It is a Friday night in Lagos. Your transfer has failed for the third time. Somewhere in a server farm in Ikeja, the bank’s API has gone quiet. And the only thing your fintech app has to say about it is: “Something went wrong. Please try again later.”
I want you to sit with that sentence for a moment. Read it the way the user reads it at 11 pm with rent due on Monday. Read it the way the user reads it while standing outside a pharmacy, trying to pay. Read it the way your mother would read it, the one who finally agreed to try the app after months of you swearing it was easy.
The sentence is grammatically correct. It is also, for lack of a better word, completely useless. It tells the user nothing about what actually happened, nothing about whether their money is safe, nothing about whose fault this is or what they should do next. Worst of all, it doesn’t sound like anyone wrote it. Because they are just automatic messages now.
That’s what I want to talk about. Over the last two years, while every product team I know was scrambling to plug AI into their stack, something quieter was happening in the background. The apps got smarter. They also stopped sounding like anyone you’d actually want to talk to.
I designed checkout flows at Yannis, a Nigerian e-commerce platform, for three years. We had a problem that looked technical on the surface and turned out not to be. About a quarter of our customers were abandoning their carts at the exact same point, not the payment screen, not the address page, but the moment the app tried to fetch live shipping rates over a slow mobile connection. They’d stare at a spinner for seven seconds, decide the thing was broken, and leave.
The fix was a single line of copy. Under the spinner, we added: “This may take a moment on slower networks. Your cart is safe.”
Cart abandonment dropped. Conversion went up. The truth of the matter is that the app didn’t get smarter; it only got kinder. And the only reason that sentence existed was because someone had sat down and thought: what does our user feel right now, and what do they need to hear?
That’s emotional design. It’s a discipline as old as the field itself. Don Norman wrote the foundational book on Emotional Design back in 2004, and for twenty years it has been the quiet, slightly unfashionable part of the craft that designers in mature markets always understood mattered most. It’s the work of asking, at every step of the product: what is the user feeling right now, and is the product meeting them there?
Here’s the thing nobody is saying clearly: AI cannot do this work for you. AI is very good at producing sentences. It is not good at detecting when a user is anxious, frustrated, or about to give up, and quietly reshaping the entire tone of the interface to meet them where they are. That is still a human design decision. And as more of what your app says gets generated in real time, your support responses, your transaction summaries, your error messages, it has become the most important human design decision left.
I spent a year working through exactly this question for my master’s at Glasgow Caledonian. The project was an interactive journaling app I called Jona, a deliberate attempt to design for emotional well-being as the explicit goal, not just the marketing line.
What I learned ran counter to almost everything my engineering training had told me about how to add value. The interventions users responded to most were the ones with the least going on. A blank screen with a single prompt. A confirmation screen that didn’t rush them to the next thing. A previous entry surfaced not as a data point but as something closer to a letter the user had once written to themselves.
The lesson, I think now, is that emotional design is mostly the discipline of restraint. The temptation, especially with AI in the loop, is to fill every screen with something useful. The intervention usually involves taking something away. To leave the user alone for a beat. To trust them.
There’s a specific cost to getting this wrong in African markets, and I don’t see it talked about enough. We are not the default user in any global AI dataset. The empty states, the error messages, the fallback flows, and the parts of the product where emotional design matters most are almost always written by models trained on the internet’s average tone. Which is the tone of someone who has never had a transaction fail at 11 pm with rent due on Monday. When that voice meets a Nigerian user at the worst moment of their day, the product doesn’t feel neutral; it feels foreign.
That’s the structural problem. The emotional tone of your app is no longer decided once at design time. It’s being decided sentence by sentence, in real time, by a model that doesn’t know your user. If you haven’t made the emotional intent of your product explicit, if you haven’t done the work of telling the model who this product is talking to and how it should make them feel, the model will default to a voice you never chose. And the default is not warm. The default is corporate American customer service circa 2018. Your user will notice.
There’s a version of the next five years in which African product teams use AI to build smarter, faster, and cheaper digital experiences for hundreds of millions of people. That’s the version everyone is selling. But there’s a quieter version, less photogenic, harder to put in a pitch deck, where the same teams remember that smarter doesn’t mean kinder. And that the products people actually love are still mostly the ones that seem to know exactly who they’re talking to.
It is a Friday night in Lagos. Your transfer has failed. Your app could tell you what happened in a sentence written by a human who actually imagined you. Or it could say: Something went wrong. Please try again later.
You’re going to remember which one you got.
