How to build an AI configurator for a technical category
From the right questions to recommendation logic and a value calculator. The 4 layers of a configurator that guides, plus the Websem framework for a category with complex decisions.
An AI configurator isn't just a form with a chatbot on top. The difference between one that guides and one that frustrates lies in four layers, built in order: what you ask, how you recommend, what value you demonstrate, and how you move to action. Skip one and the configurator becomes a cute gadget that doesn't sell.
This article maps out the four layers and the framework we apply at Websem to build configurators that produce results — with Haier AC as a recurring example.
The 4 layers of a configurator that guides
- 01
The right questions
You start from the customer's real confusions, not the product's specs. Simple questions, in their own language (“what area do you want to cool?”), not technical jargon. The best source: the questions the sales team hears most often. They're already validated by reality.
- 02
Recommendation logic
The answers translate into technical criteria and get matched to your products. Everything here depends on the quality of your product data: a configurator built on messy data recommends the wrong thing, confidently. Product logic (rules, matches) ensures correct, consistent recommendations.
- 03
The value calculator
You turn the abstract decision into a concrete number: yearly savings, total cost over 5 years, ROI. The customer sees what they gain, not just what they pay. The calculator moves the conversation from cost to value — and it's often what keeps the customer in the configurator long enough to decide.
- 04
The bridge to action
The recommendation has to lead somewhere: cart, quote, booking, or hand-off to a human. A configurator that recommends and then leaves the customer hanging misses its point. The bridge is designed from the start, plus escalation to sales for special cases.
AI for interaction, logic for correctness
The most common confusion: a configurator is either “just a chatbot” or “just a smart form.” It's neither, and both. AI does the interaction part: it understands the question in natural language, carries the guiding conversation, explains why each step matters. Product logic does the correctness part: the rules, matches, and calculations that guarantee the recommendation is valid, not just plausible.
The balance between the two is what separates a configurator that sells from one that impresses in the demo and fails in production. Too much AI without logic = inconsistent recommendations. Too much logic without AI = a rigid form the customer abandons. You build both, deliberately.
And, like any AI system that answers from your data, a configurator needs a clean, up-to-date product data foundation. Stale embeddings or wrong specs produce wrong recommendations — which is why the data foundation matters just as much as the interface.
For the data foundation a configurator rests on, see Data Foundation; for the conversational side, the Chatbot AI hub.
How to build it: a 4-step framework
- 01
Collect your customers' real questions
From the sales and support teams. The most frequent confusions are exactly what the configurator needs to solve — already validated.
- 02
Map the answers to your product data
Translate the simple questions into technical criteria and make sure your product data is clean and up to date. Without that, the recommendation is unreliable.
- 03
Add a value calculator
Savings, total cost, ROI — a concrete number that moves the decision from price to value. This is often what keeps the customer engaged.
- 04
Build the bridge and the escalation
Cart, quote, booking — plus hand-off to a human for special cases. The configurator is the start of a conversion, not the end.
Frequently asked questions
What does a good AI configurator start with?
With the questions the sales team already gets. Your customers' most frequent questions are exactly what the configurator needs to solve — and they're already validated by reality. The common mistake is to start from the product's technical specs; the right move is to start from the customer's real confusions and translate them into simple questions, in their own language.
How does the configurator know what to recommend?
From your product data, structured as recommendation logic. The customer's answers (surface area, use, budget) translate into technical criteria (capacity, class, configuration), and the logic matches them to the right products. The quality of the recommendation depends directly on the quality of your product data — which is why the data foundation matters just as much as the interface. A configurator built on messy data recommends the wrong thing, confidently.
Why does a value calculator matter?
Because it turns an abstract decision into a concrete number. A customer who sees “you save X lei a year” or “the total cost over 5 years is Y” decides far more easily than one who is shown only the price. The calculator moves the conversation from cost to value — exactly what a good salesperson does. At Haier AC, the savings calculator was a central part of what kept customers in the configurator for 3-5 minutes.
What happens after the configurator recommends?
You build the bridge to action: add to cart, generate a quote, book a consultation, or hand off to a human. A configurator that recommends and then leaves the customer hanging misses its point. The bridge has to be designed from the start — the configurator isn't a goal, it's the beginning of a conversion. For cases that exceed the automated logic, escalation to the sales team picks up the thread.
How much of this is AI and how much is classic logic?
The combination. AI excels at understanding the customer's question in natural language and holding a guiding conversation; product logic (rules, matches, calculations) ensures the recommendations are correct and consistent. A good configurator is neither “just a chatbot” nor “just a smart form” — it's AI for interaction, on top of solid product logic. The balance between the two is what makes the difference.
Conclusions
An AI configurator is neither magic nor a repainted form. It's a system with four layers — questions, recommendation, value, action — where each rests on the next. Build all four and you get a digital consultant that guides the decision; skip one and you get a gadget the customer abandons.
The key, as with any useful AI system, is balance: AI for interaction, solid logic and clean data for correctness. The question you start with isn't “what technology do we use?”, but “what are our customers' most frequent confusions — and how do we guide them to a decision?”
Dan Cristian Alexandrescu is the founder of Websem, an agency that builds AI platforms and systems for serious business. The Websem team has delivered AI configurators and calculators for categories with technical decisions — including Haier AC România and Eurial Selection.
Want a configurator that guides, not one that frustrates?
30 minutes in which we map your customers' key questions and how a configurator would guide them — plus 3 concrete actions. No strings attached.