6 min read
AI “digital twins” of your customers are a genuinely clever research tool, and almost no solo business should use them to make decisions yet. Five real conversations with five real customers will tell you more, cost you nothing, and build the relationships a one-person business runs on.
That is my position. Here is why, starting with the news that prompted it.
What Listen Twins are, and why they made headlines
On October 27, 2026, the AI research company Listen Labs launched Listen Twins. The idea: Listen Labs interviews real people in a target audience, then builds a one-to-one AI model of each person from those interviews. You can then put questions to a panel of 100 or more of these twins and get answers with reasoning, quotes from the original interviews, and confidence scores. The company says its twins matched real answers 93% of the time in its own testing.
This is not a fringe experiment. Salesforce signed a definitive agreement to acquire Listen Labs earlier in October, in a deal press reports put at about $2 billion, with plans to bring the technology into its marketing and service products. When a company that size bets on “synthetic customers,” the idea will trickle down to small business tools within a year or two. So it is worth deciding now what you think of it.
The honest case for AI customer twins
I want to be fair, because the idea is not silly.
- Speed. Real research takes weeks of recruiting and scheduling. A twin panel answers in minutes.
- Scale. Asking 100 simulated customers is cheaper than asking 100 real ones.
- Grounding. These twins are built from actual interviews, not invented personas, which is more rigorous than asking a chatbot to “pretend to be a 35-year-old mom.”
For a big brand testing twenty ad concepts, that is useful. Even Listen Labs advises running a full study with real people for high-stakes go or no-go decisions, which tells you where the company itself draws the line.
Why it does not fit a solo business
The 93% number is not what it sounds like
The accuracy figure comes from the company’s own testing, and the release does not give a sample size or method. Even taking it at face value, “matched 93% of the time” on average says nothing about which 7% it missed. In research, the surprising answers are usually the valuable ones. A twin that agrees with the crowd on routine questions and misses the one customer who says “I would pay double if you did this” has missed the point.
Your sample is already small, and that is your advantage
A solo business might have 40 regular clients, not 400,000. You can talk to a meaningful share of them directly. Big companies build twins because they cannot call every customer. You can. Swapping that advantage for a simulation means giving up the one research edge a small business has.
Conversations do two jobs at once
When you call a customer to ask why they almost did not book, you learn something and you also strengthen the relationship. Customers remember being asked. Several will refer someone or book again partly because you cared enough to call. A twin can only do the first job.
The tool is priced and built for teams
Listen Labs did not publish pricing in its announcement, and the product is aimed at marketing and research teams. That pattern looks familiar: our take on Kargo’s ad-buying agent reached a similar conclusion. Enterprise AI tools are often impressive and simply sized for someone else.
What I would do instead: the five-conversation method
Here is a free alternative that takes about three hours spread over a week.
- Pick one question you genuinely need answered. “Would clients pay for a monthly maintenance package?” is better than “What do people think of my business?”
- Choose five people: three happy customers, one who went quiet, and one who almost bought but did not.
- Ask for 15 minutes by phone or video. Offer a small thank-you, like a discount on the next order.
- Ask about past behavior, not future wishes. “Tell me about the last time you needed this” beats “Would you buy this?” People are poor predictors of their own future spending.
- Record and transcribe with permission. Our beginner’s guide to speech to text covers free options.
Then, and only then, bring in AI. Paste the five transcripts into ChatGPT, Claude, or Gemini and ask it to list the repeated themes, the surprises, and the exact phrases customers used. That is the right job for AI in small business research: summarizing real voices, not inventing them. The exact phrases are gold for your website copy too.
Three questions that work for almost any solo business
If you are not sure what to ask in those 15 minutes, start with these. They are open enough to surprise you and concrete enough to act on.
- “What were you trying to get done when you first looked for someone like me?” This reveals the real job customers hire you for, which is often different from how you describe your service.
- “What almost stopped you from booking?” The answers point straight at the friction on your website, your pricing page, or your response time.
- “If I disappeared tomorrow, what would you do instead?” This tells you who your actual competition is, and it is sometimes “nothing,” which is useful to know too.
No twin panel, however accurate on average, will give you the follow-up question that comes from hearing a real person pause before answering.
When I would change my mind
I would revisit this position if three things happen: independent testing (not vendor testing) shows twins catching unusual answers, not just average ones; a version appears that can be built from a small business’s own customer interviews; and pricing drops to the level of a normal monthly software subscription. Some of that may arrive once Salesforce folds the technology into tools used by smaller companies.
Until then, I side with the view in our piece on why most small businesses rarely need frontier AI: use simple, cheap tools for the parts of the job that are truly repetitive, and keep the human parts human. Talking to customers is the most human part of all.
Do you disagree? If you have tried synthetic personas or AI customer panels and found them useful for a small business, I would like to hear the specifics in the comments.
