Zum Inhalt springen
Individually trained
E-Commerce

Guided Selling in Chat: Finding the Right Item

Guided selling in chat: the AI assistant asks about needs, narrows the range and recommends the right item. Fewer abandonments and fewer returns.

12 min read Guided SellingProduktberatungOnlineshopConversion

In a physical store, advice is a given: someone asks what the product is needed for, listens and in the end puts exactly the right thing on the counter. Online this step is almost always missing. The visitor stands alone in front of a range with hundreds or thousands of items, a search box and a handful of filters, and has to work out what fits them on their own. For many this ends in abandonment: around 31 percent (Baymard Institute) of product searches end with test participants not finding what they were looking for or getting so frustrated that they give up. An AI chat assistant can close this advice gap. It asks about the needs, narrows down from the existing range and finally recommends a concrete, fitting item, just like good advice in a store. This article shows how guided selling in chat works, why it differs from a mere product display, how it reduces abandonments in large catalogues and wrong purchases, and where its limits lie.

Guided Selling in Chat: From Catalogue to Matchshop-example.comWhat will the bike mainlybe used for?Commuting in the city, dailyBudget up to 800 euro,low maintenance?Yes, as little effort as possibleYour matchCityFlow 3 ComfortLow-maintenance, hub gears749.00 euroAdd to cartThe best match from 3 suggestions for youWrite a message...From Catalogue to Match1,240 items in the range180 fit the use24 within budget3 favourites1 recommendationWhy chat advice works31%of product searches end without a find(Baymard Institute)71%expect advice that fits them(McKinsey)40%more revenue from personalization(McKinsey)Ask about needs, narrow the range, recommend the right item

When the Large Catalogue Becomes a Barrier

A large range is both a promise and a problem. The more items a shop carries, the harder the choice becomes when the visitor is left to their own devices. Research shows how often this goes wrong: in around 31 percent (Baymard Institute) of product searches test participants fail to find what they are looking for or abandon in frustration, and in around 65 percent (Baymard Institute) of cases it takes more than one search attempt, often three or four. No wonder, because 56 percent (Baymard Institute) of the sites tested have mediocre or worse search usability, and 67 percent (Baymard Institute) of leading sites also score only mediocre to poor on navigation. Anyone who does not already know exactly what they are looking for often fails to find their way through a large catalogue.

Behind this lies a familiar pattern: too much choice without guidance paralyses the decision. A search box assumes someone knows the right term; a filter assumes they master the right criteria. Whoever has only a goal but not the jargon faces a wall of tiles. That is expensive, because German online retail turned over around 104 billion euro (IFH Köln) in 2024, and for 2025 IFH Köln considers an online share of about 21 percent (IFH Köln) of total retail possible, that is every fifth retail euro. In this market it is not only the range that decides, but the ability to guide the visitor to the right item. This is exactly where the shop assistant comes in.

Briefly explained: guided selling

Guided selling describes advice that leads the customer to the right product through targeted questions, instead of sending them alone through catalogue, search and filters. In the chat an AI assistant takes this on: it clarifies need, budget and preferences, narrows the range step by step and ends in a concrete recommendation. The goal is not to show more, but to show the right thing.

Advice Like in a Store, Just in Chat

Advice in a store does not begin with a product list but with a question. This is exactly how a guided selling assistant works in the chat: it asks two or three targeted follow-up questions before it shows anything. What is the product needed for, what budget, what experience, what occasion. From the answers it forms a small, curated selection rather than a results list with three hundred entries. The visitor does not have to know the jargon of the range; they describe their goal in their own words, and the assistant translates that into the matching product attributes. That turns a helpless search into a guided conversation that ends in a manageable decision.

That customers expect this kind of approach is documented. 71 percent (McKinsey) of consumers expect companies to deliver personalized interactions, and 76 percent (McKinsey) get frustrated when they do not. An assistant that listens and responds to the individual need meets exactly an expectation that online retail often disappoints. What matters is that the personalization arises from the conversation itself and not from a secret profile: the assistant asks what it needs to know instead of tracking visitors across pages. Which questions it asks and in what order can be defined precisely through tool control.

Purpose of use

What is the product needed for and in what context? This question decides everything that follows.

