August 6, 2026
AI shopping assistants improve product discovery, build baskets and handle post-purchase support. See where they add value, how to measure ROI and what to look for.
Amrita Bhambhani
~t.compress(q:75)~t.toFormat(f:webp)/cms_uploads/Ai-Shopping-Assistants-for-Ecommerce.jpg)
Tesco chose a simple place to begin with its AI shopping assistant: helping customers decide what to cook for dinner.
The retailer is trialling the assistant with around 280,000 colleagues ahead of a wider customer rollout. Every week, millions of customers know they need to buy groceries but have yet to decide which meals to make, which ingredients those meals require or what should go into the basket. That decision can take longer than the purchase itself, and most of it happens before the first product reaches the basket.
Tesco says its assistant will initially support meal planning, inspiration and basket building. Grocery is only one use case. The same idea works across almost every retail category. A customer shopping for a birthday gift, a new sofa or a waterproof jacket often knows what they need, but not exactly what they want to buy. A shopping assistant helps them get there faster by turning a broad request into a small number of relevant recommendations.
An AI shopping assistant is a customer-facing tool that shoppers can talk to in ordinary language. It reads what they are asking for, searches the retailer's catalogue, prices and stock for products that answer it, and stays involved through comparing, choosing and buying.
What an assistant can do falls into three levels: answering, advising and acting. Each one asks more of the retailer's data than the last, which is usually what decides how far a deployment gets.
Answering
The simplest job an assistant does is field the questions a customer would otherwise have to hunt through the site for. Does this dress run small, what separates two washing machines at the same price, what is the sofa frame actually made of, will it arrive before Saturday.
Questions like these account for most of the use these tools currently get. Walmart found that 81% of shoppers who have tried Sparky, its AI assistant, used it to check product details or availability before buying.
The answers are only as good as the information behind them. Product copy, specifications, reviews, delivery terms and returns policies determine whether the assistant can respond at all, and live prices and stock determine whether the answer is still true when the customer acts on it.
Advising
A customer looking for a new laptop rarely opens with "show me a 14-inch model with 16GB of RAM." They ask for something light enough to carry every day, with a battery that lasts through meetings and the journey home. Someone shopping for a wedding outfit talks about the occasion, the season and a budget before they mention a brand.
That is how people ask for advice, and a shopping assistant should answer in the same terms. It takes in the situation, fills the gaps with a few questions, and comes back with products that suit. It also explains why it chose them, which is what lets the customer weigh the options rather than simply accept them.
Nosto's research found that 69% of shoppers shown irrelevant suggestions abandoned the assistant and searched elsewhere, having already spent time explaining what they wanted.
Acting
Choosing a product is only part of the shopping journey. Customers also want things done. They want to reorder the coffee they buy every month, check whether a parcel will arrive before they leave for the weekend, update a delivery address or return a pair of boots that didn't fit. Simple enough to ask for, and people now expect to handle them in the same conversation.
For an assistant to manage that, it needs a connection to the retailer's systems and not only its product catalogue. Reordering, order tracking and returns all depend on it reaching purchase history, live order data and the returns process behind them. An assistant with the language ability to discuss any of this and no access to the data will sound helpful while getting things wrong. This is the part of the problem that conversational commerce platforms are built to handle.
Amazon's shopping assistant, renamed from Rufus to Alexa for Shopping in May 2026, does most of it. The assistant reorders previous purchases, tracks prices, buys automatically when a customer-defined price is reached, and helps manage deliveries, returns and refunds. Amazon says more than 250 million customers used it last year and that they were over 60% more likely to make a purchase during the same visit. That figure comes from Amazon itself and reflects an association rather than a proven cause.
Customers are generally comfortable letting an assistant do the legwork. They are far less willing to let it make decisions on their behalf. Research conducted by Logica Research for PayPal and Commerce.com found that 43% of UK shoppers were most concerned about AI making purchases without their approval.
Retailers already rely on site search, chatbots and recommendation engines to help customers find products. A shopping assistant takes a different approach. It starts by understanding what the customer is trying to achieve, asks questions where it needs more context and refines its recommendations as the conversation develops.
Consider a customer asking for walking boots for winter hiking, under £120.
