AI’s Forcing Function: Turning Retail’s Debt Into Its Differentiator

We are thrilled to be sponsoring this year’s FT Future of Retail Summit, where retailers will be addressing the urgent issue of how AI can boost operational efficiency and deliver tangible ROI. With this in mind, Equal Experts Strategic Advisor Emma Payne spoke with Andy Redpath, EE engagement manager, to ask what it actually takes to make AI deliver behind the scenes.

Emma Payne: There’s a lot of talk about AI driving operational efficiency, but as someone working directly with retailers are you seeing this in reality?

Andy Redpath:  There’s a massive temptation to treat AI as a silver bullet that you can fire at any operational problem, but if you bolt AI onto messy, immature foundations, all you’re doing is automating poor processes at a faster pace.

The reality is that AI is proving to be a game changer in reading and reasoning about end-to-end processes across the business’s buy-move-sell operation. It is giving us renewed courage to tackle problems in commercial and supply chain that have felt insurmountable for years. Techniques like RAG (retrieval augmented generation) and data agent patterns enable retailers to stitch together disconnected, unstructured datasets, to separate signal from noise for operators and build peripheral vision across the supply chain. We can use AI to reverse-engineer legacy mainframe logic and tables into modern APIs. That’s not hype, that’s new incremental capability.

You know it’s hype when a promise of an ambitious multi-agent workflow is made without understanding your process or asking what’s underneath. Most operations carry heritage debt, batch processes and inconsistent taxonomies and are fundamentally siloed. Point an agentic workflow at that and you get a demo, possibly a very local optimisation if you’re lucky, but certainly not an enduring result.

Once you’ve got a better handle on the problem, the solution doesn’t have to be agentic. Sometimes it’s a set of skills, ML, or just good deterministic software (which can be shaped and shipped faster via AI-assisted cycles of course). The right tool for the job applies as much as ever.

Emma Payne: The next question is inevitably,  how quickly can we move from operational pilot to roll-out, and how best do we measure ROI?

Andy Redpath: Moving from pilot to production is as much a business adoption question as a technical one. Identifying the pain points with super-user operators and co-creating solutions increases buy-in and matters just as much as the engineering. We can certainly move valuable, well-crafted knowledge systems, data agents and RAG interfaces into production-beta through good service introduction discipline just as we did for deterministic software. The real risk is letting an unsupported, brittle system quietly become load-bearing.

ROI is rarely one number. It might be availability improvements measured in pounds, working capital that’s freed up by better forecasting, or the SI and licensing costs you’re no longer paying for aging commercial off-the-shelf software that has been replaced by AI assisted development processes. Of course, we also need to consider security risk mitigation, which is priceless.

The harder problem sits in the organisation structure. The impact and benefit may be enterprise-wide, so a senior sponsor needs to have ownership beyond whichever silo the data and systems happen to start in with wide ranging support from the executive board.

Emma Payne: And more broadly, how can AI assist in-store colleagues and deliver much needed modernisation?

Andy Redpath: It’s the same principle as with commercial and supply chain colleagues, just on the shop floor: engage first, deliver a real, quick win, and apply proper product thinking in fast, AI-assisted cycles. It should feel very different from having a traditional software programme rolled out at them.

Focusing on the kind of things that help staff to delight a customer on the spot is more likely to get them on-side. It might be a lightweight tool flagging storeroom-to-shelf stock discrepancies to drive up on-shelf availability, or a voice assistant that gives colleagues instant answers on products about allergens or stock.

The trust ladder still applies: staff watch the tool be consistently right to build confidence. For agentic applications, we can dial the human-in-the-loop down when the evals and the metrics say so as a joint, informed decision with head office.

The store is an interesting focus alongside Commercial and Supply Chain. Strip out the COGS and labour is consistently the biggest single store cost line, ahead of energy, ahead of rent and business rates. Meaningful cost from pick routes, replenishment and rostering is in the workforce management, and workforce management is exactly the kind of bounded, observable problem AI is good at.

To the last point on ROI, it can be easy for store-based AI POCs to stop at hours saved benefits, and not at what was done with the time that was saved. Was this meaningfully used to enable a store colleague to move a number finance already tracks e.g. availability, waste, growth without additional headcount? Or did it just evaporate?

Emma Payne: Many IT departments are considering governance,model retraining and the ongoing operational costs. Are retailers setting themselves up for a maintenance trap?

Andy Redpath: Retraining for me feels like more of an MLOps lifecycle consideration, less of a generative AI one. Retraining can become a real question if you’re experimenting with open-weight models for cost or lock-in reasons, because then the lifecycle is more yours to own.

A lot of what looks like a retraining problem is actually a grounding problem. Models were never meant to know today’s stock or pricing. They know what the world looked like at training time and nothing since. That’s where RAG again can be used to ground the answer in live data at the point of ask, rather than trying to bake it into the model itself.

For more on what the shift towards agentic commerce means for retailers, The Joshua Problem: Why intent is everything in agentic commerce explores how AI agents are changing the customer journey and what retailers need to consider as they prepare for that future.

The bigger overhead is governance; a living contract for what each agent is allowed to touch, how it’s monitored, and when it gets retired. Skip that and you get dozens of disconnected scripts across merchandising, logistics and customer ops with nobody watching what they’re doing.

And everyone’s talking about tokenomics. An AI gateway with team-level accounting builds licence and token costs into the business case from day one, enables a sliding scale of access based on maturity, and lets you measure cost and ideally value.

See how we put this into practice with HMRC, building cost visibility and control into its Generative AI Landing Zone (GAILZ).

We’re seeing success patterns emerging on all counts.

Emma Payne: If fixing the plumbing and sorting data governance takes time, what are no regret investments right now ?

Andy Redpath: Wait for foundations to be perfect and you’ll never ship.

The better move is to use the AI experiment itself as the forcing function. Run a focused, real pilot and it immediately shines a light on your most acute gaps — the data locked in heritage systems that needs liberating via modern APIs, the load-bearing spreadsheets that have accumulated around the edges, the slice of semantic layer needed to provide meaning across multiple definitions, the batch that needs to be realtime. You’re not guessing which foundations to fix first; the pilot tells you.

And the foundations fix doesn’t need its own slow programme – as mentioned, product thinking and agentic harnesses can do the heavy lifting of fixing the foundations at the same pace as the pilot itself.

About the contributors

Emma Payne is a digital transformation and technology executive with over 25 years of experience across retail and the private sector. She has held senior leadership roles including Global Director of Technology and Product at John Lewis & Partners and Pret A Manger, reporting into the Executive Board. She has deep expertise and a proven track record in digital transformation, technology strategy, customer experience and AI adoption.

As a long-standing customer of Equal Experts, Emma joined the advisory team in 2023. She now works with senior leaders and high-performing teams on strategic planning, complex transformations and building lasting partnerships that deliver measurable business value.

Connect with Emma on LinkedIn.

Andy is an Engagement Manager with 25+ years’ experience across telecoms, energy, banking, capital markets, retail and entertainment. He started out as a software developer and has since worked as a field engineer, project manager, delivery manager, management consultant and operations director. He enjoys helping clients understand, navigate and simplify the complexity inside their own organisations, delivering £-shaped value and building capability that lasts.

Connect with Andy on LinkedIn.