Explainable Artificial Intelligence and Price Influence with Elena Pizzocaro

0

A transformative change in pricing is almost tangible: the adoption of new technologies and data analytics can help insurers create long-term value. McKinsey spoke with Milan office partner Elena Pizzocaro to learn more about pricing and current challenges and opportunities.

McKinsey: What are the latest pricing requirements for insurers?

Helen: Pricing can be the source of sustainable competitive advantage and a key differentiator for long-term value creation in P&C insurance. Profitable P&C insurers emphasize and invest in pricing innovation and underwriting excellence.

We recently studied several pricing transformations at global insurers and codified different possible archetypes reflecting the maturity level of insurers and the strategic priority assigned to pricing. The most common archetypes include:

  • Consistent application of Generalized Linear Models (GLM) with emphasis on risk modeling
  • Use of AI and machine learning based pricing tools
  • Implementing robotic pricing to improve market understanding and transparency, and adjust prices automatically and dynamically (subject to regulation)

Large-scale pricing transformation focuses on putting the right infrastructure in place to achieve substantial and sustainable improvements across the entire pricing value chain, including underwriting strategies and risk selection, pricing technical, market-based and behavioral pricing, where permitted by regulation. Price execution and governance are equally important. The key factors here are organization and talent, as well as advanced analytical and numerical capabilities.

Embracing artificial intelligence (AI) throughout the pricing value chain means gaining the ability to recalibrate models at any selectable frequency and transform the operating model of the pricing team.


McKinsey: What opportunities does large-scale rate transformation offer insurers?

Helen: The opportunities for more sophisticated pricing and immediate P&L impact have never been better. The pursuit of rating sophistication can enable insurers transformative shifts toward advanced analytics, automation, new data sources, and the ability to respond quickly to changing market environments.

From a pragmatic perspective, embracing artificial intelligence (AI) across the pricing value chain means gaining the ability to recalibrate models at any selectable frequency and transforming the operating model of the pricing team. Actuaries can then spend most of their time improving data quality and identifying new sources of data, allowing senior management to easily understand pricing challenges through easy visualization and augmented intelligence. .

Insurers must not refrain from experimenting with advanced techniques and they must push the frontier of knowledge even further. For example, the application of deep learning, a subset of machine learning techniques, enables the processing of rich and complex datasets, including images and raw machine data.

McKinsey: How can insurers reconcile pricing sophistication with the need for “explainability”?

Helen: There is a growing demand for model explainability and an easier way to access and understand models, even from the perspective of regulators. Complex models can, at some point, be considered less transparent. Traditionally, this would involve a trade-off between model performance and explainability, and would result in prioritizing or eliminating certain types of models from analysis.

There is a growing demand for model explainability and an easier way to access and understand models, even from the perspective of regulators.


Nowadays, state-of-the-art methods aim to explain every post-hoc sample, effectively overcoming the trade-off between interpretability and performance. In particular, “Explainable AI” (XAI) focuses on the development of post-hoc explanatory methods that can be applied to black-box machine learning models.

This typically involves creating additional explainer algorithms that augment the machine learning model, providing insight into how certain predictions or outputs are achieved. If high prediction accuracy is paramount, using a black box model and explaining it after training and optimizing with the explanation algorithms is the best performing approach. In this way, AI is not a science reserved for data scientists, but becomes intelligible and contestable for a wider group, including insurers, internal and external stakeholders, actuaries, product developers, business strategists, risk managers and intermediaries.

***

Elena Pizzocaro is a partner in the Milan office.

For more information on insurance pricing, see:

  1. The Post-COVID-19 Underwriting Imperative for P&C Insurers
  2. Harness the power of external data
  3. Insurance 2030—The impact of AI on the future of insurance
Share.

Comments are closed.