What we learned building a voice-first storefront

We spent this year building a shopping experience with no search bar, no filters and no product grid. Customers talk to it instead.

It works. It also surfaced problems nobody warned us about. Here are the ones that mattered.

People do not know what to say

This is the big one.

A filter list, for all its faults, tells you what is possible. Size, colour, price: the options are visible. A blank prompt tells you nothing.

We watched first-time visitors arrive, look at the screen, and leave without asking anything. Not because the assistant was poor, but because they had no idea what kind of sentence it wanted.

Teaching that in the first few seconds turned out to be most of the design work. Example phrases help. So does showing what the catalogue contains, even though the whole point is that you no longer have to browse it.

The conversation has to carry context

A recommendation engine answers one question at a time. A salesperson remembers.

When someone says “I like the second one, does it come in a wider fit,” the assistant has to know which pair “the second one” is, and carry the size, occasion and budget already mentioned. Getting pronouns and references right across turns took far more work than the recommendations themselves.

This is the difference between a conversation and a series of queries, and customers notice immediately when it breaks.

Specifications are not enough

Customers close to a decision rarely ask about specifications. They ask “what is the difference between these two?”

That is a question about character, not attributes. Answering it needs the brand’s stories, not just its product data: why a boot is named what it is, who it was made for, what separates one brown from another.

We gave the assistant the brand’s heritage alongside its catalogue, and the answers changed completely. The recommendation stopped being a filter result and became a reason.

Some shopping has no goal

Not everyone arrives with a problem to solve. People browse. They enjoy scrolling with no intention of buying, and a conversation is poorly suited to that.

Any serious version of this has to keep a way to simply look. Removing navigation entirely is a purist position that costs you customers.

What we would tell anyone building this

The hard part is not the AI. Models are good enough.

The hard part is the first ten seconds, where a customer decides whether this is something they know how to use. Everything downstream depends on getting that right.

FAQ

What is conversational commerce?

Conversational commerce is online shopping where the customer describes what they want in their own words and the system interprets it, rather than selecting from filters and categories. It can work by voice, by text, or both.

Why do customers struggle with voice shopping interfaces?

A blank prompt gives no indication of what is possible, while a filter list does. Customers often arrive without a vocabulary for the system and leave without asking anything. Teaching that vocabulary quickly is the main design problem in the category.

What makes an AI shopping assistant different from a chatbot?

A chatbot answers questions. A shopping assistant carries context across a conversation, remembers what has already been said, and narrows a catalogue toward a decision. The difference shows up most when a customer refers back to something mentioned earlier.

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 retail brands. Its first live implementation is Travis, built for the boot brand Alvies.