A Report That Names the Gap Between Talk and Scale
On June 24, Boston Consulting Group and The Consumer Goods Forum published a study with a deliberately blunt title: "AI in CPG and Retail: How Winners Are Pulling Ahead." The headline finding is uncomfortable for an industry that has spent two years announcing AI partnerships. Of the 39 senior executives surveyed across consumer packaged goods and retail, roughly 75 percent of CPG respondents remain in pilot or exploration mode, and only 18 percent say they are scaling impact. Retail looks like a two-speed world, with 45 percent scaling and another 40 percent barely started.
We read this as a correction to the narrative that AI adoption is a steady, rising tide. It is not. The report describes a widening separation between a small group of operators turning models into measurable outcomes and a much larger group still running disconnected experiments. For CIOs and CTOs in this sector, the survey is less a benchmark to celebrate and more a mirror. The question it forces is not whether you have AI projects, because almost everyone does, but whether any of them have crossed from demo to durable, revenue-affecting capability.
The Quote That Ends the Experiment Era
Wai-Chan Chan, managing director at the Consumer Goods Forum, put the stakes in plain language. "For CEOs, the honeymoon phase with AI is officially over. It cannot be a tech experiment anymore; it is a direct lever for your bottom line," he said. That framing matters because it shifts accountability out of the innovation lab and onto the income statement. When AI is a curiosity, a failed pilot is a learning. When it is a bottom-line lever, a failed pilot is a missed quarter.
Chan went further, warning that the next stretch will sort the field. The leaders, he argued, are the ones actually scaling these systems rather than the ones still talking about them. We think that distinction will harden fast. The companies that treat 2026 as another year of proofs of concept are quietly conceding ground to rivals who have already wired AI into pricing, assortment and demand planning. The honeymoon language is rhetorical, but the consequence is real: budgets that once tolerated open-ended exploration now expect a return that shows up in margin.
Why Measurement Is the Real Divider
The most telling statistic in the study is not about technology at all. More than half of the companies surveyed do not formally measure the return on their AI investments. That single data point explains much of the pilot trap. If you cannot quantify what a model delivered, you cannot defend scaling it, you cannot kill what is not working, and you cannot redirect spend toward what is. Measurement, not model choice, is what separates the operators who compound gains from those who accumulate stranded experiments.
This is a governance failure as much as a technical one. We have argued before that AI programs collapse under their own optionality when every team runs its own pilot with no shared definition of success. The BCG and CGF data supports that view. The leaders described in the report do not simply have more models in production; they have the discipline to attach each deployment to a number a CFO recognizes. For technology leaders, the practical takeaway is to install ROI measurement as a precondition for funding, not an afterthought once a tool is already live.
The Strategy-Execution Gap on Core Processes
The report exposes a striking mismatch between where executives say AI matters and where they have actually scaled it. Nearly half of CPG executives named idea to market as their most strategic process for AI, yet only 11 percent have scaled AI there. On the retail side, 46 percent named offer to assortment as the top priority, but only 34 percent have scaled it significantly. In both cases, the processes leaders care about most are the ones they have been slowest to automate at scale.
This gap is where competitive advantage will be won or lost. It is comparatively easy to bolt a chatbot onto customer service or generate marketing copy. It is much harder, and far more valuable, to rebuild the core commercial engine, how a product is conceived and brought to market, or how a retailer decides what to stock and at what price. The companies that crack those processes change their cost structure and their growth rate at the same time. The ones that keep their AI confined to the periphery will find that their pilots, however numerous, never touched the parts of the business that decide who wins.
The Prize Is Measured in Basis Points
BCG attaches hard numbers to the upside. The firm estimates that getting AI right could be worth 220 to 350 basis points of EBIT for CPG companies and 180 to 360 basis points for retailers. In a sector where operating margins are notoriously thin and competition for shelf space and shopper attention is relentless, several points of EBIT is the difference between leading a category and defending one. That is why the report frames AI as a margin lever rather than a moonshot.
Nicolas De Bellefonds, BCG's global AI lead and a coauthor, was careful to dispel the idea that the winners simply had better starting hands. "The reason that some CPG companies and retailers are pulling ahead isn't necessarily that they started from a stronger position," he said, pointing instead to higher ambition, focus on a few priorities, and disciplined management of data and technology. We find that reassuring and demanding in equal measure. Reassuring, because it means the advantage is built rather than inherited. Demanding, because it removes the excuse that a company is simply behind on data or talent. The differentiator, on this evidence, is choice and discipline, and both are available to anyone willing to stop experimenting and start scaling.
What This Means for Technology Leaders
For CIOs and CTOs in CPG and retail, the study reads like a checklist disguised as research. Pick a small number of core commercial processes, idea to market or offer to assortment among them, and concentrate AI there rather than spreading it thin. Insist on ROI measurement before funding, because the absence of it is the clearest signal of a laggard. And treat ambition as a strategic input, since the leaders set higher targets and mobilized their organizations to hit them.
The broader signal is that the patient capital phase of enterprise AI is ending in this sector. Boards that funded exploration on faith now want margin in return, and a survey-backed estimate of up to 360 basis points of EBIT gives them a number to demand. We expect the next two years to produce a visible split between companies that industrialized AI in their core operations and those that have an impressive list of pilots and little to show on the income statement. The honeymoon, as Chan said, is over. The marriage, with all its accountability, has begun.
