Sure, there are the inflated promises and buzzword bingo. But there are genuinely impactful ways AI is transforming online retail. It’s no longer limited to chatbots or basic personalization. Instead, AI is woven through content management, search, merchandising, fraud detection, customer service, and analytics. The potential payoff: smoother, faster, and deeply personalized customer experiences, alongside leaner, smarter operations for the business.

Yet, as usual, the theory is a lot prettier than the reality. Actual results on the ground remain uneven, and the hype often outruns the substance.

The truth is, AI-driven gains don’t materialize without the right digital foundation. If your infrastructure isn’t composable, structured, and ready to support rapid experimentation, AI won’t magically solve your problems—it’ll just create new ones that you won't be equipped to solve. The faster AI changes things, the further behind you will get.

One of the clearest wins for AI in e-commerce has been search. Legacy keyword-driven systems often miss the intent behind what someone is actually trying to find. AI-powered search tools have stepped in to interpret that intent more effectively, serving results that align with what users mean, the context around their search, the behaviors that might illuminate the right response - not just what a user types. This leads to fewer dead-ends, less friction, and better conversion rates, especially on large catalogs.

Another area where AI has proven valuable is in content automation. AI-assisted copy, particularly translation, and increasingly image generation has started to help content teams scale campaigns faster without sacrificing cohesion. For brands with hundreds or thousands of SKUs, these tools can speed up product launches, landing page updates, and A/B testing strategies. It doesn't eliminate creative work - in fact you have to be somewhat more creative over time to get AI to do what you want - but it shifts the team’s focus to higher-impact content, the sort of thing you pay people their money for. AI isn't going to have a lot of original ideas. If you can't compete with that then the problem isn't AI.

Customer service has been the canary in the mine, with Chat being at the irritating vanguard of AI impact for multiple years now. Using AI to do routine ecommerce work today has all the frustrations you were giving customers two years ago when you thought Bots could replace your customer service agents - the forgetting things, the cheerful stupidity, the 'nearly but not quite' nature of their responses. But customer service is where to look to see how effective AI can be when it's well trained. How many times in the past few months have you interacted with a bot, wondered if it was a bot, and not asked because you didn't want to insult a human being. (For me: three times).

Generative AI is powering self-service layers that go beyond traditional FAQ bots. When properly trained on accurate business and order data, these models can help resolve common requests instantly, while escalating more complex issues to a human without losing the thread. That blend of automation and human handoff is pivotal in keeping costs down without degrading the customer experience. That's where the rubber is still making an unpleasant mark on the road. When I tell a bot my details and the human doesn't appear to have been given any of it, I'm mad. The humans and their systems are the problem now.

These gains, while sometimes incremental, add up. When AI is fully integrated into the architecture, it becomes a quiet performance multiplier. Teams move faster, experiences adapt to real user behavior, and internal data becomes more actionable. But you need a partner who understands how to put this all together, an engineering company. (Okay okay, get me a glass of water).

Despite all this, not every AI implementation delivers value. In fact, some create more problems than they solve.

A common issue is technical mismatch. Many organizations are trying to plug advanced AI tools into outdated platforms. If your e-commerce architecture is tightly coupled or monolithic, real-time personalization and AI-driven experiences will struggle to perform. The stack simply can’t handle the speed or complexity that modern AI tools require. You’ll end up with laggy site performance, inconsistent experiences, and frustrated teams. The silent part here is that if you're still on a "very" legacy version of your platform the problem might be that you don't run towards change. Get at least competent in managing the pace of technical change in our business or get rolled over.

Another problem is dirty or disjointed data. AI depends on clean inputs. When product data is incomplete, customer profiles are fragmented across systems, or tracking data is unreliable, the models being used won’t work well. This isn’t a tool issue, it’s an infrastructure issue. AI doesn't fix data problems. It magnifies them. You can't go to a trade show and buy something that does AI and hope that this solves your problems. You've got to start with a clear picture of what your capabilities are today, an honest one that acknowledges maybe you haven't changed as fast as you might have. Once you know your capabilities, know what you want to do. Focus on developing skills aggressively but incrementally, pushing and pushing for several months to develop greater understanding and ambition among your teams. Think of yourself 6 months from now and imagine AI actually can do something useful in early 2026. What needs to be true for you to take advantage? What will you wish you did now? Get on with it.

