An illusion – the ‘Next Best Action’

In customer management, the “next best action” is a recommendation to staff on the most relevant next conversation to have with each customer – the next product or service they are likely to require. It is usually determined by analytics and machine learning.

The ‘next best action’ promises much to insurers hoping to engage better with their in-force customer base. We can steer the business towards a more timely, more relevant and more commercially valuable interaction with customers. This would be great for insurers, motivating for customer-facing staff, and engaging for customers.

But there’s a problem

Increasing numbers of software providers and consultants are now offering to plug in a next best action for a given customer base.  After all, we can see how well it works for Amazon. Unfortunately for most insurers, next best actions are little more than an illusion today. This is because insurers need to take some other steps first.

Mathematically, it is true that we can calculate a next best action for each customer using analytics. We have the tools to do that. But the vast majority of these next best actions are unusable (i.e. not specific enough to be actionable by staff) and inaccurate (i.e. incorrect in too many cases for most customer-facing staff to trust them). We had this reconfirmed recently in direct feedback from customer-facing staff at one insurer that has experimented with next best actions for some time.

Why is it an illusion? Today, life and pensions insurers simply do not have enough interaction data and experience from running test and learn experiments to be able to create a set of useable and reliable next best actions. It’s also more complicated for insurers because most of the interactions are with customer-facing staff or distributors, so capturing the interaction data is also more complicated than pure web-only interactions.

In time this can be developed, but first we need to take the baby steps and build up the insights and data for the algorithms to learn from through a series of disciplined customer analytics campaigns.

Campaign experiments are the starting point

A campaign only has one specific focus (for example helping to increase pension contributions after an annual salary increase). In a campaign we will test a handful of promising methods and learn what works, and for who. We use machine learning to help us identify the customers with a higher propensity to contribute more, and help uncover potential hooks for these conversations with customers.

Importantly, these campaigns need to be carefully managed and structured. In particular, we must capture the data and feedback from the campaign interactions to help test and learn and improve.

As most insurers cannot handle running more than one or two campaign experiments in parallel today, it can take some time to build up the data and experience. But ideally each campaign experiment will wash its own face in terms of value creation.

Campaigns feed the next best action

Over a period of 18+ months, we run many campaign experiments. These give us the data and experience to determine a next best action for our in-force customers that is both reliable and actionable. Without the experience and data from many campaign experiments, we cannot do this.  But the prize is enormous for insurers – much higher customer engagement, more effective customer-facing staff, and more commercial success.

Until then, we caution insurers to focus on the campaign experiments and baby steps and not be drawn in too far by the next best action illusion …


Get in contact with Matt