Retail has always evolved through waves of innovation – from bricks to clicks and from monolithic systems to microservices – with each shift demanding investment, new capabilities, and operational and cultural change.  

But transformation in the AI era is different, says Nimitt Desai, Head of AI, Innovation & Technology at PMC, and requires retailers to adopt a different innovation playbook if they are to succeed.

New rules, new playbook

Unlike previous technology cycles or innovation waves, AI isn’t confined to one function, department or channel, Desai told audiences at last month’s Retail Technology Show. It cuts across every part of a retail organisation – from CX and merchandising to supply chain, operations, and everything in between.

It is also no longer optional. For retailers, AI adoption is quickly becoming a question of competitiveness rather than experimentation. 

Data in PMC’s Race to Unified Commerce report shows this pressure is only intensifying, with 86% of retailers saying customer expectations are evolving faster than their digital capabilities.

Yet, amid the rush to deploy AI and move from theory to execution, many retailers risk focusing on – and investing in – the wrong problem. Getting AI right isn’t about deploying more AI.  It’s about getting the fundamentals right, and that starts with strong data foundations.

Getting AI past the pilot

Across the market, a familiar pattern is emerging where early AI promise doesn’t match real-world execution.

All too often we see retailers jumping ahead, investing heavily in AI-powered software, platforms or integrations to solve use cases, such as personalisation, intelligent search, forecasting or customer service automation, before getting the basics right.   

That means, even when early proof-of-concepts or pilots look promising, the reality hits when the initiative goes live and doesn’t match initial expectations. Recommendations still feel irrelevant, search results remain glitchy, AI decisioning conflicts with commercial reality and internal users revert to the very manual processes AI sought to automate.

The instinct is to blame the AI model or the technology stack but, in reality, the issue is much more fundamental. This isn’t a failure of AI – it’s a failure of data. We’ve all heard the phrase “garbage in, garbage out” and, while overused, there’s no denying that AI systems depend entirely on the quality of the information they are fed.

When customer data is fragmented, stock files are displaced or product information is inconsistent, no amount of AI investment will fix that issue while the underlying data foundations remain fragile. If anything, it will only expose – and in some cases amplify – those weaknesses before any meaningful ‘fix’ can be achieved or ROI delivered.

Creating context – not just content

Many retailers still confuse data volume with quality when it comes to feeding AI programmes, assuming that centralising information in a data lake or warehouse will solve their problems. While this is part of the journey, it is not the same as AI readiness.

Collecting or storing data in one place does not automatically make it usable, trusted or meaningful. Strong data foundations require information to move cleanly across systems and functions, to be governed with clear ownership and quality controls, and to reflect the latest operational reality. More importantly still, for AI to succeed, data must carry context. 

This is where semantic layers, organisation-specific definitions and knowledge graphs are becoming increasingly relevant. While technical in nature, their purpose is commercial: giving AI a clearer understanding of how enterprise data fits together, so decisions and outputs are grounded in reality.

No two AI strategies can ever be the same

A one-size-fits-all strategy for AI can never exist because no two retailers ever start from the same place.  

While some will have modern cloud platforms and advanced analytics estates, others will be managing legacy technology and decentralised ownership models.  

Take for example, Ann Summers, which recently underwent a multi-year transformation programme focused on modernising what it described as the “heart and lungs” of its business. By replacing complex legacy integrations with a more modular architecture using PMC’s Graphene platform, Ann Summers created stronger operational foundations and better data flow across the organisation – a reminder that AI readiness can start with simplifying an existing estate rather than building from scratch.

Similarly, the needs and desired AI outputs will also be unique to each retail business and will impact transformation strategy; some may need real-time operational intelligence, while others require productivity gains or personalisation enhancements. As such, there is no universal blueprint for AI readiness.

The most effective approach is to begin with priority business outcomes, then align the data domains needed to support them. For example, if the objective is personalisation, customer identity resolution and product data quality may be the immediate priorities. If the focus is on supply chain optimisation, inventory accuracy and supplier visibility will become more important.

While retailers don’t need to delay their AI ambitions when embarking on a multi-year transformation programme, they do need to build data capability in parallel.

Substance, not volume, will drive AI-led advantage

Critically, the “winners” in retail are unlikely to be those announcing the most AI pilots or buying the most expensive AI-powered tools. They will be those able to scale use cases confidently and effectively because their foundation – i.e. data – is strong.

AI undoubtedly has the potential to transform retail, but data will determine whether that transformation is real, repeatable and commercially valuable.