Every retail executive I speak to is being told they need an AI strategy. The pressure comes from boards, from competitors’ press releases, from vendors who’ve rebranded their existing products with “AI-powered” badges, and from the general sense that falling behind on AI means falling behind on everything.
The reality, as usual, is more nuanced. After working with several retailers on AI adoption over the past two years, I’ve developed a reasonably clear picture of where AI is creating genuine commercial value in retail and where it’s still mostly theatre.
Where AI is actually delivering value
Demand forecasting and inventory optimisation is probably the most mature and commercially impactful application of AI in retail right now. The algorithms aren’t new — statistical forecasting has been around for decades — but the combination of better data infrastructure, more granular data, and modern ML models has meaningfully improved accuracy. For retailers with significant inventory carrying costs, even small improvements in forecast accuracy translate directly to bottom-line impact.
Search and product discovery is another area where AI is delivering measurable results. Natural language search, visual search, and personalised ranking are all producing meaningful improvements in conversion rates. The key is having clean product data — the algorithms are only as good as the catalogue they’re working with.
Customer service automation through conversational AI has improved dramatically. Modern language models can handle a surprisingly wide range of customer enquiries — order tracking, returns processing, product questions — with quality that’s getting close to acceptable for many use cases. The economics are compelling: even partial automation of customer service volumes can free up significant resources.
Content generation for product descriptions, marketing copy, and merchandising content is an area where AI is reducing cost and increasing speed. It’s not replacing editorial judgment, but it’s handling the high-volume, lower-complexity content tasks that used to consume significant time.
Where it’s mostly theatre
AI-powered personalisation remains, for most retailers, dramatically oversold. The theoretical case is compelling: show every customer a uniquely tailored experience. In practice, most retailers don’t have the data foundation, the content pipeline, or the testing infrastructure to make personalisation work at scale. I see a lot of investment in personalisation technology that delivers marginal improvements while consuming significant engineering and marketing resources.
Autonomous pricing is another area where the ambition tends to outrun the execution. Dynamic pricing requires not just good algorithms but robust competitive data, clear margin guardrails, and organisational alignment on pricing strategy. Most retailers I work with aren’t ready for fully autonomous pricing — and the risks of getting it wrong are significant.
AI for AI’s sake is the most common pattern I see. This typically looks like a “Centre of Excellence” or an “AI Lab” that exists somewhat independently from the core business. These teams often build impressive proofs of concept that never make it to production because they weren’t anchored to a specific business problem with clear economics.
What separates successful AI adoption from theatre
The retailers I’ve seen succeed with AI share a few common characteristics.
They start with the business problem, not the technology. They don’t ask “where can we use AI?” — they ask “what are our most expensive or most impactful operational challenges?” and then evaluate whether AI is the right tool.
They invest in data infrastructure first. Clean, accessible, well-governed data is the prerequisite for every AI application. The retailers who are getting the most value from AI are the ones who invested in their data platform before they invested in AI models.
They measure ruthlessly. Every AI initiative has a clear commercial metric and a defined period for proving its value. If it doesn’t demonstrate impact within that period, it’s deprioritised.
They treat AI as an operational capability, not a project. Successful AI adoption doesn’t end with model deployment — it requires ongoing monitoring, retraining, and integration with operational processes.
The pragmatic approach
If you’re a retail leader thinking about AI adoption, my advice is to resist the pressure to have a comprehensive AI strategy. Instead, identify two or three specific operational problems where better prediction, automation, or content generation could create measurable value. Invest in the data infrastructure those use cases require. Build or buy the simplest solution that works. Measure the results. Then decide whether to expand.
The retailers who will benefit most from AI in the next few years aren’t necessarily the ones making the biggest investments — they’re the ones making the most disciplined investments.