Theme 1: Grocery's AI Gap Is an Operating-Model Gap
No one at Groceryshop questioned whether AI works. Access to models like ChatGPT and Claude is now table stakes. Scaling them is the problem. Only 7% of firms have scaled AI across the enterprise, and of every 100 AI ideas, only about 16 make it through security, IT and business reviews to testing or implementation. The most common reason projects stall is that no one can prove the return: 75% of retailers and CPGs call AI strategically critical, but only 16.5% can quantify its ROI.
The fix the panels kept coming back to was organizational, not technical. That means one accountable owner with budget authority instead of a committee, and governance written so the people using it can understand it. It also means a culture that treats failures as lessons instead of repeating them. As Mars Snacking put it: "Speed without governance is risk. Governance without speed is irrelevant."
Five pillars: Incisiv's research points to five things that close the operating-model gap: champions below the executive level, a culture of learning, single ownership, an integrated business–IT relationship, and evidence that proves value.
From customer experience to margin: CEO attention has shifted from AI for customer experience toward AI for operational efficiency, which protects thin grocery margins and funds the next round of innovation.
Three horizons: Church & Dwight plans AI across everyday productivity, workflow redesign and transformational bets, and trains employees by persona (analyst, researcher, creative, project manager and executive).
Hub and spoke: an AI council or center of excellence sets shared priorities and guardrails, while local champions and functions handle execution, so governance doesn't become a bottleneck.
Good enough to ship: Rohlik Group treats an agent as ready for production once it reaches 60–70% of human accuracy, as long as feedback loops keep it learning after launch.
Build trust before autonomy: Fruitist moved its harvest models from optional tools to the primary driver of decisions in seven months by following a "build trust, train, co-own, prove" sequence.
- Put AI accountability with the P&L owner, not the technology team. Technology provides the tools; the business owns the outcome.
- Before funding any AI investment, answer three questions: what will be true if it succeeds, how will it show up in a metric or the P&L, and who is responsible for delivering it?
- Redesign the process with the people who run it. Layering AI onto an old workflow automates the old workflow.

