In my previous essays, I explored why generic AI shopping agents struggle with both product discovery and vice products in grocery retail. Highlighting that while individual products might seem unremarkable, collectively they reveal highly personal preferences that prove difficult to automate. This challenge reveals a significant opportunity for retailers themselves.
Why Product Discovery Matters
The discovery challenge in grocery is both real and persistent. Modern stores stock anywhere from a few thousand SKUs for hard discounters to 50,000+ for full-service grocers. According to Stray Partners' 12-year analysis, thousands of new products launch annually, yet most fail within their first year. Discovery isn't a one-time problem but an ongoing challenge.
What AI Can Do (and Can't)
AI could genuinely add value here. Machine learning excels at pattern recognition and can surface relevant products customers might not think to search for. Generative AI interprets search intent and natural language, providing a relevant response to "Christmas breakfast ingredients" rather than simply showing products with matching words in their title.
But generic AI agents face fundamental limitations. They lack purchase history, struggle to differentiate between thousands of similar products, and can't grasp the deeply personal nature of grocery decisions. The most sophisticated language model can identify what you're asking for, but it can't know what you actually like.
So the question is: who's best positioned to deliver great AI-powered shopping experiences?
The Retailer's Data Advantage
Grocery shopping data, captured through loyalty programs, offers extraordinary insight. Groceries represent significant household spending, interactions happen weekly or more, and every basket contains dozens of choices across hundreds of categories. Each shopping trip is a detailed preference survey completed with actual wallets.
This data reveals preferences at the individual product level and emerging trends. On a customer level it shows not just that you like "pasta" but preference for a specific brand and variation. It shows substitution patterns, captures seasonal shifts and life stage changes, and because this happens repeatedly, retailers can distinguish genuine preferences from experiments.
This creates a powerful feedback loop: better data enables better service, which drives more transactions, which generates richer data.
Targeted AI, Not Generic Chatbots
Retailers don't need to compete with state-of-the-art LLMs on general capabilities. That would be doomed to fail. Just look at the pace of improvements and eye-watering investments required. The winning approach is targeted AI capabilities deployed where retailers have genuine advantage.
Think meal planning that knows your family's preference for convenience during the week but willingness to spend more effort over the weekend. Or "what can I make with what I have" features that suggest recipes based on actual purchase history, knowing you likely already had fresh pasta this week and have certain staples in your pantry.
This isn't about having the smartest general AI. It's about combining good-enough AI with irreplaceable data. A model that knows you always substitute oat milk for dairy but never buy soy products doesn't need GPT-5-level capabilities. It just needs to know the individual customer's preferences.
The approach should blend generative AI for natural interaction with analytical AI for pattern recognition. Generative AI handles conversational interfaces and creative tasks like meal suggestions. Analytical AI mines purchase history and predicts needs. Together, they deliver experiences generic agents simply can't match.
Why This Matters
Grocery operates on razor-thin margins, typically 1-4% net profit, and faces unique complexities around perishability, supply chain, and availability. This makes unfocused AI investment dangerous.
But targeted AI capabilities offer real returns at manageable cost. Building internal AI capabilities, supplemented with state-of-the-art LLMs via API, might require a team of 5-10 people and modest compute resources. In an industry where every percentage point matters, improvements don't need to be dramatic to make an impact. Increasing basket size by 2-3% or reducing shrink through better inventory predictions pays for itself quickly.
This requires investment in technical capability and organizational change. But for retailers willing to make focused investments, the reward is substantial: better customer experiences and a competitive advantage that strengthens with every transaction.
The retailers who focus their AI efforts where their data advantage is strongest, rather than trying to build everything, will create shopping experiences generic agents can't replicate. They'll turn the complexity that challenges AI agents into their greatest strength.
The race isn't about having the best AI. It's about having the best combination of targeted AI and proprietary data. For grocery retailers, that's not just a race they can win—it's one where they start with a decisive advantage.