McKinsey’s research division McKinsey Global Institute, in April 2018 published a discussion paper on the impact of artificial intelligence (AI). Applications and relevance of AI are broken down by industry and by corporate function, for an in-depth and nuanced analysis.

“Notes From the AI Frontier: Insights From Hundreds of Use Cases” in particular focuses on the relevance of so-called “deep learning” techniques. Deep learning refers to techniques known as reinforcement learning; and feed-forward, recurrent, convolutional, and generative-adversarial neural networks.

Limited Relevance With Respect To Insurance

According to the paper, the insurance industry can continue to rely on “traditional” analytics such as tree-based ensemble learning, classifiers, clustering, and regression analysis. Among newer, deep learning techniques, only recurrent neural networks seem to hold significant promise for insurers.

However, marketing and sales in general can make good use of some of the advanced techniques such as feed-forward and recurrent neural networks. And that applies to sales & marketing of insurance products as well. Conversely, risk analytics are better served by classical regression analysis and scenario testing techniques such as Monte-Carlo.

From a functional standpoint, McKinsey sees little relevance of the more advanced techniques for Finance & IT or for human resources departments.

In the end, the insurance industry ranks among the least susceptible of deriving added value from AI over other analytics techniques. The relevant index for insurance is 38, compared to an average of 62. It is near the bottom of a range that goes from 30 (aerospace and defense) to 128 (travel industry).

But Nevertheless A Significant Impact on ROE

At the same time, the study states that adopting AI may result in a 3 to 7% improvement in insurers’ combined ratios. In most circumstances under Solvency II, three percent of revenues is equivalent to 15% ROE (return on equity) and seven percent to 35% ROE, both (very) healthy levels of profitability. In other words, AI may bring about very significant improvements in profitability.

The apparent contradiction between the limited relevance of AI for insurers and its significant impact may be explained by the fact that on the one hand, insurance actuaries have for a long time developed and applied sophisticated statistical procedures to very large sets of data for pricing and underwriting purposes, leaving relatively little room for improvement by self-learning techniques such as neural networks. But on the other hand and to put it mildly, such a high degree of sophistication does not typically exist in insurance sales and marketing; in fact, most marketing segmentations are borrowed straight from pricing/technical segmentations.

Hence, there appears to be an use for AI to segment products, marketing techniques, and sales processes from a purely sales and marketing perspective. Perhaps the key insight here is that insurance clients are both buyers of insurance and insured risks; but their behavior as buyers and their risk characteristics should not be segmented in identical ways.