AI 'Digital Twins' Are Replacing Focus Groups To Predict Consumer Buying Behavior

Who needs humans when algorithms can replicate the data that more accurately predicts their purchase preference decisions? JL 

Isabelle Bousquette reports in the Wall Street Journal:

AI can take data on a person’s individual characteristics—such as appearance, shopping preferences and health profile—then predict how they would look in an item of clothing, how they would answer a question or be affected by a disease. This AI content, referred to as a person’s digital twin, allows companies to ask AI focus-group questions, generate digital twins of people based on their health data to predict how disease might progress for those individuals or generate answers to questions like whether a female in her 30s would pay a 10% increase on a streaming service subscription.

Artificial intelligence is making it possible for companies to replace humans in tasks that range from modeling sweaters to participating in clinical trials.

AI systems can take in data on a person’s individual characteristics—such as appearance, shopping preferences and health profile—then predict how they would look in an item of clothing, how they would answer a question or be affected by a disease. This AI content, sometimes referred to as a person’s digital twin, is already being used for a variety of tasks.

Los Angeles-based startup AI Fashion uses photos of real models to generate completely new AI images of them modeling various pieces of clothing for fashion campaigns and e-commerce sites. Another startup, Brox AI, created digital versions of 27,000 individuals, with information about their brand preferences and shopping habits, that allows companies to ask the AI focus-group-style questions. And San Francisco-based Unlearn is using AI to generate digital twins of people based on their health data to predict how disease might progress over time for those individuals—aiming to make clinical trials more efficient and effective.

While the technology raises questions over the future of the human workforce, these companies say humans continue to play a vital role, and can be compensated for their willingness to share their data for the creation of AI twins. For businesses, the digital people are a way to scale faster and save on costs.

Consulting firm Gartner refers to the technology as “digital humans”—and estimates that in five to 10 years companies might even have digital twins for every single one of their customers. It is still early and there are a number of challenges. “Consumer perceptions and attitudes may form a backlash against brands if terminology, data and use cases aren’t handled with care,” said Gartner analyst Marty Resnick. Nevertheless, enterprises are starting to invest in the idea of using AI to digitize and monetize some aspects of humanness.

In fashion

Women’s clothier Anne Klein is testing technology from AI Fashion that generates fashion shoots based on photos of real-life models. AI Fashion said it uses a mix of proprietary technology and industry-leading open-source models.

An AI-generated image of model Leticia Jacobson. PHOTO: AI FASHION

“Consumers are looking for higher personalization, while also being able to see the product in a wide variety of different environments. AI enables us to do this at scale,” said Doug Weiss, senior vice president of digital, e-commerce and AI for Anne Klein parent company WHP Global.

Weiss said the tool won’t necessarily completely replace photoshoots, but “this enables us to build out the broad assets our shoppers are looking for when they shop,” he said.

A number of startups offer services that use AI to generate images based on a brand’s clothing line. In some cases models are completely AI-generated, a practice that has led to criticism for potentially putting real models out of work.

AI Fashion tries to differentiate itself by putting a real human model in the center of the process, said co-founder and Chief Executive Daniel Citron, a former creative lead at Google who founded the company in 2020 with its chief technology officer, John Chirikjian.

Model pay rates vary on a variety of factors including the brand, number of images and popularity of the model, and models can turn down any campaigns they don’t feel comfortable representing, AI Fashion said.

Weiss said more personalization and cost efficiencies are benefits he expects to see from using the tool, but added that it is too early to estimate exactly how much money could be saved.

In focus groups

Brox AI’s focus group tool gives companies the chance to get answers to questions without the expensive monthslong process of setting up a real focus group.

The tool is powered by digital twins of 27,000 real individuals, said Brox co-founder and CEO Hamish Brocklebank.

“We know where they shop, what they buy, what they like to buy. And we’ve predominantly collected a lot of this information through interviewing them at great length,” Brocklebank said.

Based on interview data, Brox’s proprietary AI algorithm can generate answers to questions like whether a female in her 30s would pay a 10% increase on a streaming service subscription. Participants were paid anywhere from $20 to $150 depending on how many interviews they participated in, he said.

 

Using the tool, companies can type in queries about what offers might resonate with particular consumers, how sensitive they might be to price increases, or what might drive them to pay for new services.

The tool costs anywhere from $25,000 to several hundred thousand dollars annually depending on how companies use it—potential savings for companies who spend millions annually on focus groups, Brocklebank said.

In clinical trials

Far away from the fashion photoshoots or corporate focus groups, startup Unlearn is using AI to generate digital twins of people that predict how a particular disease might progress over time.

The company, founded in 2017, has raised over $130 million and has 69 employees.

Typically in clinical trials, one group of people receives an experimental drug and is monitored to assess its side effects and effectiveness. A second group receives a placebo and is monitored to assess how the disease would progress without that experimental drug.

Unlearn takes in baseline data points about a given participant’s health, runs it through a bespoke model that is trained on vast amounts of clinical data, and generates a digital twin for that individual that forecasts how their disease would progress if they were in the placebo group, CEO Charles Fisher said.

 

Using a digital twin for the placebo means more real people could be in the experimental group, and gain access to the potentially lifesaving treatment, he said.

“One of the number one reasons patients don’t want to participate in trials is they don’t want to be randomized into placebo groups. And this gives everyone who participates in a trial access to the experimental therapy, which is pretty much the reason why people are participating in the first place,” Fisher said.

Patients would consent to having digital twins created as part of their normal consent to participate in the clinical trial, Fisher said, and likely wouldn’t receive any additional compensation for it.

The technology could be especially meaningful for a disease like ALS, also known as Lou Gehrig’s disease, where patients typically pass away within three years, said Kasper Roet, co-founder and CEO of biotechnology company QurAlis.

QurAlis is discovering and developing precision medicine treatments for ALS, frontotemporal dementia and other neurodegenerative diseases. Roet said it is aiming to begin testing Unlearn’s technology—alongside full human placebo groups—as early as next year.

“It is unfortunate that at this time we have to give people that have a lethal disease a placebo drug just to be able to run the trial,” Roet said, adding that there is more work to be done before technology from companies like Unlearn can fully eliminate placebo groups.

“But that is what we’re driving towards,” Roet said. “And I’m optimistic that eventually we will get there.”