AI Will Transform Retail More in the Next Five Years Than It Did in the Last Twenty Five

Online retail has not fundamentally changed since the late 1990s. We still browse grids of products, apply filters, and read recommendations generated from what other people bought. Generative and multimodal AI break that model.

Over the next five years, shopping will shift from navigating a fixed catalogue to describing what you want and having the experience assembled around you. This is not a better interface. It is a different category of product.

The 25-Year Plateau Nobody Talks About

E-commerce solved logistics, payments, trust and delivery speed. It never really solved discovery.

Open almost any retail site today and the underlying architecture is the same one Amazon shipped in 1998: a homepage, a category tree, a search box, a product grid, a detail page, a cart. Two decades of investment in personalisation produced a modest layer on top of that skeleton. Slightly different banners. A carousel of related items. A retargeting email.

That is segmentation, not personalisation. You are being shown what people statistically similar to you clicked on. The catalogue itself never changes. Even the most sophisticated performance marketing and content marketing playbooks were built to feed traffic into that same fixed structure rather than to change it.

The reason the plateau lasted so long is that the technology to break it did not exist. It does now.

Why This Moment Is Different: The Adaptive Commerce Stack

Four capabilities matured at roughly the same time, and a fifth arrived to connect them to money. Together they form what I would call the adaptive commerce stack.

1. Language Models That Understand Intent

A shopper typing "something comfortable for a small north-facing living room, nothing beige" is expressing constraints, taste and context in a single sentence. Keyword search discards almost all of it. Language models retain it, which is why objective-driven AI systems are reshaping machine intelligence away from pattern matching and towards goal interpretation.

2. Generative AI That Produces Experiences, Not Just Answers

Copy, comparisons, buying guides, room descriptions and entire page layouts can be created at request time rather than authored in advance. The way GenAI transformed e-commerce during the Diwali season was an early, seasonal preview of what becomes permanent infrastructure.

3. Multimodal Models That Reason Across Text, Image, Video and Voice

A shopper can photograph a room, describe a mood, and point at a reference image. The model treats all three as one input.

4. Image Generation That Visualises the Outcome

The single biggest friction in considered purchases is imagination. AI removes it by showing the shopper the result instead of asking them to picture it. Tools in the lineage of Meta's Movie Gen AI video generator are making that visualisation cheap enough to run on every product, in every context.

5. Agentic Commerce Infrastructure

This is the part most retailers underestimate. Discovery is moving to surfaces that retailers do not own, and those surfaces now have standards. Anyone who has followed the rise of agentic AI and autonomous intelligence will recognise the pattern: capability first, protocols second, distribution shift third.

In January 2026, Google launched the Universal Commerce Protocol (UCP) at NRF, an open standard co-developed with Shopify, Etsy, Wayfair and Target, designed to let AI agents handle discovery, purchase and post-purchase support across platforms while the retailer stays the merchant of record. At Google I/O in May 2026 it added Universal Cart, a cross-retailer cart that follows a shopper across Search, Gemini, YouTube and Gmail, plus updates to the Agent Payments Protocol (AP2), which lets agents complete payments within limits a user has authorised. OpenAI is building the competing Agentic Commerce Protocol with Stripe, following its Instant Checkout launch in September 2025.

McKinsey has estimated that AI-driven tools and agentic commerce could represent a three to five trillion dollar global retail opportunity by 2030.

The point is not which protocol wins. The point is that the checkout button is leaving the retailer's website, in much the same way that Google's AI is quietly killing the traditional search box.

The End of the Static Catalogue

Here is the clearest way to see the shift.

Static commerce asks the shopper to translate a fuzzy human need into database queries. Adaptive commerce does the translating.

Home, Furniture and Interiors Will Change First

If you want to know where this lands first, look at the categories where the product is not really the product.

Nobody buys a sofa because they want a sofa. They are buying:

• how it sits in a specific room with specific light
• whether it works with the rug, the walls and the chair they already own
• the atmosphere it creates when people walk in
• the version of their life it signals

Every one of those variables is visual, spatial, contextual and emotional. Every one of them is precisely what a multimodal model can now hold at once, and what a filter on "three-seater, fabric, under 60,000" cannot. This is the same logic behind augmented reality's revolution in Indian retail, only now the rendering engine is generative rather than geometric.

Home and interiors also have the worst returns economics and the longest consideration cycles in retail, which means the value of showing someone the outcome before they commit is enormous. Fashion, beauty, eyewear, tiles, paint, kitchens, bathroom fittings and real estate all sit in the same bucket for the same reason: high imagination cost, high emotional weight, high return risk.

