Personalization means the shop adapts to one buyer: what they see, in what order, with what help. Yet plenty of stores still push winter boots in July. Here is what works, what is being tested, and where it goes wrong.
Where personalization stands
Amazon described its item-to-item recommendations back in the early 2000s: people who bought this also bought that. Netflix built a whole business logic around suggestions, and Spotify’s Discover Weekly taught millions of listeners to trust a playlist that no human assembled by hand.
The big players cracked it years ago. Everyone else?
Most mid-sized retailers sit somewhere between a “bestsellers” widget and a plugin installed in 2019. Those that want to move on often start with outside help, usually AI transformation services that open with an audit of data quality and a ranked list of use cases, well before anyone trains a model. Sounds dull. It saves months, because personalization more often flops on messy product data than on a weak algorithm.
What the word covers
It gets stretched a lot. In practice there are a few layers:
- Recommendations. Carousels, “complete the look”, email picks
- Search and ranking. The same query giving different results to a runner and a hiker
- Offers. Coupons, bundles, loyalty perks tuned per customer
- Merchandising. Banners, category order, product descriptions
- Service. Assistants that know the order history
Most shops do the first layer badly and skip the rest.
The cold start problem
A new visitor has no history, and a new product has no clicks. Classic recommenders choke on both. Newer systems read the session itself (what was clicked in the last two minutes), turn product text and images into embeddings, and add plain context like device, time and country. Someone who lands on a rain jacket and scrolls past three black ones has already said something.
What’s new on the market
A few of these changes are shipping. Others are still prototypes.
Shopping assistants
Amazon launched Rufus in 2024, a chat assistant that answers things like “what do I need for a first camping trip?” inside the app. Walmart followed with Sparky. Klarna made noise with an AI helper in customer support, and Shopify’s Sidekick works on the merchant side, helping write copy and read reports.
People type a sentence instead of fiddling with filters. Nobody enjoys ticking nine checkboxes for “waterproof, under €150, size 43”.
Try-on and visual search
Sephora’s Virtual Artist tested lipstick on a live camera feed years before generative AI. IKEA Place dropped a sofa into a living room through a phone in 2017. Google’s try-on in Search now shows clothes on different body types, and Google Lens or Pinterest Lens let a photo replace a query. Fashion is the obvious target, since wrong-size returns eat margins.
Agentic checkout
Odd one, but real. In 2025 OpenAI introduced Instant Checkout in ChatGPT with Etsy and Shopify merchants, built on the Agentic Commerce Protocol developed with Stripe. Google answered with its Agent Payments Protocol, AP2, and Perplexity has tested in-chat purchases with PayPal.
Think about it for a second. If a bot searches, compares and pays, who looks at the homepage? Retailers are starting to wonder whether bots need optimizing for as well as people. Clean feeds and real stock levels begin to matter more than a pretty banner.
Still in the lab
- Product pages rewritten for each visitor
- Voice agents that take a full order by phone
- Personal “shopping twins” trained on one customer’s taste and budget
- Smart mirrors and in-store assistants, tested by Zara and Uniqlo in flagship stores
Some will stick. Plenty won’t. Remember when every brand needed a Messenger chatbot in 2016?
Where it goes wrong
Personalization has a nasty failure mode: it gets creepy. The story everyone retells is Target, which, according to a 2012 New York Times piece, guessed a customer’s pregnancy from her purchases and mailed her baby coupons before the family knew. Details are debated, the lesson isn’t. Knowing something and acting on it are two different things.
Usual ways it breaks
- Bad data in, bad picks out: duplicate SKUs, missing attributes, stale stock
- Recommending what was just bought (a fridge, again)
- Echo chambers, where the shop keeps showing the same narrow slice
- Assistants that invent a size chart or a return policy
- A cookie banner nobody understands, feeding a profile nobody can see
Regulators are watching. GDPR already covers profiling, and the EU AI Act is phasing in, with transparency duties for chatbots and limits on manipulative design. A bot that pretends to be human is a legal headache as well as a PR one.
A quick test
Three questions for any personalization feature:
● Would the customer be fine seeing exactly why this was shown?
● Does it still work when the model gets it wrong now and then?
● Can a person step in when it fails?
One “no” and the feature isn’t ready.
Building it without burning the budget
Failed AI projects rarely die from a bad model. Data and process usually do the damage, a point Bintime makes on its own AI transformation page, and one that matches what practitioners keep saying.
An order that tends to work
- Audit the data. Product attributes, customer IDs across channels, order history.
- Rank use cases. Value against feasibility. Search relevance and product-data enrichment often beat flashy try-on.
- Prove one thing on real data. A contained pilot with success criteria written down first.
- Connect it to the stack. ERP, PIM, CRM, storefront. This unglamorous part decides everything.
- Scale with guardrails. Evaluation, human handover, audit trails.
The build-or-buy decision should depend on the use case: off-the-shelf solutions often work well for standardized processes, while custom development may be more appropriate when a workflow provides a competitive advantage. Ultimately, successful AI personalization depends on reliable data, thoughtful integration, and a clear understanding of customer needs. Retailers that prioritize practical improvements over technological hype are better positioned to deliver shopping experiences that build trust and long-term loyalty.
