3 leading AI concerns for insurers

Register now

On the 11th of October Celent hosted an invite-only roundtable discussing AI in insurance. In contrast to the earlier event, this one hosted professionals from the IT and architecture teams within insurers rather than data science professionals themselves.

In this roundtable, the discussion centered much more on enabling AI within an enterprise. The challenges discussed had a great deal of overlap with those of the data science community, namely:

  • Productionising AI applications once they’re built
  • Engaging with Data Scientists
  • Education about AI and its capabilities

Of course this group had a slightly different take on these three points. I think it’s worth taking a little time with each one.
While I’ve phrased the first bullet as productionising AI, actually the habits of budding insurance data scientists pose significant issues for their colleagues in IT. The preference for open source solutions, and the latest and greatest open source solutions at that, presents a challenge for IT governance -- particularly where support and security professionals need to stay on top of an ever larger and more complex technology stack. These challenges stretch from desktop deployments (we’ve been asked to support Linux for the desktop) through to governing appropriate access to data through to what is run and supported in a production environment.

A balance of some standardisation and a fast route to supporting new tools where necessary seems to be the sweet spot here, although as always the devil is in the detail.

On this topic, one last note. I was asked, “Is DataOps just DevOps but with Data swapped in?” Well, in a way, the simple answer is yes. We discussed how an AI application or a model is typically a small deployable piece of code with an API (some tools deliver models with REST APIs; we didn’t dig into the nuances of “API”). Such a component can be readily deployed with most insurers' existing investments in DevOps, including all the good practices around testing, automation, roll back and so forth.

My personal view is IT could be a great asset and enabler to an AI or data science function, albeit it requires some education and empathy on both sides, which brings us neatly to …

Any organisation dealing with highly talented individuals has a challenge ahead in retaining them, challenging them, finding cool projects, etc. This was a topic discussed at some length among data science leads -- but isn’t an issue just for that team. Insurance is an industry with an abundance of talent.

However, talented individuals who want to deliver quickly, who are IT-savvy and time-poor, are particular challenges for an internal IT department. Discussions regarding governance structures, security procedures, and evaluation procedures for new applications (or open source libraries) can all be seen as needlessly slowing down good work within the insurer -- even if they do serve a purpose to protect the enterprise, its customers, and its shareholders/members.

Much like early digital teams a couple of years ago, a lack of desire to even engage with IT or to deliver change through shadow projects were discussion points around the table. This is a great shame because the rise of digital programs and adoption of automation and DevOps can be a great enabler for the data scientist and AI teams.

Some attendees were further along the journey than others, and this feels like a maturity point, rather than an inevitable pothole all insurers must fall into.

Some of the answer is to increase the level of education about AI.

From a week spent with various parties in the insurance engaged in or on the periphery of adopting AI: Education and increasing understanding of AI feels like a critical step now.

For IT professionals, this is as always another example of IT sharing and selling their capabilities and what can be done with the rest of their colleagues, including new ones. It is partly IT’s role to educate on the power of AI, but also on how to deliver change quickly by engaging with them.

As noted previously, for AI in insurance the conversations have moved beyond the use cases and now on to the pragmatic details, the nuts and bolts, of delivering AI-empowered insurance organisations.

Again, do take a look at our reports looking at the global insurance position on AI, the use of DataOps for productionising AI applications, and the rise of citizen data scientist communities.

This blog entry has been reprinted with permission from Celent.

For reprint and licensing requests for this article, click here.
Artificial intelligence Machine learning