Abstract
We present a practical implementation of applying LLMs to enable micro-personalization in a mass market context through the use of customer digital twins. The proposed pipeline was tested using two field experiments with approximately 330,000 customers of a global financial institution, receiving cross-sell offers generated either by the business-as-usual approach currently used by the bank, or by the proposed digital twins pipeline. Success was measured based on actual conversion rates. We find that the proposed pipeline generated a +6% conversion uplift relative to the control condition and a +11% prospective three-year gross margin uplift for credit card offers, and a +25% conversion uplift and a +30% prospective three-year gross margin uplift for payroll portability offers. To the best of our knowledge, this is the first implementation of digital twins (i) at scale, (ii) leveraging first-party behavioral and transactional data, (iii) validated using randomized controlled trials, (iv) using actual business outcomes as success metrics.