AI in ecommerce: what's actually worth your attention
The AI conversation in ecommerce has a hype problem. Every tool claims to be AI-powered, every conference deck leads with it, and it's hard to separate what's genuinely useful from what's marketing. Here's my honest view on where AI is actually delivering value for Shopify brands.
Where AI is delivering real value
Personalisation and product recommendations
This is probably the most mature application. Tools like Rebuy and Nosto analyse browsing behaviour, purchase history, and patterns across similar customers to surface relevant product recommendations in real time. When set up properly, this lifts average order value without feeling pushy. The key word is "properly." Out-of-the-box recommendations are often worse than well-curated manual selections. If you implement a personalisation tool, spend time on the rules and logic rather than just switching it on.
Search
Native Shopify search is limited. AI-powered search tools like Boost Commerce, Klevu, and Algolia understand natural language queries, handle misspellings, and learn from what customers click versus what they ignore. For stores with large catalogues, better search is often one of the highest-return improvements available.
Email marketing intelligence
Klaviyo's predictive analytics – churn risk, next purchase date, customer lifetime value – are genuinely useful for deciding who to target, when, and with what. This is AI working quietly in the background of something most brands are already doing. If you're using Klaviyo and not using its predictive features, you're leaving value on the table.
Customer support automation
Modern AI chatbots such as Gorgias and Tidio are the most common on Shopify – and can handle a meaningful volume of routine queries: order status, return policy, product questions. They've improved significantly and are no longer the frustrating dead-ends they used to be. They work best when there's a clean handover to a human for anything complex, with the conversation history intact.
Inventory and demand forecasting
Tools like Inventory Planner use sales history, seasonality, and trend data to help brands make better stocking decisions. For brands that regularly overstock slow lines or stock out on winners, this is worth looking at.
Fraud detection
Shopify has decent built-in fraud signals, but tools like Signifyd and Riskified go further for brands with high transaction volumes or elevated fraud risk. This one largely runs itself once configured.
Where to be more cautious
AI-generated content at scale
Using AI to write product descriptions can save time for large catalogues. But the output needs editing – generic AI copy is obvious, and product pages are where your brand voice matters most. Use it to draft, not to publish.
Virtual try-on and AR
Getting better, but still works best for specific categories – eyewear, makeup, furniture visualisation. If you're not in one of those categories, it's not worth the investment yet.
Dynamic pricing
Technically impressive, but can damage customer trust if people notice prices changing. Worth understanding but approach carefully.
The practical question: where do you start?
The mistake most brands make is treating AI as a project rather than a question. The question is always: what specific problem am I trying to solve, and is there an AI-powered tool that addresses it better than what I'm currently doing?
Start there rather than with a list of tools. If your search is letting customers down, look at search. If your email list is underperforming, look at Klaviyo's predictive features. If customers are asking the same five questions repeatedly, look at a chatbot.
One well-implemented tool that solves a real problem is worth more than five half-configured ones that add cost and complexity.
A note on pace
This is one area where content dates quickly. The tools available today are meaningfully different from eighteen months ago, and they'll be different again in eighteen months' time.
If you're making decisions about AI investment, it's worth getting a current view of the market rather than relying on a blog post – including this one.
That's a conversation I'm happy to have as part of a broader tech stack review.
