Some surprisingly honest revelations are available in a booklet promoting a conference on Machine Learning in the Insurance Industry. It can be downloaded here. The obvious applications revolve around claims processing, fraud detection, policy pricing and customer acquisition. The firms that are prototyping with machine learning all see benefits, but there are associated hurdles that need to be overcome. Senior management are often not comfortable with models that lack transparency, since it places total reliance on algorithms and the skills of those who are developing models. There are also regulatory requirements for transparency that need to be met, and a black box is by definition not transparent. Another issue is the need to integrate machine learning with existing IT infrastructure – something that could slow down development and deployment of capability quite considerably. Some firms are opting to use cloud resources to bypass these limitations to some extent, and others are simply building another layer of IT infrastructure to run along-side the legacy systems.
The main benefits are focused on the responsiveness of machine learning to changing markets and environment. Instead of building a model and having to optimize every six months or so, it is possible with machine learning to rebuild models on an ongoing basis, so the models are as accurate as data allows. Reduced cost is also another benefit, with one company seeing model building reduce from a month down to a week.
Skills are obviously key in all of this, and managers report difficulties recruiting the right skills and validating those who say they have got them. It seems the only answer to this is some level of education in what machine learning is and what it is capable of.
Machine learning and AI will transform most businesses, but it’s not going to be an easy ride from a cultural and organizational point of view. But this conference does look like it will be addressing real issues instead of the usual hype and over-expectation.