For years, personas have been a fixture in the toolkit of marketers and insights professionals. They help you get a grip on the customer, for example when making choices about positioning, product development or targeted campaigns. At the same time, personas are a huge simplification of reality. Most organisations work with four to seven personas: recognisable, workable and not too many to visualise well. Yet this does not do justice to the diversity of society.
Blog
“You’re not mad if you hear a hundred voices”
Thanks to AI, we are seeing a promising innovation in the field of personas: personas are being brought ‘to life’. You can communicate with them: they answer questions, brainstorm with you on an interactive platform, or talk to each other in a virtual focus group. Precisely because they are scalable, these digitally accessible, interactive personas open up a world of new possibilities. After all, why limit yourself to 6 personas, 8 Mentality milieus or 4 Lifestyles? Wouldn’t it be brilliant to have no fewer than a hundred synthesised personas at your disposal, each representing 1% of the Dutch population? This allows you to work in far more detail at once, by combining general demographic characteristics with domain-specific situations and personal motivations.
I picture a virtual panel of digital characters, based on real data, domain-specific knowledge and behavioural profiles. AI agents that respond from their own unique perspective as soon as you put a research question to them. As if you were talking to 100 people at once. Not as a gimmick, but as a valuable addition to your research process.
In this article, I show how to build a synthetic panel like this, what challenges come with it and, above all, why now is the time to experiment with it. The potential is huge, but in practice it is still in its infancy.
How it works: from microsegment to AI agent
The synthetic persona panel is built on a set of 100 microsegments, based on large-scale, representative consumer research. We cluster the respondents into 100 microsegments with a clear profile, each with its own behaviours, preferences and context. Each microsegment represents 1% of the Dutch population and adds colour and depth to the customer landscape.
Every AI agent gets its own identity
For each microsegment, we then build an AI agent: a digital representative that “thinks” and “responds” from the perspective of that specific segment. The agent is given a unique prompt based on the research data. This prompt contains information about demographics, context and personal preferences, but also about tone of voice, language use and attitude towards the topic.
Ask the panel an open question and you get 100 different answers, each from a different perspective. Ask a closed question and you get a frequency distribution of the answers. This gives an indication of how different types of people in the Netherlands might respond.
Accuracy grows with targeted input
How accurate those answers and frequency distributions are depends heavily on how far the topic is from the original data. Questions that lean heavily on measured behavioural or attitudinal data produce robust patterns. For topics that are far removed from the underlying data, you should see the results more as inspiring input: a qualitative simulation, not a representative judgement.
Over time, you can keep improving the panel’s accuracy by explicitly researching new themes. By carrying out additional research into specific topics and adding this data to the AI agents, you enrich their perspective. The result is a panel that can be used ever more widely and that grows with the questions that come up.
First case in practice: working with KPN
In a pilot with KPN, we are building a panel of 100 AI agents, based on segmentation across three dimensions: demographic profile, telecoms situation and personal preferences. The result is a set of digital conversation partners that represent clearly different perspectives. KPN can use the panel to test early ideas, discover nuances in the customer experience or fine-tune communication. It is especially valuable in the exploratory phase, when you are still working out what works, for whom and why.
AI agents are just like people (well, almost)
AI agents are of course not real people. They base their responses on data and context, but remain simulated representations. Especially for topics that are further removed from the original input data, it is important to be careful about drawing firm conclusions. So do not see them as a replacement for quantitative or qualitative research, but as an additional tool that lets you gather input from your target audience in an accessible way.
We are only just getting started. The technology is there and the first pilots are running. So here is my invitation: let’s run a pilot together. Not theory, but practice. Not a promise, but discovering together what works, and what doesn’t.
So will you soon hear a hundred voices when you ask a question? Don’t worry: you’re not mad. You’re simply using a synthetic panel the way it was meant to be used.