Budget

A price range narrows the selection quickly and prevents recommendations that do not fit from the start.

Experience level

Beginners and experts need different products; the assistant adjusts the selection accordingly.

Preferences

What matters, what is out of the question? Exclusions are often as helpful as wishes.

Compatibility

Does the accessory fit the main product, the size fit the need? The assistant checks this against the real attributes.

Occasion and urgency

A gift for the weekend has different requirements than a long-term purchase without time pressure.

From Need to Recommendation: How the Dialogue Narrows

The core of guided selling is narrowing down step by step. The assistant starts with the whole range and makes it smaller with every answer, until one or two fitting items are left at the end. Twelve hundred items may become a few dozen after the question about use, a handful after the budget, and exactly the match that fits after the preferences. This funnelling takes off the visitor the work they would otherwise have to do themselves with search and filters, and it does not feel like a machine but like a conversation.

  1. Clarify the need: what is the product needed for, in what context?
  2. Set the frame: capture budget, size, quantity or technical requirements.
  3. Note preferences: what matters, what is explicitly out of the question?
  4. Narrow the range: only the items that truly fit remain from the catalogue.
  5. Show the recommendation: one to three matches as a product card, each with a short reason.

At the end of the funnel there is no list of links but a concrete recommendation. The assistant shows the match as a product card with image, price and buy button right in the conversation and gives a one-sentence reason why it fits. Someone hesitating between two models gets them side by side as a direct comparison. These cards are created from real catalogue data, so they are clickable and current instead of a rigid advertising image. How these product cards are built is described in detail on the page about product cards in the chat.

One match beats a list of results

Guided selling is not measured by how many products the assistant shows, but by how accurately. A single, well-justified recommendation leads to a purchase more often than twenty tiles the visitor has to dig through themselves. Less, but more fitting, is the better choice here too.

Guided Selling Is Not a Product Card

A product card shows a product; guided selling leads to a product. The difference sounds small but is decisive. When the assistant displays fitting items as cards in response to a clear question, that is active product display, as the article on product cards and cart in the chat describes. Guided selling starts one step earlier: the visitor does not yet know exactly what they want. The assistant works that out in the dialogue first, before any card appears. The product card is thus the result of the advice, not its beginning.

Both building blocks belong together and mesh. The advisory conversation narrows and justifies, the product card makes the result visible and clickable, and a cart assistant puts the recommended item straight into the cart on request. Whoever leaves out the advice and only shows cards sells past the undecided; whoever only advises without making a tangible offer in the end lets the purchase intent fizzle out. The art lies in the transition: first listen and narrow, then show and lead to completion.

AspectPure product displayGuided selling in dialogue
Starting pointThe visitor already knows the product soughtThe visitor has only a goal, but no product
FlowThe card is shown on requestThe need is clarified first in the conversation
SelectionA list of results to browseOne to three justified matches
Role of the visitorChooses from the offer themselvesIs guided and decides in the end
Fits especially withClear, already formed purchase intentA large range and real need for advice

Fewer Abandonments, Fewer Returns

Guided selling works at two expensive points at once: at the abandonments before the purchase and at the returns afterwards. Whoever does not find what they are looking for in a large catalogue leaves the shop, often without ever putting an item in the cart. And even those who make it to the cart often abandon again: around 70 percent (Baymard Institute) of online shopping carts are not carried through to purchase. An assistant that guides the selection already lowers the first hurdle and can, once the cart is filled, clear open questions about shipping and cost before the visitor leaves. How to catch this late phase is shown in the article recovering cart abandonment with a chat.

The second effect concerns returns. Many returns arise because the expectation of a product does not match reality: the wrong size, the unsuitable model, the missing accessory. An advisory dialogue clears exactly that before the purchase. The assistant asks about fit, purpose of use and requirements, reads size tables and compatibility from the catalogue and points out differences before the order is placed. Whoever gets the fitting item from the start sends less back. From our own projects (project experience) it is often exactly the cases of missing advice that later come back as returns. A fixed return rate cannot be assured, because it depends heavily on the range, but the cause of many wrong purchases, the missing advice, is tackled at its root.

The most expensive purchase is the one that comes back. Advice before the order is cheaper than any return after it.

