There is a lot of excitement, as well as misunderstanding, around synthetic respondents and digital twins. We’d like to explain exactly what these are, why and when you should use them, and when they need to be treated with caution. And with our behavioural science hat on, we also offer a solution for how you can make these AI respondents more human.
Synthetic Respondents
Synthetic respondents are AI-generated personas built to mimic how a target audience (e.g. HCPs and patients) answer a survey or react to a concept. We can think of them as representing a sample of respondents.
Removing the need for recruitment and panel fees, the main benefit of synthetic respondents is that you can produce results in minutes instead of weeks.
This is especially appealing when working with hard-to-reach respondents, such as specialist physicians and rare-disease patients, where traditional recruitment and research can be slow and expensive. Synthetic panels are great for message and concept testing, able to screen dozens of messages and concepts before a single real respondent is recruited.
But with that added speed can come a loss of depth and novelty. Synthetic respondents struggle with questions around a new area that hasn’t been explored before, and answers often fail to capture the emotional nuance that drives real decisions.
AI systems are inherently logical, humans are not; the irrational behaviours of real doctors and patients tend to be missing from synthetic respondents.
Digital Twins
While synthetic respondents give you a snapshot of a sample, digital twins model a specific individual doctor or patient, built to persist beyond a single study.
This makes them especially valuable after the initial research is done, to follow up on findings or test scenarios without going back into the field each time. Instead of a stand-in survey respondent, we can think of them as ongoing simulations that teams can return to with “what if” questions.
However, similarly to synthetic respondents, digital twins can fail to capture the irrational behaviour of real HCPs and patients.
Knowing when and where to use each
Synthetic respondents and digital twins solve different problems, and each comes with its own trade-off.
Synthetic respondents buy speed by sacrificing depth. They are useful for screening ideas at scale but shouldn’t be used as a substitute for understanding how an HCP or patient behaves. For example, synthetic respondents can be used to test several messages and concepts at scale, allowing you to reduce the number of stimuli to test with real respondents during the market research, as well as identifying areas for further research and exploration.
Digital twins are great to use after market research. For example, you can model them on real respondents from your study, allowing you to continue research through scenario testing and follow-up questions.
For deep insight into how an HCP or patient behaves, such as in a patient journey, synthetic respondents and digital twins shouldn’t replace real respondents but can supplement and sharpen your research.
How we are using them at Branding Science
Our behavioural science expertise gives us an advantage here. By applying behavioural science principles and over 12 years of our institutional knowledge, we can engineer both types of AI respondent to carry certain “irrational” behaviours, habits and biases we already know are exhibited by the respondent type. In this way, we can get closer to the truth of how real people behave.
Contact
To find out more about how we can help you make your AI respondents more human, contact our team:
Elizabeth Brown – Senior Data Scientist – [email protected]
Lucy Ireland – Senior Research Director – [email protected]