What Changed in Retail AI Last Year - And What to Focus on in 2026

Last year AI finally made the shift in retail from mostly experimentation to practical solutions. Mindera’s Phill Gillespie reviews the key innovations and the exciting opportunities they offer retailers for 2026.

Wherever you were with AI at the end of 2025, the New Year brings a fresh start. Optimising its potential across your business can be your New Year’s resolution to drive success across 2026.

Drawing on our findings from last year, we’ll kick start the process by helping you:

· Identify the five most important AI advances in retail last year
· Understand how to overcome the biggest challenge to AI adoption
· Pinpoint the best opportunity to focus on for the year ahead

First, let’s look at how 2025 finally saw AI transform from fantasy to reality…

Shifting From Pilots to Projects

In 2023, AI opened up a world of new possibilities with mass-market access to impressive generative capabilities. Then came 2024, the year of mass experimentation, but limited impact. Last year marked an inflection point - some retailers were soul searching, trying to figure out what AI could actually do for them. Others donned their hard hats and shifted away from shiny demos and pilots that weren’t going anywhere, to building practical integrations to drive longer-term measurable returns.

If you felt AI in 2025 was a grind, you were doing it right. While the general market remained distracted by noise, the smart retailers embraced the unglamorous work of integration, putting AI foundations in place.

The most important were using gen AI to power extensive product attribution tagging and draft product descriptions for review and iteration by in-house experts. These two seemingly simple projects have helped retailers open up far more commercial possibilities, including better commercial and trading analytics, better product discovery for users and even product to customer personalization. If you don’t have these in place or on your roadmap for 2026, there’s still time. Doing this will help you lay the foundations to make the most of last year’s AI breakthroughs (summarised below).

The AI Breakthroughs That Changed Retail

‘Thinking’ AI models unlocked reasoning & decision making

Chain-of-Thought AI models arrived in 2025, with tools like GPT-o3 and DeepSeek R1 ‘thinking’ before answering. Breaking down complex requests into a step-by-step plan, they test their own logic, often iterating on the answer. Although more costly in time and tokens, they deliver far better results when deployed in the right way.

Why it matters: These tools enable the use of AI in judgement-heavy workflows that have been too risky or complex for earlier gen AI models (from finance to operations to writing code for real-world environments).

A Standard For Integrating AI Emerged

Widespread adoption of Model Context Protocol (MCP) meant it has become the de facto standard for connecting AI models securely to external tools and internal systems without requiring extensive bespoke glue code for every integration.

Why this matters: This means all AI tools will use the same integration, reducing the risk or potential regret from using a specific tool while ramping up customer service chatbot capabilities. Think Customer Service bots processing instant refunds for customers, rather than just apologising in the chat window.

Agents Made AI Modular & Deployable

We were promised that 2025 would be ‘the year of the Agents’ and ‘the rise of Agentic AI’. While agents have started to be deployed in production, we’ve realised this is a marathon not a sprint. Retailers are using agents to automate existing workflows that weren’t possible with earlier RPA solutions. There are incremental improvements to be made here, but those companies with the vision to embrace holistic workflow redesign are the ones primed to make the substantial gains.

Why this matters: On page merchandising can be adjusted hourly by agents to ensure all products get visibility with customers that match their stock profile. This will help ensure the optimum sell-through rate for the season. i.e. don't show customers sold out products, but also keep the visual appeal of the page while promoting products that have more depth at the right time.

‘Jagged Intelligence’ Paradox Appears

AI models can now answer PhD-level questions yet can struggle with simple maths - like brilliant graduates with no common sense. Redesigning workflows to verify their outputs helps overcome this problem and unlock value.

Why this matters: Rather than asking AI to manipulate data for you, prompt it to write reusable code that can do it reliably and repeatedly.