From e-commerce project practice

On Real Catalogue Data Instead of Guesswork

A recommendation is only as good as the data it rests on. A guided selling assistant does not guess but reads the shop's real catalogue: titles, descriptions, variants, prices, stock levels, categories and technical attributes, plus shipping rules and the content of the website. If someone asks about a specific size or property, the assistant checks the current stock instead of recommending a sold-out item. For Shopware in the freely available Community Edition this happens through the existing interfaces, so content does not have to be maintained twice. Where this knowledge comes from and how it is structured is described in the article building a knowledge base: which content belongs in it, complemented by the page on the knowledge base.

Besides reading comes acting. Through clearly delimited functions, known technically as function calling, the assistant can add the recommended item to the cart, choose a variant or lead to a fitting page. Each of these actions is tied to a permission, so the assistant only does what is intended. Which functions make sense and how far they reach can be extended all the way to custom functions. When the advice reaches a limit, for example with a very specific requirement, the assistant hands over to a human, together with the conversation history as context. How this handover succeeds smoothly is described in the article AI assistant: handover to staff.

  • Clarify the need through targeted follow-up questions instead of letting the visitor guess
  • Narrow the range from real catalogue data, not from a rigid script
  • Check availability and variants before an item is recommended
  • Show one to three justified matches as a product card, not an endless list
  • Add the recommended item to the cart with permission
  • Hand over to a human for very specific questions

Personalization With Measure and Data Protection

Good advice pays off. Companies that personalize consistently generate around 40 percent (McKinsey) more revenue from those activities than average players, and personalization typically lifts revenue by 10 to 15 percent (McKinsey). What matters, though, is the measure. An assistant that pushes three additional products at every opportunity achieves the opposite and comes across as intrusive. Guided selling stays advice: it leads to the fitting item and offers sensible accessories when they really fit, instead of inflating the cart at any price. A fitting recommendation is experienced as service, five arbitrary ones as advertising.

Because the assistant processes information in the conversation, data protection is part of the design from the start. The personalization arises from what the visitor tells it themselves, not from a secret profile across many visits. The assistant asks only what is needed for the recommendation, does not track anyone across pages and collects no data on stock. Hosting and processing take place in Germany, complemented by a data processing agreement, a deletion concept and clear data sovereignty. Details are set out on the page about privacy and hosting. Especially in German retail, this visibly economical handling of data is not an obstacle but a vote of confidence that counts at the moment of the purchase decision.

Advise, do not pressure

Define how many questions the assistant asks before it recommends, and when it makes an additional offer. Two or three targeted follow-up questions reach the goal; an interrogation puts people off. A justified recommendation and fitting accessories at the right moment work as advice. If every click is buried in suggestions, the impression tips into the promotional. Less, but more fitting and at the right time, is almost always the better choice.

What the Advice Delivers Becomes Measurable

An advisory dialogue is only as good as what you learn from it. The conversation analytics evaluate at which point conversations break off, which questions come up most often before a purchase, which recommendations are accepted and where the assistant lacks knowledge. From these patterns come concrete improvements: a product text that keeps triggering follow-up questions is sharpened; a variant that is often searched for but rarely found moves to the front; a recommendation that is never accepted disappears. This improves not only the assistant but the whole path to purchase, because it becomes visible where visitors get stuck. How to use this evaluation is explored in the article analysing and optimising conversations.

For all its impact, an honest expectation is part of it. An assistant can err, which is why it ties its recommendations to its own catalogue and knowledge data, hands over to a human in cases of doubt and is evaluated continuously. A specific increase in revenue, conversion or return rate cannot be assured, because it depends on range, prices, target group and competition. What can be influenced are the factors that demonstrably work: less friction in the selection, advice that fits the need and clear recommendations at the right moment. Because a growing share of retail runs on small screens, where search and filters are especially tedious, a guiding conversation pays off particularly on mobile. How such an assistant works in an online shop in practice is shown on the corresponding industry page.

This article is based on data from: Baymard Institute (product search, navigation and cart abandonment in online retail), McKinsey (expectations of personalization and its revenue impact) and IFH Köln (revenue and online share in German retail) as well as our own projects. The figures cited can vary depending on range, prices and target group; entries marked with (project experience) are based on our own projects. A specific increase in revenue, conversion or returns cannot be assured.