The gap between imagining and seeing is where the next decade of retail margin lives.

Recommendation Engines Optimise Products, AI Optimises Experiences

For fifteen years the industry poured money into a single sentence: "customers who bought this also bought."

Recommendation engines are ranking systems. They take a fixed set of items and reorder it. They are extremely good at increasing attachment rate on an existing journey, and structurally incapable of changing the journey. Even a well-built TOFU, MOFU and BOFU funnel model assumes the shopper walks through stages the retailer has designed in advance.

Generative AI is not a ranking problem. It is a composition problem. Instead of choosing which of ten thousand products to show, it decides what the shopper should encounter, in what form, in what order, with what explanation, and what it should look like.

That is a harder problem and a much larger prize.

The Five Shifts to Plan For

The next generation of retail will not be won with a better search bar or a bigger catalogue. It will be won on five shifts, and each one maps onto the broader AI trends redefining business impact in 2025:

1. Visual rather than textual. Shoppers will point, photograph and gesture more than they type.
2. Conversational rather than navigational. The journey becomes a dialogue, not a hierarchy of menus, echoing how WhatsApp became a direct brand-to-customer channel.
3. Context-aware rather than rule-based. Room, budget, climate, occasion, existing possessions and mood become live inputs.
4. Adaptive rather than static. The page is generated, not retrieved.
5. Individual rather than segmented. One customer, one experience, at scale.

What Retailers Should Actually Do in the Next Twelve Months

Most of the useful work here is unglamorous.

• Fix your product data before your AI. Agentic surfaces read structured attributes, not marketing copy. Missing dimensions, materials, care instructions and lifestyle tags are the single most common reason a brand is invisible in conversational discovery, which is why knowledge engineering and domain graphs underpin AI data connectivity.
• Assume discovery happens off-site. Treat AI assistants as a channel with its own ranking logic, and measure your presence in it the way you once measured organic rankings. A working knowledge of SEO and AI optimisation for content professionals is now a merchandising skill, not just a marketing one.
• Build a visual asset library, not a photo shoot. Models need products in varied contexts, angles and lighting to place them convincingly in someone's space.
• Pick a protocol position deliberately. Adopting UCP, ACP or both is a distribution decision, not an IT decision.
• Instrument intent, not just clicks. If you cannot see what shoppers asked for and did not find, you cannot improve a generative experience. Modern martech stacks built for smarter marketing are only as good as the intent signals feeding them.
• Decide what you will not delegate. Brand voice, pricing integrity and after-sales relationship are worth defending even when an agent stands between you and the customer, a tension already visible in the rise of vibe marketing built for the AI generation.

The Bottom Line

AI is not improving e-commerce. It is redefining what shopping is.

The retailers who understand this will not simply sell more units. They will build experiences that feel as though they were designed for one person, because they were. Everyone else will keep optimising a 1998 storefront while the customer has the conversation somewhere else.

That is the real transition of the next five years.

Frequently Asked Questions

Will AI replace e-commerce websites?

No, but it demotes them. The website becomes one endpoint among several, with a growing share of discovery and even checkout happening inside AI assistants, agents and aggregated carts. The retailer's site becomes the place for depth, trust and after-sales rather than the only front door.

What is agentic commerce?

Agentic commerce is shopping in which an AI agent carries out steps on the shopper's behalf, from researching and comparing options to completing payment within limits the user has set. If you are new to the underlying technology, this practical guide to how AI agents work is a useful starting point. Open standards such as Google's Universal Commerce Protocol and OpenAI's Agentic Commerce Protocol exist to let those agents transact with merchants without custom integrations for every platform.

How is generative AI different from a recommendation engine?

A recommendation engine reorders a fixed catalogue. Generative AI composes the experience itself, including the options presented, the explanation, the imagery and the layout. One optimises products, the other optimises journeys.

Which retail categories are affected first?

Categories with high imagination cost and high return risk: home and furniture, interiors, fashion, beauty, eyewear, tiles and paint, kitchens, and property. In all of these, the shopper is buying an outcome they cannot currently see.

What should a mid-sized retailer prioritise?

Structured product data and visual assets. Both are prerequisites for every AI capability that follows, and neither requires betting on a specific model or vendor.

This Series

This is the opening article in a series on AI and commerce. Upcoming pieces:

1. AI-powered product discovery: what replaces the category tree
2. The future of visual search: why the camera becomes the search bar
3. AI shopping agents: what happens when your customer never visits your site
4. The personal AI salesperson: why every retailer will eventually have one
5. Measuring what you cannot see: analytics for generative and agentic journeys