Site Search | Chatbot | AI Assistant | |
How it interprets the request | Matches keywords in the search | Identifies the question being asked | Understands the customer’s goal and any constraints |
Follow-up questions | None | Limited to predefined questions | Asks for details synch as terrain, budget or preferred fit |
How recommendations are made | Based on keywords matches | Doesn’t recommend products | Recommends products based on the customer's needs and explains why |
Remembers earlier context | Starts each search from scratch | Keeps track of the current query | Remembers everything shared during the conversation |
Product and stock information | Depends on the latest search index | Usually separate from product and inventory data | Uses live product and inventory information when connected |
Explains its recommendations | No | No | Explains why each product has been recommended |
Tasks it can complete | None | Simple service requests, such as order tracking | Adds products to a basket, checks availability, tracks orders and starts returns |
When a person takes over | No handover | Transfers the conversation | Passes the full conversation and context to the next agent |
Categories where choosing is difficult
A dishwasher listing tells a customer about decibel ratings, place settings, energy classes and wash programmes. What most of them really want to know is whether it will be quiet enough for an open-plan kitchen and whether it will fit under the counter. Shopping for skincare, hiking boots or paint works the same way.
Narrowing the choice to two or three products doesn't necessarily make it easier. Specifications are spread across different pages, reviews are mixed, and the details that matter most can be difficult to pick out. An assistant can bring those comparisons together, explain the trade-offs and recommend the option that best fits what the customer has already said they're looking for.
Pour Moi, the UK lingerie and swimwear retailer, reported an 11% increase in conversion after introducing AI-powered search and shopping assistance, according to a case study from Findologic.
Building the basket
Some shopping trips are about completing a project rather than choosing one product: a week's meals, a bathroom, a nursery for a first baby.
Tesco's assistant, still in beta, builds the basket as a whole, suggesting what belongs together, checking products are available and keeping the total inside the budget.
Repeat purchases
Customers who buy the same cat food, coffee or washing powder every few weeks already know what they want. The decision has been made, so the assistant's job is to rebuild the previous order, suggest alternatives where something is unavailable and help the customer complete the purchase quickly.
If a substitution is needed, customers should always be able to review it, change it or decline it before the order is confirmed.
After the order has been placed
A customer might want to know where their parcel has reached, whether the delivery address can still be changed or how to return something that doesn't fit. Those are some of the most common reasons people contact a retailer after buying. When an assistant has access to order and fulfilment data, it can usually answer them in the same conversation, without sending the customer elsewhere for help.
Shopping beyond the retailer’s website
A shopping assistant doesn't always belong to the retailer. Customers are increasingly discovering products through ChatGPT, Gemini or Amazon's Alexa for Shopping rather than on a retailer's own website or app.
Google's Universal Cart, announced in May 2026, lets shoppers buy from multiple merchants across Search, Gemini, YouTube and Gmail. ChatGPT recommends products directly within conversations, while Amazon's Alexa for Shopping does the same inside Amazon.
Being recommended in those experiences comes down to product information. Structured attributes, accurate stock data and clear descriptions are what let an external assistant work out what a retailer sells and put it forward when it fits what the customer asked for.
Almost two-thirds of consumers across the UK and Ireland, 64%, say they want retailers to use AI to improve how they shop, according to CI&T's survey of 2,000 people. Yet in the same research, 68% could not name a single AI shopping experience that had impressed them. Plenty are using these tools regardless. Adyen's 2026 retail report found UK adoption doubled in a year, from 12% to 28%, and CI&T found 61% had used AI while shopping, with 53% doing so regularly.
ACI Worldwide asked YouGov to survey 2,080 UK adults in June 2026 and found that half trusted AI to find the best available price. Only 19% trusted it to follow their rules and make everyday purchases. By comparison, 55% would trust a human adviser, while just 18% believed AI would act in their financial interests. The same survey found that 60% would stop using an AI shopping assistant after a single mistake. Retailers can automate much of the shopping journey, and the purchase itself should still be confirmed by the customer.
A shopping assistant can answer thousands of questions without improving the business. The only way to know is to compare what changed after it was introduced with what was happening before, which means recording conversion, average order value, returns, customer service contacts and cost per contact while the assistant is still switched off.
Conversation volume can be misleading. A higher number might mean customers are finding the assistant useful, or it might mean they're struggling to get the answers they need.
Imagine a homeware retailer running a three-month trial. Half of its visitors shop with the assistant and half without it, with each group generating 50,000 sessions a month. The figures below are illustrative.
