The Insurance Industry’s Buy vs Build Dilemma for AI Solutions

The Build vs Buy dilemma for AI-powered solutions is a tough one for business leaders even for those with deep pockets, significant scale, and technology teams packed with cutting-edge talent. The reason it is so tough boils down to the fundamental nature of artificial intelligence and next-generation machine learning projects: They don‘t behave like traditional software build-outs and that bad behavior can be quite a shock to those not prepared for it.

In the article available for download below, EvolutionIQ CEO and Co-Founder Tomas Vykruta shares his firsthand experience building our market-leading claims guidance platform. In it, he outlines the many steps required and the many risks and pitfalls that insurers will need to address when making the decision on whether to build an in-house AI solution.

As Tom underscores, Even our team of PhD data scientists from Google, Meta/Facebook, Bloomberg and other tech leaders could not solve the problem in the first six months of effort. Even after the 14 month mark, although we had made progress, we were getting zero breakthroughs in terms of complex claims analysis and insights. Over 15 different approaches were taken, and only one worked. Only at the end did the breakthroughs happen and then continue to cascade. We‘re still seeing them and the value is substantial. But the ability to tolerate this kind of success-desert for very long periods of time is essential and not for the faint of heart.
Key challenges that Tom explores in the article include:

• The substantial costs of the investment
• That as many as 87 percent of AI projects fail to launch
• The need to augment proprietary carrier data with external data sources
• How marginal returns in AI projects are nonlinear, with the last 5 percent of engineering effort producing most of the value
• That long time horizons are required before prototypes can even begin to be tested
• The need for builders to develop a framework of evaluating risk vs success across the full life cycle of the initiative
• How machine learning initiatives have distinct tailend risks not present in traditional software development
• How adoption by frontline users can be a major stumbling block if the software does not immediately perform as expected
• The need to plan for ongoing fine tuning and calibration due to data drift, in which statistical properties in the model can change in unforeseen ways as new data continually hits the system

As Tom points out, there is a need for urgency. Advances in machine learning in just the last five years have created a situation in which the insurance industry which has long been underserved by advanced technologies  is about to have its own Uber moment as it undergoes a once-in-a-generation shift to replace outdated claims management methodologies.

That means the decision about build vs buy isn‘t just academic  it‘s one that C-suites across the industry are making now.

Click here to download the full article, The Insurance Industry‘s Buy vs Build Dilemma for AI Solutions.

‍