How on-site chat features are turning into full AI assistants

Shoppers who talk to Rufus (now Alexa for Shopping) before buying are 60% more likely to complete a purchase, according to CEO Andy Jassy.

That's not a one-off feature launch.

The same chatbox that used to answer a single product question on Amazon, Walmart, Home Depot, and elsewhere keeps getting handed a bigger job every few months. This is about where that's headed, and what it's actually moving once it gets there.

How four retailers expanded the same tool

Home Depot's Magic Apron launched in March 2025, answering how-to and product questions on Home Depot's product pages. EVP of Merchandising Billy Bastek said that in May it was "contributing to growth in online conversion."

By January 2026, it was able to understand a whole project, not just one item: ask it what grout works with glass tile, and it recommends the product - and in stores running the pilot, it tells you the exact aisle to find it in.

Klarna's assistant launched in 2024 to handle customer service: refunds, payment questions, account issues. It now has a full shopping mode. Search, compare, and get recommendations across a database of millions of products from thousands of stores.

Amazon shipped more than 50 technical upgrades to Rufus in November 2025: account memory, automatic cart-adds, price tracking, auto-buy at a target price. That same quarter, on the Q4 earnings call, they started calling it "the AI agent."

In May, they retired the Rufus name entirely, folding it into Alexa for Shopping, now the default AI layer built into Amazon's own search bar for every signed-in US shopper. I counted it on Amazon - from the search box to the first product it recommends by name is seven prompts now. See on LinkedIn →

Walmart Sparky's early use, in CEO David Guggina's words, centered mostly on general merchandise discovery. Guggina told analysts on Walmart's Q1 FY27 call that usage has since expanded into everyday essentials, food, and personalized reordering, with Sparky now live across Walmart's app, website, and stores.

What it moves: Revenue

Amazon tracks what it calls "downstream impact".

A seven-day rolling window that credits Rufus for a purchase even if the shopper doesn't buy the same day they asked the question. Under that model, shoppers who engage with Rufus are 60 percent more likely to complete a purchase than shoppers who don't.

Walmart's version of the same finding is that shoppers who use Sparky carry a basket about 35 percent bigger than shoppers who don't. By the following quarter, Walmart added that weekly active users on Sparky were up more than 100 percent year over year, and units purchased through the assistant had more than quadrupled.

Both retailers are pointing at the same thing - a shopper who gets to describe what they actually want buys more than a shopper stuck filtering through a grid.

What it moves: Retention

Zalando's Size & Fit AI uses real body measurements from more than a million customers to help shoppers pick the right size before they check out.

Zalando reported the feature has cut size-related returns by more than 8 percent. A separate feature, its foundation-model-powered matchmaking, drove a 13 percent increase in items added to bags in the same results.

Getting the right size or the right match on the first try is the same thing as giving a shopper a reason to come back. An assistant that prevents a bad purchase is doing the same job as one that closes a good one, just earlier in the relationship.

What it moves: Engagement

Zalando's Assistant, which now handles both fashion and beauty recommendations in a single conversation, went from 6 million users in 2025 to 10 million in Q1 of 2026 alone.

Wayfair built a similar bet around visual browsing instead of a chat window.

The tool generates photorealistic, shoppable room scenes from a style prompt and feeds those results into the app's Discover tab.

Constellation Research, an analyst firm tracking Wayfair's AI rollout, found that shoppers shown Wayfair's AI-curated design recommendations were 33 percent more likely to add an item to cart or complete a purchase. CTO Fiona Tan told investors on the company's Q3 2025 call that the updated tab had "boosted visit duration and conversions".

Neither of these numbers is about a single transaction.

They're about a shopper choosing to open the tool again.

Why some of these work better than others

Dozens of individual studies point to the same design principles.

A 2026 meta-analysis pooling 468 effect sizes from 82,751 participants across 262 studies found that AI agents move a shopper's stated intent to buy more reliably than they move how much the shopper likes the agent or the brand behind it. Intent is the metric that predicts revenue. It's also the one AI assistants are best at moving.

A separate meta-analysis of 32,172 consumers across 122 studies (Journal of Retailing) found that not all recommendation styles perform equally. Interactive, conversational formats outperform static, one-shot recommendation lists. Recommendations a shopper actively asks for outperform ones pushed at them unprompted.

When put together, the version of this that works answers a question the shopper actually typed, not the one that pops up uninvited with a suggestion nobody asked for.

What every assistant above depends on

None of the assistants above are smart on their own.

Rufus reads Amazon's product catalog, reviews, and Q&A threads to answer a question, nothing else. Magic Apron is grounded in Home Depot's product data and project knowledge base. Zalando's Size & Fit AI needs real, structured measurement and material data attached to every product to make a size call worth trusting.

Swap in a catalog full of thin, inconsistent product data, and the assistant still generates an answer. It just won't be as good. The structured data behind it is the key.

Install Atomz on Shopify - try it out for free - the AI Assistant runs on the same enriched catalog, so it recommends what you actually sell instead of guessing.

Book a strategy call - if you want to see a demo on your own store.

- Ankit