Podcast: Play in new window | Download
Dr. Olivier Toubia, Glaubinger Professor of Business at Columbia Business School, joins Sima Vasa to discuss his landmark study building digital twins from over 2,000 real participants — and what the results reveal about the genuine limits of synthetic data in market research.
Olivier explains why digital twins skew hyper-rational, why a 0.2 correlation with real human behavior is the honest benchmark, and why the “better, faster, cheaper” promise of synthetic data still has a question mark on “better.”
Olivier also covers the hybrid panel model for keeping digital twins calibrated over time, the structural advantage of within-person A/B testing with synthetic respondents, and what the neuromarketing hype cycle can teach the industry about moving faster toward evidence-based answers.
KEY TAKEAWAYS
00:00 Introduction.
02:07 From operations research to marketing, Conjoint analysis and capturing human preferences with math.
03:54 The adoption cycle repeats: every new technology prompts replication before reimagination.
05:44 How synthetic data evolved from basic LLM personas to data-rich digital twins with real heterogeneity.
11:54 The 0.2 correlation finding: digital twins and humans, and calibrating what that actually means.
14:41 Twins skew hyper-rational, struggle with affect-based decisions, and perform better on text than video.
17:06 The “holy grail” of “better, faster and cheaper,” and why “better” still carries the biggest question mark.
23:33 The hybrid panel model: synthetic at scale, small human sample running alongside to keep twins honest.
Thanks for listening to the Data Gurus podcast, brought to you by Infinity Squared. If you enjoyed this episode, please leave a 5-star review to help get the word out about the show, and be sure to subscribe so you never miss another insightful conversation.
RESOURCES MENTIONED
Columbia Business School Digital Twins Lab
https://business.columbia.edu/ai-in-business/labs/digital-twins-lab
Prolific
Hugging Face (Digital Twins dataset)
https://huggingface.co/datasets/LLM-Digital-Twin/Twin-2K-500
Qualtrics
https://www.qualtrics.com
#Analytics #Data #MRX
