Emerging risks – time to consider algorithm governance?

Emerging risks – time to consider algorithm governance?

By October 6, 2020News, Uncategorized

Earlier this year, the New Zealand Government launched its “Algorithm Charter”. The charter establishes a set of principles government institutions are expected to follow, with the ultimate aim of mitigating the risks that emerge from increasing operational reliance on algorithms and building public confidence in algorithms. The Australian Government also released its own “AI ethics principles” in 2019. Both Australia and New Zealand’s initiatives represent a voluntary (and aspirational) set of principles to help businesses and governments build trust with algorithms as they start to design, develop and implement AI.

Government guidelines are an essential step….

We were pleased to see these initiatives as this area presents an emerging risk for insurers and society more broadly. Both sets of principles are consistent with the principles and approach endorsed by us following our research (published in the Actuaries Summit and Finity's Personal Lines Pricing & Analytic seminars earlier this year).

Why should insurers care?

Both the New Zealand Algorithm Charter and the Australian Ethics Principles come at a time when most sectors are investing heavily in the use of algorithms and automation in all aspects of their business.  The insurance sector is (and has been for many years) among the more advanced users of algorithms to assist with core business processes, not least those processes that influence front line customers directly. The Hayne Royal Commission brought the conduct of insurers into intense spotlight, particularly around the treatment of customers. Insurers are now intensely busy readying themselves for the risk management and compliance requirements coming out of Commission’s recommendations.

There has also been spate of public controversies surrounding high profile algorithms used across a range of industries.  Some examples include:

  • Australian social services (Online compliance intervention a.k.a. ‘Robodebt’).
  • The US justice systems recidivism algorithm (COMPAS).
  • UK A-level results.
  • Technology companies (Amazon’s recruitment algorithm, Google’s photo categorisation).
  • A bit closer to home – scrutiny placed on price optmisation practices of US and UK insurers.

Over the last couple of years we’ve also seen AFCA take a lot of interest in the way underwriting, pricing and claims algorithms impact customer experiences.

The increased regulatory burden has not, however, dampened insurers’ appetite for investment into algorithms. While algorithms and automation will be a key frontier in competitive insurance markets, like many frontiers, the risks leave insurers exposed to small but deep cracks to fall into.

Australian insurer research

Earlier this year, we examined the risks and issues around algorithm governance in depth. Our research relied upon:

  • An extensive industry survey supplemented by individual stakeholder interviews.
  • Post-mortem analysis of algorithmic governance failures (both publicly known and privately experienced).
  • Consultation with risk management experts.

Where’s the industry at?

In a nutshell, we found that:

What should insurers do?

It is our view (and perceived similarly by the industry) that it is up to individual companies to decide how best to establish governance responsibilities and how to embed good algorithmic risk management practices within their operations and throughout their organisations. If it’s not done well enough, or soon enough, regulators will eventually prescribe rules to us.

In an environment where insurers need to demonstrate they are acting honestly, efficiently and fairly, we think it is imperative that insurers do the following:

We encourage interested readers to read slide 12-14 of this presentation for an overview of the domain specific questions to explore to support efforts in better identifying, analysing and mitigating algorithmic risks. The publication focuses on the areas of pricing and claims, as these areas involve the most immediate customer impacts, but the concepts can be applied in a variety of other contexts.

Clearly, the issues and suggestions we’ve outlined here would involve a considerable amount of effort for any business to act on immediately, and it is not practical far any one individual to manage all of it. The issues need a collective and orchestrated effort across entire organisations.  To this end, there are a broader set of considerations around culture, structure, KPIs and management approach that will play into how different organisations embed these disciplines into their operations.

In the meantime, in terms of what we can all do as individuals in our respective organisations:

  • We need to maintain the enthusiasm and momentum that’s built around algorithms and automation.
  • Start to ask questions and probe into the issues we have outlined, raise awareness within the organization, and try to be a champion for the customers’ perspective where possible.

Like all successful initiatives, this one will need a combination of top-down endorsement paired with a grassroots effort from algorithmic, risk and analytics practitioners.

To discuss the above article, contact the authors: 

Ashish Ahluwalia

Ph +61 2 8252 3272

Marcello Negro

Ph +61 2 8252 3438