The AI Gap in UK Retail: Most Investments Are Falling Short

Research commissioned by ecommerce agency Quickfire Digital and published in November 2025 put a number on something I had been observing from inside a mid-market UK retailer. The study found that 77% of UK ecommerce retailers admit their AI initiatives are falling short of expectations, with roughly £92 million of the sector's 2024 AI spend considered at risk of underperformance.

The £92m is a derived figure, not a surveyed number, based on applying e-commerce's approximate 25% share of retail to total UK retail AI investment, then multiplying by the 77% failure rate. The methodology is worth noting, because it means the headline is more indicative than precise. But the underlying finding — 201 retailers surveyed, 77% dissatisfied — is the part worth dwelling on.

More than three-quarters of retailers who set out to do something meaningful with AI are not satisfied with what they have achieved. That is not a technology failure. It is a deployment problem, and it tends to look different depending on where you sit in the market.

The Enterprise Advantage

Enterprise retailers (the ones with significant technology budgets, dedicated data science teams, and the IT infrastructure to support large-scale integrations) are operating in a different world from mid-market operators. Headline adoption rates across company sizes are closer together than you might expect, but the quality of what is being deployed varies considerably.

An enterprise retailer's AI footprint includes genuinely sophisticated personalisation systems, real-time demand forecasting, and AI that has been integrated with clean first-party data at scale. A mid-market retailer's AI footprint often includes some AI-generated content, a chatbot that handles basic queries, and an analytics tool with a machine learning badge on the box.

Not the same thing. And the Quickfire Digital research confirms the gap. When retailers were asked which AI applications had underperformed, the list was dominated by exactly the surface-level tools mid-market operators tend to deploy first: AI-powered chatbots (cited by 29% as underperforming), data analysis applications (27%), AI-driven marketing activities (23%), and content generation tools (20%).

Why Mid-Market Implementations Struggle

I can speak to some of this from the inside. The barriers are not primarily about budget, though cost matters. They are about data infrastructure and organisational capacity.

The AI applications that deliver the clearest value (personalised product recommendations, dynamic pricing, predictive demand forecasting, AI-powered customer service) all require clean, connected, accessible data. Not as a future-state aspiration but as a pre-condition for deployment. An AI personalisation system trained on fragmented, inconsistent customer data does not personalise effectively; it generalises badly, which can be worse than segment-level targeting because it creates an expectation of relevance that is not met.

Most mid-market retailers have the data, somewhere. It is in multiple systems, it is inconsistently attributed, it lacks the customer identity resolution to connect a person's online behaviour to their purchase history to their service interactions. Fixing that is not glamorous AI work. It is data engineering, and it tends to be both slow and invisible to the people who approved the AI budget.

The organisational capacity problem is related. Building and maintaining AI-powered systems requires a type of technical capability that most mid-market retail tech teams do not have in depth: data engineering, ML ops, API integration, prompt engineering at production scale. The standard response is to buy a vendor solution, which gets you part of the way there but creates a dependency on vendor capability and vendor roadmap that can constrain what you can actually build.

The Shopify Question

I wrote earlier about Shopify Sidekick's first-use numbers and what adoption actually means in practice. The democratisation narrative around Shopify Magic and similar platform AI tools is real in the sense that access has improved. The tools exist. They are affordable.

The caveat worth adding is that platform-level AI tools are optimised for the common case. They work well for the things most merchants need most of the time. They work less well for retailers with specific, complex, or category-specialist requirements where the off-the-shelf model does not have the training data to handle nuance.

Fashion is a useful example. A general-purpose AI content tool will write a product description. It may not capture the brand voice, correctly identify the fabric hand, or describe the fit in the language that particular retailer's customers respond to. Getting that right requires customisation that platform tools do not typically offer without meaningful engineering effort.

What's Actually Working

The mid-market retailers getting genuine value from AI are mostly doing one of two things.

Some are using AI in contained, low-risk ways where the quality ceiling does not matter much: first-draft content generation that gets reviewed and edited, marketing copy variants for A/B testing, basic query handling in customer service. These deployments do not require sophisticated data infrastructure and deliver consistent value.

Others have made the data infrastructure investment first and are now building on top of it. This approach is slower and more expensive upfront, but the AI layer actually works when the data underneath it is clean. These retailers are in the minority.

The £92m at-risk estimate largely represents the gap between those two approaches: AI investments made before the data infrastructure was ready, or vendor solutions deployed without the customisation needed to make them actually work for a specific business.

That is not a reason not to invest in AI. It is a reason to invest in the boring foundations first.