An AI assistant on an e-commerce site: instant answers to product questions

“I’m looking for a 10k 0805 resistor, show me what’s in stock.” In a store that sells technical products, questions like this arrive every day and most do not fit in a search box. An e-commerce AI assistant is a chat layer that answers customers’ product questions with the store’s catalog and stock data and, when needed, routes them to a quotation or to a sales representative.

In this article we explain how such an assistant differs from a search box, how a conversation progresses and what preparation is needed for the assistant to be useful. We give our examples through the AI product assistant of YazSis Smart E-Commerce.

How does an assistant differ from a search box?

A search box matches the typed word against the catalog. A customer, however, is often describing a need rather than a word: a particular technical specification, a budget limit, a delivery date. The assistant tries to understand this need within the conversation and bases its reply on current catalog data.

• Understanding the need: a part number, brand, technical criterion, budget and delivery expectation can be given in a single sentence.
• Answers tied to catalog and stock data: the answer is based on the store’s own product information; instead of general internet knowledge, your catalog and your stock are what is discussed.
• Technical comparison: several products can be placed side by side, feature by feature; the customer sees on a single screen how each product differs and in what respect.
• Collecting quotation and sales requests: if the customer is making a corporate purchase, quantity and lead-time details are collected within the conversation and the sales team receives a ready request.

Take the example sentence above. The customer gives two separate criteria: the value of the resistor and the package size. The third expectation is that only products in stock be shown. A classic search leaves it to the customer to turn these three pieces of information into separate filters; the assistant reads the sentence in one go and maps it to the relevant fields in the catalog. That way the customer reaches the comparison screen directly instead of browsing a category tree.

How does a conversation progress?

1. The customer writes the question in their own words: product, feature, budget or delivery. They do not have to find the right keyword.
2. The assistant checks stock, price and technical specifications in the catalog and works out which product group the question concerns.
3. It presents suitable options in comparison; the customer narrows the criteria, states a budget or asks a follow-up question.
4. The customer chooses a product or asks for a quotation for a corporate purchase.
5. At the point where the assistant cannot answer or the customer wants to talk to a person, the conversation is handed over to the live sales team together with a summary.

The last step matters. The customer does not have to explain the same question again; the sales representative takes over the conversation with its summary and goes straight to the solution.

A good assistant is not one that is ready to answer every question, but one that is ready to hand over to the right person where it cannot.

The assistant’s accuracy depends on the quality of the catalog

An AI assistant cannot be more accurate than the information in the catalog. If product names are inconsistent, technical specification fields are empty or stock information is delayed, the assistant will also give incomplete or wrong answers. That is why an assistant project is really a catalog clean-up project first, and that work produces value for search, filters and SEO even without an assistant.

• Product name, brand and model information should be entered in a standard format.
• Technical specification fields should be filled in by category; the criteria customers ask about should find their counterpart in the catalog.
• Stock and price information should be kept in sync with the ERP; an answer given with outdated information is worse than a wrong answer.
• Variants, alternative products and complementary products should be linked to one another.

Limits and handover to a person

AI can misunderstand or answer with incomplete information. That is why the assistant does not close matters that require a decision, such as price negotiation, a special quotation or a return, on its own; it passes them to the sales team. Which topics are handed over is decided up front according to the business’s sales model. It should be remembered that everything the assistant says can be read as a published commitment; for binding topics such as price and delivery, a setup that sticks to current catalog information and hands over when it is unsure should be preferred.

How do you measure success?

• Questions left unanswered and handed over: if the same question is handed over again and again, there is missing information in the catalog or a rule not defined for the assistant.
• Quotation and sales requests coming from conversations: is the assistant really bringing the sales team qualified requests, or is it producing noise?
• The product features customers ask about most and the products they compare; these become input for the storefront, campaigns and the content plan.
• The sales team’s satisfaction with the conversations they take over: does the conversation summary save time, or does the representative still ask from the beginning?

An AI assistant can also help prepare product content and SEO text; but the first step is being able to answer the customer’s question with current data. If you would like to discuss an assistant setup that suits your store, you can start from the Smart E-Commerce page.