Tools for implementing Machine Learning and distributed ledger technologies (‘Blockchain’) can now be used relatively easily and cheaply in insurance processes. Used well, they present a great opportunity to improve productivity and security. However, organisation need to ensure that implementations and subsequent operations are as painless and risk-free as possible. This article aims to provide tips on avoiding common pitfalls.
Machine Learning (ML) describes software that adapts / changes / ‘learns’ when exposed to new information. Ie explicit programming by a human is not required as the software is self-adapting. A good example of ML in action in our everyday lives is spam-detection software that improves its own performance through experience (even a single email can result in learning that can be generalised across all users of the software worldwide). In the insurance industry, one fundamental process that benefits greatly from using ML is underwriting (the process by which the institution decides whether to take on the risk offered by a customer or broker, and at what price). ML algorithms can be trained on millions of pieces of customer data, actuarial information, and policy outcomes to suggest or make the best business decisions within parameters set by the firm. As a further benefit, the analyses done by ML software can unearth underlying (sometimes non-intuitive) trends as new information continues to be captured during live processing. Such trend analyses can result in better underwriting decisions and more competitive pricing (a benefit to customers). Other uses for ML in insurance include:
- combining ML with telematics in vehicles to improve road safety. Eg by using information on the road ahead to guide drivers, analyse driving patterns, and make safety suggestions.
- spotting account behaviours correlated with default and customer churn.
- increasing the speed and quality of processing (through digitalisation and information capture) from paper forms.
- improving fraud detection and prevention at the underwriting and claims-handling stages. Eg analysing handwriting (using ML) on digitalised paper documents can result in similar content being spotted across different jurisdictions, even if held in different formats (eg death certificates in different states in the U.S.).
- improving the speed and quality of insurance policy reviews by ingesting a policy, breaking it down into clauses and logical blocks, analysing and comparing etc. (when done manually this process is labour-intensive, error-prone and slow.)
Tips for adopting ML in insurance:
- start small – but do start. Do proof-of-concepts (POCs). Do not attempt to plan a large-scale transformation without ever having done a small implementation.
- early success will depend on having enough valid data to train the software. Make sure you can get those data.
- partnerships are important. Most insurers do not have enough expertise internally (as at February 2017) and will benefit from third-party assistance.
- diversify. Avoid betting the farm on a single ML approach or technology. Try different approaches in different POCs.
- Use Cloud-based, SaaS, on-demand solutions from partners, plus open-source tools. This will allow you to experiment at lowest possible cost (rent don’t buy).
Blockchain is a secure record of transactions collected into blocks grouped in chronological order and distributed over different servers to provide reliable provenance. The technology uses digital signatures and a consensus mechanism that ensures participants can agree on which transactions are valid. Recent experience with clients suggests that Blockchain is going to be an important part of the insurance technology (InsurTech) revolution. Blockchain benefits will include improved underwriting accuracy, reduced administrative costs, and improved success in preventing claims fraud. In the research paper ‘ChainReaction: How Blockchain Technology Might Transform Wholesale Insurance’, Michael Mainelli identified the three most viable use cases as: (1) placement and contract lifecycle; (2) KYC/AML (know your client / anti-money laundering); (3) claims management. A further benefit could come from (4) improved fraud detection.
- Placement and Contract Lifecycle. The placement process is often heavily paper based. Each participant must ensure that there are no mistakes by checking the documents. This often results in rework and wasted effort/money when mistakes and discrepancies (between the information held by different participants) are found. If contracts are stored on the Blockchain as an immutable ledger they are certain to be consistent for all users, thus removing the need for participants to repeat the checking process.
- KYC/AML. Distributed ledger technology can reduce the cost and time of these expensive, laborious and (very) slow processes, for example by eliminating the need for third-parties to produce reports. The Blockchain would create a digital identity system that would allow broker and insurer to manage their documents (credit reports, patient records etc.) without fear of losing control of personal data and other sensitive information.
- Claims Management. It is highly feasible to design a Blockchain ledger of all documents created in the claim process and make them available for interested underwriters and other stakeholders. This would make the process transparent and reduce cost, delay, and reputational risk.
- Fraud detection. http://www.FBI.gov reports that “The total cost of insurance fraud (non-health insurance) is estimated to be more than $40 billion per year.” Increased effectiveness in detecting fraud (eg falsified injury or damage reports) can be achieved by automatically validating and confirming ownership as well as authenticity of documents and location changes.
Tips for implementing Blockchain in insurance:
- start small. Find a small problem and use Blockchain to improve the process running behind it (make it faster / less costly / less prone to error)
- use private Blockchains such as Chain.com and Hyperledger for improved peace of mind (public Blockchains are still young and subject to security exploits and attacks).
- a Blockchain is a distributed network of trust, so consider looking for partners that share the same goals. Doing so could unlock new gateways to internal transformations and shed light on new business opportunities.
- Blockchain experts are rare and hard to employ, but it is crucial that you have a knowledgeable expert in your team. Hiring an experienced expert – even as a temporary or part-time consultant – should always be worthwhile.
Whether adopting ML or Blockchain or both the following approach should work well: brainstorm with representatives of lines-of-business and business functions to identify processes that are problematic and could benefit from either technology; create a list of POCs to execute over the next 9 to 18 months; use partners with real expertise and verifiable experience; be prepared to fail early on some POCs while finding gems in others; keep experimenting (coming up with ideas and using different technologies); and, gradually build robust business cases for change from those proofs of concept.
Good luck with your endeavours!
Cliff Moyce, February 2017
[this article was first published in Insurance Innovation Reporter on February 7th 2017 at https://iireporter.com/avoiding-pitfalls-when-implementing-machine-learning-and-blockchain-in-insurance/