Has anyone used an AI-based synthetic data product for gathering customer insights (rather than using traditional market research like a customer survey), and what have people found good and bad about the tools?
Sort by:
I've used synthetic data in a number of instances to help with customer insights. There are 3 key areas that I would highly recommend you think through: 1. Synthetic data should not be viewed as replacing, but its an augmentation for you. If you have a hard to reach audience, or you have specific gaps in your data, or if you are looking at additional scenarios which you don't have data for but need data to do "what if" planning, synthetic data can be useful in helping in these instances. 2. As best you can, you want to have some foundational data to help generate synthetic data. With this foundational data, if there are any biases, or skews, synthetic data will amplify that even further. You also need to decide how much synthetic data you're generating relative to your source data (10/90 vs.50/50 vs 90/10 splits between real vs synthetic data). Not all synthetic data generators are the same, so you need to understand what are the pros/cons of the algorithm that's being used to generate synthetic data. This will determine what degrees of confidence you have. 3. Finally, synthetic data isn't going to offer much around capturing emotional mindsets per se. You can generate synthetic responses in a qualitative sense, but you will see that you won't get the emotional state of mind. I wrote this primer on synthetic data if that helps: https://www.linkedin.com/pulse/unlocking-potential-synthetic-data-marketing-market-research-m42bc/?trackingId=i9csnFt9QsSQ5lou58YZ4g%3D%3D
I haven't, but similarly interested.
We are in the very early stages of using a synthetic panel to supplement our existing research or to extend on to research we have already completed. Our internal teams are excited about the quick response from this channel so we will need to continue to explain how it should be used and that it should not be used as a replacement. For example one promotional offer preference question gave us results where 100% of the panel chose one option most likely due to the word free in the prompt. Understanding how AI responds to that will require us to be diligent in cautioning.