Consumer goods executives have decided AI matters. It has not decided who is accountable for making it pay — and the 2026 data says that single unresolved question is the industry’s largest value leak.
There is a comfortable story circulating in consumer goods boardrooms: AI adoption is going well, pilots are multiplying, budgets are up, and the value will follow.
The 2026 evidence does not support the last clause. Across four independent datasets published this year — Deloitte’s survey of 200 retail and CPG executives, the Consumer Goods Forum’s research with BCG, EY’s study of 850+ senior executives across 24 markets, and NVIDIA’s annual industry survey — the same fracture appears. Conviction is universal. Investment is rising. Returns are largely unmeasured, and where measured, unimpressive. This is usually diagnosed as a technology problem, a data problem, or a talent problem. It is none of those. It is an ownership problem, and it is fixable faster than any of the others.
The numbers that should be on the agenda Conviction is not the constraint. Three-quarters of retail and CPG executives call AI a top strategic priority. Only 16.5% can quantify a return on it. Eighty-two percent plan to increase AI investment over the next twelve months. Neither is budget — yet. Roughly half of companies invest less than 0.5% of revenue in AI despite calling it a priority. Nine in ten retailers plan to raise AI budgets in 2026. The money is coming. The question is where it lands: most current spend flows into IT and data infrastructure rather than value-generating use cases. Scaling is where the industry stalls. Enterprise-wide deployment sits at 7–10% across both retail and CPG. The Consumer Goods Forum and BCG put it more starkly: 76% of CPG companies remain in the “explorer” phase, with only 18% actively scaling impact. Retail is materially ahead — 45% scaling — though 40% have barely begun. Best-in-class organizations report 2–5x ROI. More than half do not measure return at all. And the sector is splitting in two. Retail is converting AI into commercial outcomes: 38% report revenue growth impact, against 10% for CPG. CPG’s gains remain concentrated in productivity, cost and product development (27% vs 18% for retail). Manufacturers are extracting efficiency. Retailers are extracting growth. Given that 79% of executives already expect bargaining power to shift further toward retailers over the next two to three years, this is not a neutral asymmetry — it compounds an existing structural disadvantage. The operating model has not moved. Only 11% of consumer products executives say sales and marketing operate as a unified growth engine. Only 15% say commercial data is fully integrated and routinely used for cross-functional decisions. 71% agree structural disruption makes rapid transformation essential.
Fifty-four percent of AI strategy ownership sits with technology leaders rather than the P&L owners accountable for business results. Every other number in this article is downstream of that one. When a CTO or CIO owns AI strategy, the organization optimizes for what technology leaders are correctly measured on: platform reliability, data quality, architectural coherence, deployment coverage. Those are genuine prerequisites. They are not returns. Nobody in that structure is incentivized to kill an underperforming use case, because no one’s number improves when it dies. Nobody is forced to choose between two pilots, because neither is competing for the same P&L line. And nobody is accountable when 90%+ of deployments never leave pilot status, because “pilot completed” is a legitimate technology-function success metric. This is not a criticism of technology leadership. It is a description of a misallocated accountability, and it explains the pattern precisely: high conviction, broad piloting, infrastructure-weighted spend, and an inability to name the return. Adoption data confirms the containment. Wide AI adoption never exceeds 36% of the organization outside IT. The capability has been built adjacent to the business rather than inside it.
In our work with commercial organizations across Europe and the Middle East, four recurring patterns explain most of the value leakage. Each map to an ownership vacuum.
- The pilot that has no owner who loses. A use case is sponsored by a function that benefits from its existence rather than its outcome. It runs indefinitely in a state of “promising.” No one closes it because no one is penalized for its cost.
- Governance chosen before accountability. Retail is consolidating around centralized AI governance (43% of retailers). CPG has not converged — it splits roughly evenly across three governance models. The instinct is to resolve this by picking a model. That is the wrong sequence. Which governance model you choose matters far less than whether the person accountable for the result also controls the decision. Centralized governance with the wrong owner produces the same outcome as federated governance with the wrong owner.
- Infrastructure counted as progress. Data platform milestones are reported to the board as AI progress. They are enabling conditions. A migrated data lake with no commercial decision attached to it is a cost with an option attached, not a return.
- Strategy written after execution has already started. The clearest instance is agentic commerce: 40% of CPGs report no defined approach, while 50–60% are already piloting capabilities and prioritizing API readiness. Teams are building because the technology is available and the pressure is real. The strategy that should govern the build is being written afterwards, if at all.
1. Reassign AI strategy ownership to P&L holders — within one planning cycle. Not an AI council. Not a center of excellence with dotted-line influence. The executive who owns the revenue or margin line owns the AI agenda that affects it, with the technology function accountable for enablement and standards. This is a governance change, not a reorganization, and it can be made in a single quarter.
2. Force every use case onto a P&L line before funding. Each initiative names the line it moves, the baseline, the expected delta, and the date the delta is measurable. Use cases that cannot complete that sentence is research, and should be funded from a separate, explicitly capped research budget — not from the commercial AI envelope.
3. Rebalance the spend mix, not just the spend level. Since 82% are increasing budgets anyway, the decision that matters is allocation, not quantum. Set an explicit floor for value-generating use cases as a proportion of total AI spend, and report against it quarterly. Infrastructure will always argue persuasively for more; it needs a counterweight with equal standing.
4. Change the reporting metric from pilots launched to pilots scaled. With enterprise-wide deployment at 7–10%, “number of pilots” is a vanity metric that actively rewards the wrong behavior. Track the conversion rate from pilot to production, and the time it takes. Treat a killed pilot as a positive outcome, and say so publicly the first time it happens.
5. Close the commercial gap, not just the technology gap. The 11% figure on unified sales and marketing, and the 15% on integrated commercial data, are the ceiling on everything else. Integration here is a prerequisite for CPG closing the 38%-versus-10% revenue-impact gap with retail. AI applied to a fragmented commercial system automates fragmentation.
The window The Consumer Goods Forum and BCG estimate the AI value opportunity in the sector expands by a factor of 1.7 as capabilities mature toward autonomous models. That is the reward for getting the foundation right. It is also the penalty for not: the multiplier applies to whatever base a company has actually built. The pattern across every dataset published this year is consistent. The sector has resolved the question of whether AI matters. It has not resolved who is accountable for making it pay — and the companies that answer that question first will spend the next twelve months compounding, while the rest continue to pilot. The most valuable decision available to a consumer goods leadership team in 2026 costs nothing and requires no technology. It is deciding, explicitly and in writing, whose number moves when AI works.
