The most expensive line on a CPG P&L is also the one nobody can explain. CPG brands spend 15 to 25 percent of revenue on trade promotion. A Tellius analysis published this January, citing McKinsey, puts it bluntly: 72 percent of those promotions in the US lose money. Read that again, because it should stop a CMO cold. We are talking about a quarter of the top line, and three quarters of it sinking.
That is not a measurement curiosity. That is the single best argument for agentic AI in retail, and almost no one is making it.
We are pointing AI at the wrong workload
Walk into any consumer goods transformation roadmap right now and count the AI initiatives. Personalization engines. Next-best-offer models. A conversational layer on the e-commerce site. Another pilot for content generation. All of it defensible. None of it touching the line that actually moves enterprise margin.
I sit in the forward-deployed seat inside enterprise retail and CPG, and the pattern is consistent. Teams deploy AI where it demos well, not where the money leaks. Personalization is photogenic. It produces a tidy lift chart for the steering committee. Trade promotion is the opposite. It is messy, political, owned by three functions that disagree, and buried in a spreadsheet swamp that no one wants to open in a board review. So it gets left alone.
That instinct is exactly backwards. The criteria that justify an autonomous agent over a human-run dashboard are simple: enough data to learn from, enough friction to make manual work impossible, and enough margin at stake to fund the build. Trade promotion management hits all three harder than any other workload in the building. It is the most underused AI workload in retail precisely because it looks unglamorous.
The blocker is incrementality, not intelligence
Here is what stops most TPM efforts before they start, and it is not the model. It is that the organization cannot define a defensible baseline. Without a clean read on incrementality, you cannot separate a genuinely profitable promotion from a sugar high that simply pulled forward demand you would have captured anyway. So teams keep funding programs that feel busy and look active, and the true ROI stays a mystery until the quarter closes and the answer no longer matters.
Agents are well suited to this exact shape of problem. The work is repetitive, high volume, and rule-bound at the edges but judgment-heavy in the middle. An agentic layer can run continuous post-event analysis across every promotion, compute true incrementality against a modeled baseline, flag the losers inside the cycle rather than three weeks later, and recommend reallocation before the next planning window locks. The Tellius piece notes that organizations applying measurement-first AI typically see a 10 to 15 percent improvement in trade ROI simply by identifying and killing the underperformers. On a line worth a fifth of revenue, that is not a feature. That is a margin event.
The sequencing matters. You do not start with the autonomous agent. You start by building the measurement layer the agent reasons over, because an agent pointed at a baseline you cannot defend just automates a bad decision faster. Get incrementality right, instrument the data, then let the agent run the loop the humans never had time to run.
The CODN math is brutal and specific
Every executive I work with eventually asks the same question: what does it cost to wait one more cycle? The Cost of Doing Nothing on trade promotion is not abstract. It is quantifiable to the dollar, and it is the most quantifiable CODN in the entire AI portfolio.
If trade is 20 percent of a 5 billion dollar revenue line, that is a billion dollars in annual spend. If 72 percent of it is unprofitable, you are watching a meaningful fraction of that billion go to programs that destroy value. Recovering even a slice of it dwarfs the combined return of every personalization pilot on the roadmap. The CODN of leaving trade promotion alone for four more quarters is not a missed feature. It is a billion-dollar line operated on faith while the AI budget funds the photogenic stuff next door.
So the question for any CDO, CMO, or VP of Transformation reading this is uncomfortable but clarifying. You have the largest, leakiest, most data-rich workload in your enterprise, and you are spending your AI capital somewhere safer. The competitor who fixes trade measurement first does not get a better chatbot. They get a structurally lower cost of doing business, and they get it before you notice.
Stop deploying AI where it photographs well. Deploy it where the money is leaking, and trade promotion is where the money is leaking.