
Nothing sensitive goes in. No identity documents, no bank details, no confidential contract, no full client file. I rephrase, I anonymise, I work on an extract. That discipline costs two minutes and spares you years of trouble.
The tool has no value in itself
This is the point I repeat most often. An AI tool is worth nothing as long as it floats. It takes on value at the precise moment it replaces a chore that has been identified, timed, and that you are glad to be rid of. Otherwise it is one more subscription and a three-day curiosity.
That is where process optimisation becomes concrete. A chore named. An output format decided. A systematic review. A figure before, a figure after. Nothing spectacular, and that is exactly why it lasts.
Then the figure will move. Not because there are more mobile lines, they already circulate. Because people will have named their chores, one by one.
The gap does not close with network, it closes with a task.