I travel a lot for business. I consult with customers across Africa in a variety of industries. But no matter the country, culture or industry, the one thing that most businesses have in common is that there is inevitably a disconnect between the data scientist teams and business units. Business units often don’t understand the predictive models and how they can really add value to the organization. This can be hugely frustrating, especially for the data scientists, who have often invested a lot of time and effort into developing these models.
Bridging the gap
There is, however, a way that data scientists can help themselves. Using a more business-orientated approach can help them to convince business users, management and IT of the value that these models can add.
Let’s consider the case of lapses (where clients cancels/stops their premiums) in the insurance industry, which can result in a significant loss of future profits. This is also often an indicator of customer dissatisfaction, which is likely to have resulted from poor customer experience.
The ability to predict lapses before they take place therefore allows insurance companies to react earlier and hopefully take effective action to retain the customer. The company can then engage with those customers to provide relevant offers to keep them happier.
Suppose the data science team has developed a predictive model that enables the company to do just that. The model has been trained on historic data covering existing customers that have lapsed and those who have remained with the company. The model differentiates well between the two subsets (known as lapse and non-lapse).
This is exactly what predictive models are designed to do: differentiate between possible events.
Instead of giving business a detailed account of how their model works, and the statistics behind it, how does the data science team present this to the business unit to show how it could add business value?
Models to business value
The answer is to show the number of customers who could be retained using the model, compared to the current situation.
The table below shows the possible results of a retention marketing campaign in a population of 100,000 active customers.
The 100k customers selected by the model is based on the top 10% highest scoring in terms of high lapse risk. For the model to be compared, we have a hold out which is a random sample across the population.
The results look promising. Using the model, the business could increase its retentions by 35.7%. The data science team could then put a value on that in terms of the customers future value to the business, or they could work with the business units to provide these estimates
However you wrap it up, this is a simple but effective message for the business unit, which does not require much technical understanding of the modelling process.
Moving to the business reality
The next step, of course, is for the data science team to use the model on existing customers and identify customers that look most likely to lapse. Together with the business unit, they can analyse these customers to get a better understanding of them, and therefore identify the most relevant offer.
If business understands the kinds of problems that data science can solve, they will be more likely to come to the data scientists to solve other business problems that they face with. Working closely together to solve a business problem is likely to be the strongest way to convince business units of the value offered by predictive models. The key is to choose the right messages to get to that point.
