According to recent industry research, 95% of retailers are now using AI in some form, yet only 5% report clear, scalable ROI. That gap is real – and in most cases, it comes down to where retailers are actually putting their AI. What I see most often is organisations experimenting in pockets, without rethinking the foundations that AI needs to actually work. The result is activity without acceleration.
Most organisations haven’t yet figured out how to make AI work commercially, not just operationally. And the AI we’re using today is likely the least impressive version we’ll ever see. The ceiling isn’t technology, it’s whether you’ve built something capable of absorbing what comes next.
Adoption is not the same as integration
Willingness isn’t the issue. Retailers are investing, experimenting, and deploying at pace. The problem is that most of this activity is happening on top of foundations that were never designed to support AI. In many retail organisations, customer, product, and transaction data still sit in silos.
An AI model might optimise product recommendations without visibility into inventory constraints, or personalise marketing without understanding in-store behaviour. These aren’t failures of the models – they’re predictable outcomes of incomplete context. The more retailers automate on fragmented foundations, the more that fragmentation shows.
Take a churning customer. It’s easy to point an ad platform at a re-engagement signal, but that logic lives within its own channel and its own data model. The customer who stopped buying from you will see ads, emails, in-store messaging – all generated from different systems with no shared understanding of who they are.
A lack of organisational alignment compounds this. When customer experience functions aren’t connected to the data and decisions driving AI, the insights generated are far less likely to be used effectively. The result is AI that optimises within channels rather than acting across the business. That’s a data foundation problem, not a technology one.
The barrier isn’t technology – it’s people and culture
Four of the five biggest barriers to scaling AI in retail are organisational, not technical. Skills gaps (58%), internal resistance (57%), legal and privacy concerns (54%), and lack of trust in AI decisions (53%) all rank above integration challenges. Retail is a deeply human profession – and that context matters.
Designers, buyers, and marketers have spent years developing a craft – and now parts of that craft are being automated. The hesitation isn’t irrational. When AI gets things wrong in retail, it’s not abstract – it’s a customer who gets the wrong message, at the wrong time, about the wrong product.
Pricing errors, poor stock allocation, and tone-deaf loyalty communications all carry real commercial consequences. The organisations making progress treat this as a culture challenge – they make ownership explicit, give teams guardrails and human checkpoints, and reframe what AI is for: not replacing judgment, but sharpening it. That reframe is what turns hesitation into confident, compounding adoption. Without it, AI remains a pilot rather than a platform.
What the most successful retailers are doing differently
The retailers seeing scalable ROI share a set of structural traits – operational choices that shape how AI fits into the business. They treat unified data as a commercial priority, not a technical project. Advanced retailers draw on nearly twice as many data sources as their mid stage peers. That breadth determines how far AI can extend beyond isolated tasks into cross functional decisions – connecting customer behaviour with product attributes, inventory, margin, and loyalty signals simultaneously.
They embed AI into commercial decision making, not just campaign execution. In mid stage organisations, AI supports defined workflows; optimising email timing, automating segmentation. In more advanced organisations, AI informs planning: budget allocation, inventory trade-offs, loyalty investment. The difference isn’t the presence of AI, it’s the level at which it operates.
The most commercially impactful use cases also aren’t necessarily the most visible. Predicting churn, identifying high value customers, and triggering timely, personalised loyalty communications tend to deliver more consistent returns than headline grabbing innovations. The deeper the context, the more precisely AI can act.
Finally, mature organisations maintain a genuine ‘test and learn’ culture. Experiments have clear success criteria, outcomes are tracked, and learnings compound over time. That balance, human judgment plus AI signal, is one of the clearest markers of maturity I see.
The window is closing
AI maturity compounds. Retailers investing in the right foundations today will improve their models, accumulate proprietary data, and strengthen customer loyalty loops simultaneously. Those who delay aren’t standing still – they’re drifting further behind.
Industry research suggests 71% of retailers expect AI to be meaningfully deployed across marketing within two years, with 45% expecting measurable returns in the same window. Beyond that point, AI moves from competitive advantage to baseline expectation, and the gap becomes structural. The retailers that will lead in 2030 aren’t necessarily the ones with the biggest AI budgets today. They’re the ones treating AI maturity as a discipline: building the data foundations, skills, and operating models that make the next wave of capability usable at scale.
The highest return AI investment is rarely an AI feature. It’s the foundational work underneath – the data foundation, the governance, and culture shift – that makes AI commercially valuable rather than just operationally present. None of this requires panic, but it does require a decision.
The 95% versus 5% gap is not an indictment of AI. It’s a map. And the window to act on it is still open, but it’s closing faster than most expect.
