Testing chatbot assumptions for Prudential

Before chatbots could revolutionise customer service, Prudential had to answer one crucial question: would their clients really trust a machine with something as sensitive as their pension pot?

Before ChatGPT blew the world's collective mind, we had machine learning powered chat bots. And, like the current climate, big businesses weren't sure how customers would react to them. Prudential wanted to understand whether customers would be comfortable engaging with a chat bot around complex financial matters, and if it would be useful enough to divert calls from the contact centre. Sometimes, the only way to test an idea is to put it out into the world. So that's what I did.

Over the course of six weeks I identified a small use case (people turning 55 and gaining access to their pension pot), worked with Deloitte's AI team to develop an MVP, created a content database and got it signed off by marketing, legal, and compliance, ran an internal staff test, and ran a test with 500 customers. Not bad going for a 200 year old pensions company. Presenting the results of this test to the CMO unlocked budget to continue development and turn the MVP into a scalable product.

This project taught me a huge amount about getting things launched in big businesses. I learned how to protect the core of a proposition, and the integrity of your test, as you make concessions with other parts of the business in the name of speed and getting it out there.