Welcoming the machines: How insurers will drive value from machine learning

AI is clearly a hot topic for insurers. And, as Rick’s article aptly points out, InsurTech startups are creating compelling new use cases and applications for data, algorithms and AI within the insurance space. It is, indeed, an exciting time for insurance.

Yet many traditional insurers may read Rick’s article with concern and potential fear. In a recent survey by KPMG International, 91 percent of insurance CEOs admitted being worried about the challenge of integrating automation, AI and cognitive robotics into their existing business and operating models. They recognize the value that AI can deliver, but they also recognize the massive challenges of integrating those technologies into their existing IT estate and operating models.

For now, most leading insurers are focused on Machine Learning. With Machine Learning, algorithms are trained, based on historical labeled data and decisions, to answer a specific business question. This would allow an insurer to, for example, train the machine on historical claims data and then let the machine make the routine claims decisions.

Essentially, Machine Learning excels at the more ‘narrow’ use cases that focus on improving the speed, consistency and volume of business decisions. And it’s these applications that insurers are starting to focus on, not only because they hold a great deal of promise, but also because they are starting to deliver tangible results.

Forward-looking insurance executives recognize that – to innovate in today’s environment – they need to build a culture of experimentation. They know that to create a better customer experience, they need to improve the efficiency of their operations. And they are increasingly acknowledging that Machine Learning sits at the confluence of these two demands.

We work with many leading insurers to help drive value from Machine Learning and, based on our experience, we have identified 10 key considerations – both organizational and technical – that we believe underpin successful Machine Learning adoption, these include:

1. Secure a great team: How will you secure and develop the right talent to ‘drive the machines’? How will the teams be structured? How will they work with the business?
2. Create global alignment: Do your projects and methods drive the corporate strategy? Are your global technology stacks aligned?
3. Educate leadership: Do your executives truly understand the value that Machine Learning can generate? Do they understand the key terms and use cases?
4. Invest at scale: Are you making the right level of investment into technology, people, alliances and capabilities? Are you growing your Machine Learning team?
5. Be obsessive about data collection: Are you collecting and curating as much of the ‘right’ data as you can? Are you labeling your data fast enough to keep up with reality?
6. Take a portfolio approach: Do you balance the risks and rewards of your Machine Learning projects as a portfolio of investments, rather than hedging your bets on one big mega-project? Do you continuously track the benefits of the investments?
7. Embrace streaming data: Are your systems and processes capable of managing streaming data? Are you able to apply Machine Learning in the stream and take business decisions in real-time?
8. Review your cloud strategy: Are you able to quickly scale your Machine Learning platform to meet your computational needs? Do you have a clear DevOps strategy to ensure your cloud is automated and deployments are repeatable?
9. Open up your architecture: Does your architecture enable collaboration with technology providers and suppliers? Are you driving integration through open APIs?
10. Think about production: Do you have a model for pushing new Machine Learning models into production? Can your underlying infrastructure enable elasticity and keep up with demands?

Ultimately, the greatest challenge for insurance executives is to be comfortable with failure. The reality is that – as with any areas of experimentation and innovation – not all Machine Learning investments will turn into commercial results. The trick is to fail fast, learn from your experience and then exploit your new knowledge on the next iteration or project. The beauty of today’s technology is that it allows for just this type of speedy and cost-effective experimentation.

Like Rick, I certainly believe that AI technologies offer massive opportunities for the insurance sector. But the real spoils will go to those that start by getting the foundations right.

The author, Gary Richardson, leads a team of data scientists and data engineers at KPMG in the agile development of data science solutions.