2024 in AI Commerce: The Year It Got Real

The coverage out of NRF in January set a tone: every vendor had an AI story, the question was broadly "when" rather than "whether," and the general atmosphere was one of confident inevitability. Reading it from Manchester, it felt like a story being told faster than it was being lived.

Twelve months on, the picture is more interesting than either the optimists or the sceptics got right. AI in commerce did move in 2024. Not everywhere, not always in the ways predicted, and not as fast as the January confidence implied. But it moved.

What Moved

Customer service automation produced the most unambiguous commercial results of the year. Klarna is the headline case: in the first month of deploying its AI assistant, it handled 2.3 million conversations, the equivalent of 700 full-time agents, while resolving customer queries in under two minutes compared to eleven minutes previously. Customer satisfaction scores held level with human agents. These are Klarna's own figures, and they should be treated as such, but they are the kind of concrete numbers that change boardroom conversations.

What sits behind the Klarna headline is a model other retailers have been quietly following: start with a well-defined, high-volume, repetitive task where the inputs are reasonably clean and the success criteria are measurable. Customer service is that task. A lot of what mid-market retailers did in 2024 wasn't fully automated: it was AI-drafted responses reviewed by a human before sending. Even in that hybrid form, the efficiency gains are real.

Content generation at volume crossed a production threshold. What started in 2023 as "we ran a pilot on product descriptions" became, for many larger retailers, a standard workflow by the end of 2024. The quality ceiling for AI-generated product content turned out to be higher than critics suggested, and the economics are compelling enough that the question shifted from "is it good enough?" to "what's the review process?" This is also where the GEO implications I wrote about mid-year are starting to matter. Better-structured product content helps in both traditional search and AI-mediated discovery, and retailers who have improved their content infrastructure this year are building a head start, even if they don't think of it in those terms.

Retail media AI tools moved from sales pitch to shipping feature. Amazon's AI campaign optimisation tools are live and being used at scale. The transparency questions I raised in September about algorithmic budget allocation remain unanswered, but that doesn't appear to be slowing adoption.

What Didn't Move (Much)

Personalisation remains the gap between promise and reality. The tools exist. The data infrastructure often doesn't. The retailers doing genuinely personalised, AI-driven experiences are those that spent years beforehand doing the unglamorous work: unifying customer data, cleaning product catalogues, connecting CRM to commerce platform to inventory. Most retailers haven't done that work yet. The AI layer is ready; the foundation isn't.

Amazon's cashier-less store ambitions got a reality check. In April, Amazon announced it would no longer deploy Just Walk Out technology at its own Amazon Fresh grocery stores, shifting to Dash Cart for large-format retail. The technology is accurate at smaller scale and continues in third-party venues, including UK locations, but the large-format grocery ambition ran into the hard economics of a technology that requires substantial infrastructure investment to work reliably at scale. Amazon's own communications on this were notably candid.

Dynamic pricing was the story that went sideways most visibly. Wendy's announced then quickly abandoned surge pricing plans after immediate consumer backlash in early 2024. In the UK, Oasis's reunion tour ticket sale used dynamic pricing that saw prices double during the sale window, prompting a Competition and Markets Authority investigation. The tools are clearly there. Consumer trust clearly isn't, and the regulatory environment is starting to catch up with the commercial enthusiasm.

The Infrastructure Problem

The finding that stays with me from the year's research is this: the retailers who moved from interesting pilot to production system in 2024 generally did so in a specific pattern. They started with a use case where the data inputs were already clean, iterated quickly on a measurable outcome, and built outward from there. Customer service worked because conversation logs and resolution data are well-structured by nature. Product description generation worked because catalogue data, while imperfect, is at least in one place.

The retailers still stuck in pilot mode often share a common problem: they're trying to deploy AI against data that isn't ready, or they're trying to solve a problem too broad to evaluate clearly. "Improve the customer experience with AI" is not a deployable brief.

There's also an accounting problem worth noting. Much of the work that enables AI to function well, including data integration, product catalogue quality, and customer data unification, doesn't show up on an "AI investment" line. It shows up in data engineering, platform maintenance, or content operations. The retailers who will have genuine AI capability in 2025 and beyond aren't necessarily those with the biggest AI budgets. They're the ones that have been doing the infrastructure work, whether or not they called it AI investment.

What 2024 Actually Taught Us

The real lesson of the year is that AI in commerce isn't a capability you buy. It's something you build incrementally on top of operational and data foundations that take longer to establish than anyone admits.

The technology is better than it was twelve months ago and significantly more affordable. The tools are more accessible. The case studies are more concrete. We're at the end of the beginning, not anywhere near the end.

But 2024 taught me to be sceptical of the version of the AI story that says technology is the hard part. Mostly the technology is fine. The hard parts are the same hard parts they always were.

---

The 2025 edition of this piece will spend considerably more time on agents — autonomous AI systems that take actions, not just generate content. That's the shift that's coming, and the retailers who've spent 2024 sorting their data foundations will be far better placed to benefit from it.