Bitesize InsurTech: Emerge Analytics

March 17, 2017 George Hanks

Emerge Analytics are a machine learning startup focused on insurance.  (See our primer on Artificial Intelligence here and a blog on its application here.)  The company is developing models using proprietary methodologies to ingest and interpret insurers’ datasets, leading to highly predictive output.

The South African company is a member of the 2017 Startup Bootcamp cohort.

Emerge Analytics is currently applying its solution to use cases across the insurance lifecycle: marketing, sales, underwriting, fraud and lapse reduction.  The company will be presenting at Startupbootcamp’s Demo Day on 26 April.

The company was set up in 2012 but only recently started to focus on insurance, where the founders have experience – Laurence Rau from South African insurer Discovery (the holding company of the UK’s Vitality Insurance) and Danny Saksenberg as an actuary at Deloitte.

The six steps of a machine learning project

In a recent discussion with Oxbow Partners, Laurence described machine learning projects in six steps:

1. Identify the problem to solve
2. Identify available data
3. Feature engineering – explaining the characteristics of a dataset to the machine
4. Machine learning analysis – running the algorithms over the data
5. Implementation of findings
6. Monitoring

Emerge Analytics focus on the third and fourth steps.

Example applications

Laurence cited recent work with a life insurer.  Emerge Analytics used the client’s data to identify, with 99% accuracy, those customers who would lapse their policy within two months.  According to Laurence: “Insurers are currently reactive to policy cancellation, not proactive – we’re helping to change this.”  They have also done some work to help a bank identify the leads that generate 90% of sales by making 11% of calls.

The Oxbow Partners view

There is no doubt in our minds that machine learning can have a tremendous impact on the insurance industry.  This is certainly true if companies like Emerge Analytics can extend their early pilots to broad implementations.

But herein lies the challenge.  There is, arguably, a belief in the industry that successful pilots automatically convert to future disruption.  We urge caution.

The move to a broad implementation requires the engagement of many more parts of the insurance organisation, for example sales, operations and technology.  In other words, the ingestion and analysis of data might end up being the “easy difficult” bit, whilst creating operational impact could be the “difficult difficult” bit if the insurer doesn’t manage implementation of the models into operations.

We are strong advocates of machine learning – but the critical success factor could be the project or operational environment in which the technology is deployed.