The Money Sitting on a Retailer’s Shelves

Walk the back of almost any retail operation in Lagos and you will find two problems standing next to each other. Stock that has not moved in months, and a list of lines that ran out last week. Both are cash. One is trapped on the shelf. The other walked out of the door to a competitor. I spent two and a half years at Jumia, in growth marketing and then advertising. You could not work there without seeing this pattern. Most AI conversations in retail skip straight past it to something that looks better in a presentation. ## Stock is where the cash is With exchange rates moving and suppliers who deliver when they deliver, a retailer that orders on instinct pays twice. Once for what sits, and again for what is missing. Forecasting demand properly is the least glamorous AI project in retail and usually the most valuable. It does not need exotic data. Your own sales history by branch, salary week, the school calendar, festive seasons and, for some categories, the weather. A forecast that is right more often than the buyer's gut releases cash. When borrowing is expensive, released cash is margin. ## The last mile is a data problem In many Nigerian cities an address is a description. "Opposite the filling station, the blue gate." Couriers learn these by heart, and the knowledge leaves when they do. Recording where deliveries actually ended up, and how long each route really took at each time of day, gives you something no static map has. Fewer failed drops and less fuel. In Lagos a five-kilometre trip can take ten minutes or two hours. Knowing which is worth money on every run. ## Sell where your customers already are Your customers are on WhatsApp. Plenty of them will never download a 50MB app. Conversational AI on WhatsApp can take orders, answer questions in the English and Pidgin people actually use, and handle the routine support that currently needs a call centre. Built badly it becomes another bot people shout at. Built well, it passes the conversations that need a person to a person. ## Where I would start Get the transaction data in order first. A model is only as good as the sales records underneath it, and in many businesses those records are still partly on paper. Then pick the problem costing you most this quarter, whether that is stock-outs or failed deliveries, and solve that one properly. Train the people you already have. Your branch managers know things about demand that no dataset holds. Give them tools that let them act on it faster. ## The test Before you start, write down the number this is meant to move, such as cash tied up in stock or cost per delivery, and the date you will check it. If it has not moved by then, stop and find out why before spending more. --- ## Working Together If this is close to where your business is right now, I would like to hear about it. I work with a small number of clients at a time and I read every message myself. Email [me@martins.com.ng](mailto:me@martins.com.ng), read [how I work](/work-with-me), or connect with me on [LinkedIn](https://www.linkedin.com/in/stanmartins).