← All retail signals
Marketing turn

The Rehearsed Customer

Companies now build AI copies of real customers from their interviews and data, and test launches on them first. One study finds the copies beat demographic personas.

What is it?

Companies are building AI models of specific customers from their interviews and data, then putting launches to the models first. One study finds such twins predict reactions better than demographic personas. How far a twin can speak for its person is the open point.

Why now?

A model can now be rendered from one person’s interviews and records and questioned in their place. Simile says it raised $200 million at a $2 billion valuation, and Salesforce agreed to buy Listen Labs, which offers twins. Whether twins hold up away from their training data is unsettled.

Who’s building it?

Simile and Listen Labs build twins, CVS Health and Telstra were early customers, one study tests them.

SimileListen LabsCVS HealthTelstraSalesforce
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — twin makers such as Simile and Listen Labs.
Buyers — companies testing launches, including CVS Health and Telstra, among Simile’s first customers.
Gatekeepers — researchers testing twins against real reactions, and trade press.
What’s at stake
How far a twin can stand in: ahead of personas in one study, with errors on distant questions.
Simile says it is valued at $2 billion.
Richer interviews did not improve prediction in one study.
What’s unsettled
Whether twins hold up on questions far from their training data.
Whether thinking-style differences make some people harder to twin.
Whether one study of six marketing stimuli carries over to other products and messages.
02 · How the demand matureda progression, not a timeline
Definition
Research Live separates a twin, one person’s model, from a persona, a segment’s.
Capital
Simile raised $100 million, then says $200 million at a $2 billion valuation.
Customers
CVS Health and Telstra were among Simile’s first customers; CVS’s venture arm joined the later round.
Test
One study found interview-built twins predicted held-out reactions better than demographic personas.
03 · Where the market goesthe forks, not a prediction
→Twin or persona: interview-built twins may displace demographic personas in pre-launch tests, or the two may sit side by side.
→Later studies may test twins on products and questions further from their interview data.
→Twin makers may publish prediction results against held-out reactions, as the study did, for buyers to read.
→Twins may arrive bundled into larger research platforms, as with Listen Labs and Salesforce.
The direction depends on where twins are tested next: which questions, which people, and whether results are shown against real reactions for buyers to check.
04 · The Bespoke Default deskadjacent markets, moving now
The Digital Twin Self
Medicine models individual patients as twins.
Medical digital twins model an individual patient and update over time
The Week
The Model Kitchen
A food giant tests simulated shoppers before shelves.
Nestlé is testing AI-driven consumer simulations before products reach shelves.
Reuters (via RTÉ)
Hyper-Personal Media
Tailoring to one individual reaches media as well.
Deloitte: AI hyper-personalization becomes ubiquitous, leaving fewer shared media moments.
Deloitte Insights
05 · The Wire
01
A twin is one person’s model, a persona a segment.
Research Live ↗
02
Simile raises $200m for simulated users, it says.
TechCrunch ↗
03
Twins from interviews beat demographic personas in one study.
arXiv (Deng et al.) ↗
04
Simile’s model simulates individuals’ reactions to new products.
SiliconANGLE ↗
05
Listen Labs offers twins built from survey interviews.
Research Live ↗
The panel on this signal
They agreed on
–A twin models one specific person from that person’s own data, while a persona stands for a segment.
–In one study, twins beat demographic personas on held-out reactions.
–Richer interviews did not improve prediction, and errors were linked to distant questions.
The discrete-choice modeller disagreed on
–Matching reactions to six stimuli is not predicting choices; a choice model needs trade-offs among alternatives, prices and constraints.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.