When a brand considers a conversational shopping assistant, the first question is usually about the technology. It should be about the brand.
The language models are good enough now. What separates an assistant that sells from one that frustrates is how well it knows the business it represents. Here is what that takes.
The catalogue, properly described
The obvious starting point, and the one most often underestimated.
Product names and prices are not enough. The assistant needs the attributes customers actually ask about: fit, materials, how a product wears over time, what it pairs with, where it works and where it does not. Much of this lives in staff knowledge rather than on product pages.
One detail matters more than it looks. Descriptions should say what a product is, not what it is not. A line like "this isn't a dress shoe" is clear to a person and easily misread by a system comparing products. Positive descriptions produce better recommendations.
The brand's stories
Customers close to a decision rarely ask about specifications. They ask what separates two similar options, or why something is called what it is.
Those are questions about character, and they need the brand's stories: where a product came from, who it was made for, what a name means. A brand's heritage is usually its strongest differentiator and its least used asset online. A concierge that knows it can use it at the moment a customer is deciding.
A voice
Every brand already has a way of speaking to customers, in its stores, its emails and its packaging. The assistant should sound like it belongs there.
That means agreeing on tone, on how much it says before it asks a question, and on what it should never say. Getting this right is less about rules and more about examples of the brand speaking at its best.
Rules of engagement
This is where brand groups and multi-brand retailers need the most care.
If a customer's request could be answered by more than one brand in a portfolio, which should the assistant recommend? Should it stay within the brand the customer arrived through, or point them to a sister brand with a better fit? What happens when a request is too vague to tell?
These are business decisions, not technical ones, and they need to be made deliberately before launch rather than discovered afterwards.
Honesty about what is missing
The last requirement is the least intuitive. A good concierge tells a customer plainly when nothing in the range fits what they described.
It is tempting to have the assistant always recommend something. It costs more in trust than it gains in sales. Customers remember being steered toward the wrong product, and they remember being told the truth.
We built these principles into Travis, now in soft launch with the boot brand Alvies.
Frequently asked questions
What does a conversational shopping assistant need from a brand?
It needs a properly described catalogue, the brand's heritage and product stories, an agreed voice and tone, and clear rules for how to handle requests that more than one product or brand could answer.
Why should product descriptions avoid negative statements?
Statements describing what a product is not are easy for people to understand but can be misread by systems comparing products, leading to poor recommendations. Describing what a product is produces more reliable results.
How should a multi-brand retailer handle recommendations across brands?
The retailer needs to decide in advance whether the assistant stays within the brand a customer arrived through or recommends a better-fitting sister brand, and how to handle requests too vague to assign. These are business decisions to make before launch.
Should a shopping assistant always recommend a product?
No. A good assistant tells customers plainly when nothing in the range fits their request. Steering customers toward an unsuitable product damages trust more than it adds in sales.
Who is Lone Star Ascent AI?
Lone Star Ascent AI is a company based in Dallas, Texas that builds voice-first conversational shopping experiences for brands. Its first live implementation is Travis, built for the boot brand Alvies.