Without assistant | With assistant | |
Conversion rate | 2.0% | 2.3% |
Orders | 1,000 | 1,150 |
Average order value | £80 | £80 |
Revenue | £80,000 | £92,000 |
Returns | 25% | 22% |
Revenue retained after returns | £60,000 | £71,760 |
In this example, the assistant generates an additional £11,760 in retained revenue each month. At a gross margin of 45%, that contributes around £5,300. Add a further £3,200 saved through fewer customer support contacts, and the total monthly benefit comes to roughly £8,500.
Against that sits the cost of running the assistant: £7,500 a month covering platform fees, implementation and integration spread across the year, and internal time keeping product information and prompts up to date.
ROI = (£5,300 + £3,200 − £7,500) ÷ £7,500 = 13%
A 13% return is positive, but it doesn't leave much room for error. A smaller improvement in conversion or a higher return rate would quickly change the outcome.
Adobe's consumer research found that 69% of shoppers said they were less likely to return an item bought with help from an AI assistant, although the survey was conducted in the United States and should be treated as directional. If conversion improves but returns rise with it, the assistant may simply be moving costs from one part of the business to another rather than creating new value.
The way the comparison is made makes a difference too. Customers who choose to use an assistant are often more engaged than those who do not, so comparing those two groups almost always overstates the impact. A genuine A/B test or a phased rollout provides a much more reliable picture of what has changed.
Testing is also part of the UK's regulatory guidance. In March 2026, the Competition and Markets Authority set out that businesses should test systems before they go live, monitor how they perform afterwards and make sure important decisions are still reviewed by a person. Even when the technology comes from a third party, the retailer remains responsible for what the assistant does.
Product data: Everything starts with the catalogue. The assistant needs live prices and stock, and it needs to know what to do when that connection drops. It also needs enough product information to answer questions about size, fit and compatibility.
Recommendations: Understand what influences the recommendations. Can sponsored or higher-margin products be prioritised? What happens when the assistant doesn't know the answer? And if it builds baskets, how does it decide what belongs there?
Customer approval: Decide where the assistant stops and the customer takes over. Some actions can take place automatically but purchases should still be confirmed by the customer.
Handover: The context should move with the conversation. When the assistant passes a customer to a colleague, they shouldn't have to explain everything again.
Costs: The platform fee is only part of the cost. Implementation, integrations, usage costs, renewal terms and the ongoing work of keeping product information and prompts up to date all need to be included.
Evidence: Look beyond case studies. Ask how the results were measured. A controlled A/B test tells you far more than a before-and-after comparison, and measures such as conversion, returns and cost per resolution are more useful than conversation volume.
Customers experience a shopping assistant as a conversation. Retailers experience it as product data, inventory, commerce systems and customer service working together. That's what determines whether the conversation is genuinely useful or simply another way to search.
Fynd Kaily connects those parts of the retail stack in one platform, so every conversation runs on current information.
No. Shopify, Magento and custom builds can all support an AI shopping assistant. The assistant needs access to your product catalogue, pricing and inventory through APIs. When choosing a supplier, ask which systems they have already integrated with. That gives a clearer picture than the platforms they say they support.
An assistant can only answer questions using the information it has. If customers ask about size, fit or compatibility and those details are missing, the product data needs attention. Many retailers start with the categories where product information is already strong, then improve the rest over time.
Run it until both groups have generated enough orders to judge the results with confidence. The right timeframe will vary by retailer, depending on traffic and order volume. Try to avoid measuring the pilot entirely during a sale or promotional period, as customer behaviour is often different.
The ICO's January 2026 report on agentic AI highlights four areas: transparency, purpose limitation, data minimisation and customer control. Customers should know when they are interacting with AI, the assistant should only collect the information needed for the task, and people should be able to view or update any personal information it stores.
Discover how Instaghar partnered with Fynd to build a hyperlocal grocery platform with unified ecommerce, multi-store fulfillment, intelligent order routing and Transport Management System.
AI assistants already compare price and stock before a customer visits a website. See what agentic commerce means for retailers, and what to prepare for.
Discover how Nanobot used a unified B2B ordering platform to manage order management, improve inventory visibility and create a scalable foundation for future growth.
Fill out the form
Share your contact information to get started
Speak to an expert
A member of our sales team will get in touch with you