There’s also a strategic challenge. Many organizations adopt AI without a clear use case. They chase a trend, roll out an AI tool, and then wonder why it doesn’t move the needle. The ones who will see value are the ones that start with a specific outcome in mind (that picture of the world in 6 and 12 months is fuzzy, you are going to have to guess some of the details). Start with pain or greed. Pain? Do you need to reduce support costs. Pain: Improve search engagement. Greed: Could you increase campaign velocity. Then work backwards, match the skills and technology to the need and work out where to start. (Where you end up is too hard to predict. AI is developing too quickly.)

Lastly, AI is often introduced without a plan for internal ownership. So you start yourself a political AI death match at the very moment you need everyone to be working together. The most successful implementations happen when cross-functional teams are brought into the design process early. And when a small group of people get to deliver something real as a small cross-functional team. They develop trust. And that's the culture your AI initiatives will have. If only one department or one egomaniacal super-ambitious leader is responsible for your AI initiative, it not only becomes siloed, it becomes us versus them. If your best people are put in the trenches together and held accountable jointly for the first outputs, the extended team has a chance to see AI as a shared capability, it becomes part of how the business works.

When I was at London Business School, I had this professor, Costas Markides. He had a book. I still have it and I think of it in every conversation we have at 64labs about innovation. The people who run the fastest at the start of an innovation wave are not always, in fact are rarely, the companies who benefit the most. (MySpace is the easiest to think about).

I had a sports editor and friend at the Sunday Times, Chris Nawrat. He was one of the smartest people I knew. He had this thing called the "Do Nothing" strategy. "John," he would say. "Sometimes the right thing to do is nothing. Let everyone else waste their energy, while think about what the possible end games might be and prepare for them instead." He survived at the Sunday Times the longest of any executive other than Andrew Neil.

So two questions. What's the end game of AI in your business do you think. What would Nawrat be preparing for? And does it matter if you arrive at the right answer on AI slightly after the initial discoverers have found it. Save some money. Plan rather than act.

The businesses who will win at AI aren’t necessarily spending more money or hiring more engineers. They’re structuring their teams and technology in ways that support process, clarity, simplicity. flexibility, speed, and interoperability. The sort of stuff that will mean they understand how best to apply AI in their business. Not actually rushing right away to the AI bit. Prepare.

And that's why composable should be on your agenda. If you adopt composable principles at an architectural level, give yourself the ability to test, refine, and scale new tools without refactoring your entire stack. Pay off technical debt urgently. You won't be able to service it in AI land. Invest in cleaning up and centralizing your data. Choose vendors based on their coherence and connectivity with the rest of your stack and your ability to use them as tools rather than products. Get some useful experience in an IT and content architecture where you get equal parts more power and more responsibility. But don't spend a lot of money on fancy consultants who barely know more about AI than you do. Build an internal coalition of the willing—people curious enough to figure out what works right now and committed enough to turn that learning into action. Let them develop just enough expertise to be dangerous when the end game starts. Then make sure they’re building the muscle to train others and evolve the tools internally, not just depend on vendors forever.

AI is absolutely going to shape the future of e-commerce in some way at some point. But right now it's either doing useful but banal work or it's a cute party trick. That will change quickly. But the quicker you admit you have no clue how it will affect your company exactly and start understanding the implications of that ignorance and what you can actually do about it for real, the better.

Those that approach AI thoughtfully, match it to real use cases, and modernize their infrastructure accordingly will see measurable returns. Those that treat it as a bolt-on feature or marketing bullet point will fall short. Not because the tools don’t work, but because the foundation wasn’t ready.

If you're serious about integrating AI into your customer experience, the first step isn’t to buy a new tool. It's to make sure your architecture, data, and teams are capable of supporting it.

Want a second opinion on whether your current stack is built to support modern AI tools? Let’s talk.