Where FMCG and Retail Teams Should Start, Using Data Already in the Sales System

Most AI advice for this sector assumes clean data and an analytics team. Neither is typical in African FMCG and retail, and neither is required to start. Three that work with what is already in the sales system. ## 1. Forecast one category, not the business Pick a single category with reasonable sales history. Build a forecast using your own sell out data plus the local signals that actually move demand: salary week, season, school calendar, rainfall if it is relevant to the category. Compare it against whatever you currently use for the same period. If your existing method is a rolling average or a planner's judgement, the comparison will be informative either way. The measure is not model accuracy in isolation. It is stock in the right place and fewer out of stocks on the lines that matter. ## 2. Work out profit by outlet, not volume Take your top hundred outlets. For each one, put actual cost to serve against actual margin generated. Delivery frequency, drop size, distance, returns, credit terms, trade support given. Most businesses doing this for the first time find a meaningful share of outlets are being served at a loss, and that several of them are large accounts everyone assumed were valuable because volume is high. You do not need machine learning for the first pass. You need the arithmetic done honestly once. ## 3. Attribute one promotion properly Take a recent promotion. Compare participating outlets against similar non participating ones over the same period. Separate genuine incremental volume from sales that would have happened anyway at full price. Do this three or four times and you will start to see which mechanics work in which channels. That is worth more than any single campaign result. ## Before any of them Write down which AI tools are approved and what commercial data may go into them. Distributor files, pricing, competitor data and customer records should not be in a free account, and right now somebody in your team probably has them there. That is BROAI, and it comes from capable people with deadlines and no sanctioned option rather than from carelessness. This is the Scan and Score phase of SWITCH, the framework I use with clients. Establish what is actually true before building on it. Pick one of the three. Give it an owner and a number. Look at it again in thirty days and either scale it or stop. If you want this run with your own categories, outlets and constraints rather than as a generic session, the [speaking and training page](/speaking) sets out the formats, or reach me at me@martins.com.ng or through the [work with me page](/work-with-me).