Digital Twins for Research and Insights

Ray Poynter, 3 June 2025

---

Over the last year, there has been a growing interest in Digital Twins, which are virtual participants based on real individuals. In this note, I will outline what is happening, why it is happening, and share some key concerns that need to be considered.

Digital Twins, the Basics
The Twins are created by gathering data about an individual and using that data to create something that will answer questions in the same way the original would have done. For example, if you have an online community, you will have multiple qual and quant responses from each member of the community. This information can be used to create a virtual representation of some or all of the members of that community. These digital twins can then be asked further questions.

At a less detailed level, the data collected from qualitative projects, such as diary studies, depth interviews and focus groups can be used to create a digital twin for each real participant.

Why?
The main reasons for using digital twins, in the context of market research and insights, is to enable faster and less expensive research. If a user can get a result that is 80% as good as real research in minutes instead of days, then customer insight can be used as a real-time input to every stage of decision making

But there are other reasons, for example:

1. Reducing the burden on participants. Many research projects are tedious for participants. For example, an insurance company might be testing different formats for claims letters. If we can shift some of these studies to digital twins, we can reduce the burden on real people.
2. Sometimes users want to test ideas that can’t be shared with anybody outside of their organization. Digital twins can be trusted not to share the information (if the AI platform has been configured properly).
3. Avoiding Sensitising the Participants. If we want to test multiple versions of a new idea, we would ideally like to use separate people for each test. However, digital twins can be configured not to ‘learn’ from the tests they participate in.
4. Hard-to-reach samples: If you have a group of people who are difficult to reach, their digital twins can make life easier for both the user and the participants.

A less reassuring example of why an organisation might want to create a very large number of digital twins of its customers is that it could then conduct one-to-one modelling to predict how best to market to them. One of the goals of personalisation.

What is the current state of the Art?
In my experience, for quantitative projects, the information suggests that digital twins tend to be quite good at reflecting attitudes and being creative, but not so good at forecasting behaviour. For qualitative projects, the digital twins produce responses which are very similar to those in the sources, but tend to exhibit better grammar and structure.

Evaluating Digital Twins
As with all synthetic data, the Digital Twins need to be assessed. A good approach to this process is:

1. Check the answers to questions against the training data. This is a hygiene test; the responses should be similar, but it does not tell you that the digital twins are doing anything more than repeating the input data.
2. Before creating the twins, hold back some of the data; do not use it in the training. Then, compare the digital twins with the hold-out data to assess how closely they replicate the hold-out data.
3. Look at the picture at the total level (would the decision have been the same) and at the individual digital twin level to assess the stability of the process.

A notable example of the evaluation process is provided in the Ipsos paper referenced below in the further reading section. Ipsos interviewed 150 women about menstrual cycles, experiences and products. They used this as their training data to create the digital twins. They then interviewed the real women again, asking different questions about the same topic and talking about new product concepts. The digital twins were then asked the same questions, and the results compared – another way of creating an independent test and validation set. This study confirmed some similarity in attitudes, a good performance in breadth of creativity, but a poor match in most preferred product concept.

Ethical Issues
There are broadly two categories of ethical issues with Digital Twins, the first in respect of the users of the Digital Twins and the second relates to the individuals whose data created the twins.

The Users: For the users of digital twins, we need to be sure that they are aware of how much confidence they can put on the findings generated from the digital twins, both in terms of direction and accuracy.

The real participants: For the participants, there are three issues to consider. Potentially, the process of creating digital twins could deprive the real participants of revenue, utilising information they had contributed. The second potential problem is that participant anonymity could be compromised. For example, if the users were to ask enough questions, it might reveal somebody’s details. A third potential area for concern is that the digital twin might be prepared to answer questions that the donor person would have declined to answer, raising consent issues.

Digital Twins are not just a Research Thing
Outside of the world, the concept and term Digital Twin is in wide use, for example, Singapore has a digital twin of the city, factories create digital twins of machines, and in healthcare, digital twins of organs allow doctors to research and plan.

Want to find out more about Digital Twins and Synthetic Data?

I am running a course on Synthetic Data on 19 June, 2025. You can find out more about it by clicking here. I will also be running a course on Synthetic data in the Australian time zone for The Research Society on 30 July, click here to register.

Further Reading

• Is there Depth and Empathy with Digital Twins, Ipsos, Views, Maya Ilic, Ajay Bangia & Jim Legg
• Synthetic respondents & digital twins: Snake oil or useful? Bubble or Banger? LinkedIn, Frank Buckler
• Can Synthetic Data Rescue Survey Research? A Case Study, Crux Research, John Geraci
• Digital Twins, Synthetic Data, and Audience Simulations, Verve
• Are Digital Twins Just Virtual or a Reality for Life Science Insights? Sparq, Dan Wasserman
• Toothpaste maker Colgate testing new product ideas on ‘digital twins’, Reuters, Jessica DiNapoli
• Top 10 Digital Twin Use Cases | Industry Examples, VisionDignify, Rajat Singh