Anyone who's shopped online recently has probably noticed how weirdly good the recommendations have gotten. That homepage on Amazon? It's been custom-built for you in the last few seconds. And the same thing is happening on Nike, Sephora, and thousands of smaller retailers you've never heard of.
The numbers matter here. Retailers spent around $12 billion on AI last year, and Amazon says its recommendation engine drives roughly 35% of sales. Whatever you think of the hype, something real is happening under the hood.
Personalization That Actually Works
Recommendation engines used to be dumb. They matched you with people who bought similar stuff and called it a day. Basic collaborative filtering, and it worked fine when the internet was smaller.
Modern systems track a lot more. Dwell time, scroll speed, how you hover over images, whether you resize product photos (that one's a bigger signal than you'd think). Some also factor in time of day and device model.
Stitch Fix built an entire company on this idea, pairing machine picks with human stylists. At its peak the company hit around $2.1 billion in annual revenue. Growth has slowed since, but the core lesson stuck: people buy more when the store seems to actually know them.
Building this kind of system from scratch takes engineers most retailers don't have. Tools like uxify package the pieces (search, product tagging, personalized recommendations) into something a small team can actually run. Nobody's competing with Amazon's ML division, but they don't have to.
Visual Search Actually Changed How People Shop
Pinterest Lens does over 600 million visual searches a month. Snap a photo of a jacket in a coffee shop, get shopping links before your latte cools. Google Lens does something similar across its shopping ecosystem, and both keep getting scarily good.
This matters most for fashion and home goods, where nobody knows what to type in a search box for that specific weave or shade of blue. Nike and IKEA both report solid conversion lifts from adding visual search. Thecomputer vision techniques underneath (mostly deep neural networks trained on billions of product images) have been around a while, but only recently got cheap enough to run at retail scale.
AI Chatbots Finally Got Useful
Chatbots were terrible for years. Keyword-matching, canned responses, escalation to a human after five minutes of frustration.
Large language models fixed most of that almost overnight.
Klarna's AI assistant reportedly did the work of 700 human agents in its first month, resolving queries in under two minutes on average versus 11 for humans.
Sephora's Virtual Artist takes a similar approach. ButHarvard Business Review has pointed out the obvious catch: bots handle the routine stuff (returns, order status) great, and mess up the complicated stuff (fraud disputes, custom orders) in expensive ways.
Pricing Got Weird
Prices on major retail sites now change thousands of times a day. Amazon updates some product prices every 10 minutes. The models pull in competitor pricing, inventory levels, seasonal patterns, and whatever they can figure out about you as a shopper.
Predictive inventory works on the same principles. Walmart's forecasting models chew through weather data, local events, and social sentiment to guess SKU-level demand weeks ahead. MIT Technology Review has covered how these systems ripple out into warehouse staffing and even shipping routes.
There's a regulatory question hovering over all this. When personalized pricing starts penalizing shoppers based on device type or zip code, watchdogs notice. The EU's Digital Services Act already forces retailers to be more open about algorithmic personalization, and similar rules are showing up in California.
The Trust Problem Nobody Wants to Talk About
All this AI comes with a downside. Generated product photos can lie about fit and color, and AI-written reviews poison the well shoppers drink from when deciding what to buy.
Amazon pulled roughly 200 million suspected fake reviews in 2020 alone, and that fight has only gotten harder since.
Retailers building AI features need to build AI detection at the same time. Otherwise the same tech that makes shopping smoother poisons the whole experience, and shoppers eventually notice.
What Comes Next
Generative AI is the next battleground. It's already writing product descriptions, generating personalized email campaigns, and (more controversially) creating model photos without hiring humans. H&M and Levi's have both experimented with AI-generated model imagery, with mixed public reaction.
Whether shoppers will actually accept synthetic imagery once they know what they're looking at is anyone's guess. What's clear enough: retailers pouring money into AI infrastructure right now are pulling ahead, and the gap widens every quarter.
