ARTICLE · Revenue Growth Management
Gen AI will not fix trade spend if your data cannot be trusted.
There is a sense that the real benefits of Gen AI adoption in FMCG companies are not flowing through to the business. I have run a number of discovery workshops with department heads in FMCG companies, and the pattern is the same.
Everyone has started leaning on LLMs to write emails, do analysis and prepare presentations. Yet the commercial team still works the same way. Trade spend continues to grow, and the same arguments keep happening in meetings. The blame game has not stopped.
Meanwhile the money sits exactly where it has always sat. In gross to net.
McKinsey published a piece in October 2024 covering more than 140 use cases across the CPG value chain. Two findings from it stuck with me. First, for a 10 billion dollar food and beverage business, they put the value at stake at 810 million to 1.6 billion dollars, which they translate to 7 to 13 percentage points of EBITDA margin. Second, they estimate that traditional AI's potential impact is 2.5 to 7 times higher than Gen AI's. The areas they flag as highest impact are not the glamorous ones. Consumer insights and demand shaping. Customer and channel management. Inside that, retail trade promotion optimization.
Most FMCG companies in the GCC do not have clean data to run through an LLM, and this needs to be the big focus. It is a real blocker. It is not about the talent, the budget or which model they choose.
Ask a simple question in most businesses here: what did we spend, by customer, by SKU and by mechanic, in the last twelve months, and what did each mechanic return? In a lot of cases the answer lives in three places. A distributor claim in one format, an internal accrual in another, and a sales team's memory of what was agreed verbally with a buyer.
I have sat on both sides of that. As a principal and as a COO building distribution. The claim arrives without the backup data, and it is submitted six weeks late, sometimes even longer. You cannot automate your way out of that. You have to fix the ways of working with your customers. We have to go back to reordering the procedures, lining up the master list and unifying the data structure before we start running the data through this intelligence layer.
I recommend three things.
Terms before tools. Before anyone talks about AI, the terms need to be simpler, with more precise deliverables and with the fixed and variable elements clearly separated. The conditions need to be simple enough to be verified. If the data is not collected against agreed, measurable deliverables, it is not a term but a hope.
Gross to net on one page. Every deduction between invoice price and net revenue needs to be named, owned and recorded in a single system. On one of my projects, I noticed that the client's spends were converted from how they are executed in the field into 'SAP language'. Any report run the way expenses are defined in the system then had to be translated back into how the business actually runs in the field. On top of that, a second system was used to approve certain spends in the market. The result was two sources of truth, both incomplete.
Once the first two elements are fixed, the analytics can begin. Promotion response by mechanic and channel can then be plotted. Cost to serve by customer. Which outlets in a route actually justify the visit frequency, and so forth.
The third step is the one people want to start with. It is also the one that fails first when the first two are missing.
I am not going to pretend Gen AI is a distraction. It is not.
McKinsey put the global Gen AI addition at 160 to 270 billion dollars a year on top of the traditional AI value. That is real, and the practical wins I see are in the unglamorous middle of the day. Reading distributor contracts and pulling out the terms. Summarising field reports from six markets. Drafting the first version of a customer business review so the KAM spends their time thinking rather than formatting.
That is the ground we cover in the intelli.accelerate Gen AI Sprint sessions.
Gen AI can also help you work out whether your 20 percent off mechanic in modern trade destroyed value. But only if your data is honest.
Try this.
Take one customer. Your biggest, ideally, or the one that argues most. Pull twelve months of trade spend for that customer only. Every line. Listing, display, rebate, price support, free goods, damages, whatever sits there. Now split it into three buckets. Spend you can tie to a specific mechanic with evidence. Spend you believe happened but cannot prove. Spend you cannot explain. Do not clean it up. Just size the three buckets.
If buckets two and three together are under 10 percent, you are in better shape than most and you are ready for a model. If they are over 30 percent, you have found your Gen AI project.
So here is my question for the CEOs and commercial directors reading this.
If someone asked you today for the return on your last twelve months of trade spend, by customer, could you produce it in a week?
Tom Thomas, Managing Director
Tom has spent more than 30 years leading consumer businesses across the Middle East, Africa and Asia. Today he helps leadership teams grow revenue, improve profitability and put AI to work.