More Advanced AI Wasn’t Always Better

Last year we discovered smaller AI models (like Gemini Flash, Claude Sonnet or GPT-mini) can deliver better results than their bigger ‘thinking’ relatives by breaking complex tasks down into smaller steps. These smaller models run faster and operate at a fraction of the cost of the larger ones. Combined with micro / nano models and open models you can run on your own hardware (either open-weights or open-source) there is no longer one-size-fits-all.

Why this matters: Because you don't buy a Ferrari for grocery shopping. Instead use small / open models for speed, lower cost and data sovereignty.

The Culture Shock of AI Adoption At Work

Although 76% of executives think their employees are excited about adopting AI, in reality only 31% are, according to a recent BCG/Columbia survey. This startling disconnect is preventing companies from realising the full benefits of AI.

Companies think they are driving adoption when their teams complete mandatory AI training, but then they don't actually change their workflows. Although this educate-first approach seems logical, it doesn’t overcome the distrust and inertia many employees feel when adopting new tools to boost efficiency. This demands winning hearts THEN minds, and it’s this year’s big challenge for retailers. Failing to bridge this gap risks deploying expensive AI tools that simply sit on the shelf and deliver no benefit.

We’ve been directly addressing this issue with our most forward-thinking clients, ensuring their entire workforce is engaged and advocating a future working with AI, rather than simply rolling out the technology to passive users. It’s important to not just give teams AI tooling, but also ensure they have the time to adopt. To recognise that the more specialised the task, the less helpful AI is likely to be. To take into account that a new tool will probably add inertia initially, but will speed up processes in the long run.

The big question for 2026 is: Are you building a culture around AI that makes your people feel like it’s replacing them, while imposing unrealistic performance expectations, or one that makes them feel supported and gives them superpowers? The answer will determine whether you fall into the Engagement Gap or leap over it.

The Opportunity: Progress at Lightning Pace

The most exciting shift this year is that AI will make previously ‘impossible’ projects financially viable. For example, AI-governed modernisation (inc. modularisation) is providing a new option between ‘buy’ and ‘build’, particularly for replacing legacy systems.

AI’s ability to drastically reduce the cost of code generation is just one part of the jigsaw. People had long believed that the bottleneck was down to the ability to write code. It turns out the real scarcity is knowing how to build solutions properly. It’s about architecting, designing and securing reliable solutions that don't crash during peak periods.

Combining hard-earned insights with a proven, structured approach can help develop AI agents that can deliver initiatives previously too complex or expensive to justify.

Imagine rewriting your entire legacy picking system in 8 weeks, not 8 to 12 months. That is the scale of the ‘Impossible’ projects now within reach. We aren't just doing the same things faster, we’re unlocking an entirely new layer of opportunity along with the value that brings.

The Key To Success In 2026

The retailers who will thrive aren't those who choose speed over people, or culture over capability. Success belongs to those who recognise these forces as inseparable. Your ecommerce AI agents can optimise sell-through rates brilliantly, but the system needs a team who understands its purpose and advocates for it.

Reconciling the cultural challenge with the technical opportunity will be the key to success in 2026 and beyond. Design your AI initiatives with equal attention to technology and people. Measure adoption by genuine workflow transformation, not training completion. Build organisations where AI equips teams with superpowers rather than threatening them.

At Mindera, we’ve spent the last year working alongside retail technology leaders to separate AI hype from practical implementations. Through workshops, assessments and hands-on delivery, we’ve helped clients identify which “impossible” projects are now viable and build the technical foundations across the culture that unlock broader commercial opportunities. Our approach addresses both sides of the equation: the engineering expertise to deliver applied AI projects and the strategic guidance to ensure your organisation is culturally ready to adopt them.

The question isn't whether to pursue AI - the market has decided that. It's whether you'll bring your entire organisation along, or deploy sophisticated technology to a workforce that isn't ready.

Get this balance right, and 2026 could be the year your organisation truly transforms.

Want to do more with AI in 2026? Then let's talk!

Contact our retail technology experts today.