An AI-powered chatbot
activating 25% of 'non-checkers'
ABP | Renee
We joined ABP in their mission to activate pension 'non-checkers'.
This resulted in Renee: an AI-powered chatbot that successfully motivated 25% of participants to check their pensions for the first time.
Research revealed that 20% of ABP participants had never reviewed their expected pension income.
These participants, known as 'non-checkers', generally showed no intention to review their pensions. Our experiment aimed to activate these non-checkers to check their pensions for the first time on mijnpensioenoverzicht.nl.
The approach was based on 'The Theory of Planned Behavior', focusing on influencing attitudes and increasing perceived behavioral control.
We developed an AI-powered chatbot named Renee that independently engaged with over 200 'non-checkers'.
Built using an open-source Large Language Model, Renee was designed to influence negative attitudes by motivating individuals and boosting their confidence to review their pensions.
The chatbot maintained a concise and approachable tone while discussing pensions, ensuring compliance with ABP's legal, compliance, and risk policies, as well as technical IT requirements.
We built Renee using Sunrise, our AI development platform, which allowed us to rapidly develop and deploy a compliant AI solution that could engage with users naturally while adhering to strict regulatory requirements.
Using Renee, we successfully activated 25% of 'non-checkers' to check their pensions.
The results exceeded expectations, achieving unprecedented success in participant activation. The AI environment we developed not only delivered impressive results but also established a compliant framework for future AI-driven experiments at ABP.
The numbers behind Renee's success
- 916 conversations with Renee
- 8711 messages sent
- 257 messages blocked by our guardrails
- 25% of participants converted
- 1 pilot, 1 scientific experiment
- 7 Large Language Models tested
- 4526 evals ran
In the 7 years that we've been looking for ways to activate participants, we have never had this kind of promising results. On top of that, the great thing about this experiment is that The Main Ingredient developed an AI-environment that is compliant with our legal, compliant, risk and IT policies. Which means we can also reuse it for other AI-driven experiments.
