Artificial Intelligence and the Future of Underwriters
Part I: Artificial Intelligence (AI)
The phenomenon dubbed “artificial intelligence” could (and likely will) be one of the most impactful in terms of how our profession’s future sorts out by the mid-2020s.
Every year, the January issue of LOMA’s Resource magazine leads off with a lengthy article recounting the views of a dozen insurance company executives, along with those of what appear to be “strategically selected” technology consultants.
The contributors opine on a range of issues including two (information technology and operational challenges) that evoke responses germane to our profession.
The preamble of the 2017 installment speaks to how “intelligent underwriting technologies…have the potential to shorten the new business cycle, cut expenses…” and so on.
The senior VP of a major carrier says “…like most companies, we are working on [technology] capabilities, especially in the area of intelligent underwriting.”
A consultant whets the appetites of prospective customers by chiming in with “…the impact and benefits of machine learning on risk assessment are one of the major developments we observed in the last 12 months and we are investing heavily in this area…”
To understand why heads are being turned and predictably Pavlovian responses elicited at the mere mention of AI, consider these recent revelations:
A factory producing mobile phones in China recently announced that AI replaced the lion’s share of its workforce, increasing productivity 250% and reducing the incidence of defective merchandise by 80%.
Only 60 out 650 human workers were retained, chiefly to monitor and finely tune their system.
Meanwhile, a Japanese insurer embraced an AI system with the capacity to meticulously dissect “medical certificates” as a prelude to paying claims.
Once again, top management celebrated their success with steep staffing cuts.
These and similar accounts of AI payoffs are fueling the imaginations of those who covet equivalent returns from deploying AI for the determination of insurability.
A commentary in the The Guardian UK (January 6, 2017) pointed out that embracing AI centers upon breaking down complex problems into simple algorithms suited to machine intelligence.
It is one thing to swiftly sort and then analytically process objective data and quite another to translate these data into adequately nuanced assessments of their implications in real world settings with subtle and often subjective issues at play.
We are not the only profession that must come to terms with the potential impact of AI.
In their December 13 Journal of the American Medical Association essay on adapting clinical medicine to AI encroachment, Jha and Topol offer this reassurance to their physician peers:
“Jobs are not lost; rather, roles defined; humans are displaced to tasks needing a human element”
The roles of both doctors and underwriters will change to a greater or lesser degree. On the other hand, while physicians’ jobs may not be lost, the present density of our professional community cannot be sustained in an AI-driven environment.
An industry CIO made this relevant observation in the February 11 issue of LOMA Resource:
“In the future underwriters will do less actual underwriting and servicing and they’ll become decision-makers and decision-arbitrators as required instead.”
Three executives with Accenture, writing in the Harvard Business Review (November 2, 2016), noted that when the Associated Press put AI to good use, it allowed journalists “to conduct more investigative and interpretive reporting”…
…which, when you think about it, is broadly akin to what we do when reconciling insurability issues on complex cases.
These gentlemen recommend adopting AI “…to automate administration and to augment but not replace human judgment.”
What is the Achilles heel of artificial intelligence?
When asked if AI has any shortfalls, a Silicon Valley technology innovator replied succinctly that it “…can’t write poetry.”
Which is not to say AI cannot string together words in a manner imitating the form of poetry…and even “get the rhyme right!”*
Poetry is art, not to be confused with limericks or clever ditties used in a marketing campaign.
In the context of this essay, poetry is akin to the art of underwriting and thus the domain of risk assessment immune to usurpation by machine thinking.
The art of underwriting requires intuitive “right brain” skills that are vested in humans, not software programs.
And therefore – despite the misconceptions of productivity obsessed, underwriting insight-impoverished new business executives – this critical aspect of what we do has precious little in common with AI-friendly factory and clerical occupations.
Part II: Three Types of Underwriters
How will AI impact the future of underwriters?
To put this in perspective, I would argue that one is best served by carefully distinguishing between 3 types of underwriters.
Type 1 Underwriters
Type 1 underwriters survive the culling process inherent in initial training by thriving largely on cases more appropriately adjudicated by an underwriting engine.
Their limitations come into focus when they struggle to sort out what truly matters from the morass of information at hand in more complicated, multifactorial cases.
They fail to acquire a sufficient knowledge base and the necessary skill sets to master the science of underwriting.
They have no clue whatsoever as regards the art of underwriting.
They parasitize learned colleagues relentlessly and ad infinitum.
They forward files up the referral pyramid or on to medical directors without making a viable recommendation as regards the applicant’s insurability status.
We know 3 things about type 1 underwriters:
- Most eventually wander off into other occupations.
- Those that remain will predictably languish at authority levels increasingly incongruent with the duration of the time they have spent as underwriters.
- They cannot survive the advent of AI in underwriting.
Type 2 Underwriters
Type 2 underwriters are skilled in the science of underwriting, as evidenced by their ability to identify most if not all objective factors germane to insurability on a given case.
They are adept at tracking down these factors in their manual (notwithstanding the fact that their comfort zone is best served by misconstruing guidelines as rules).
They hold their manual’s risk calculators in high regard because these dumbing-down devices cater to their rule vs. guideline mindset.
Their limitations emerge when attempting to see the forest for the trees.
They are accorded the authority to credit certain applicants with normal NT-proBNP readings. Nevertheless, they are likely to fail to appreciate how this might apply to an asymptomatic, otherwise healthy 60 year old with a bicuspid aortic valve and isolated “mild-to-moderate” aortic insufficiency discovered by ultra-sensitive Doppler ultrasonography.
They struggle to integrate relationships between the objective and the more subjective elements of a given case.
For example, they cannot consistently connect the dotted lines between age at diagnosis, level of education, occupational and social histories, (more subtle) risk-related behaviors, prescription history and so on in order to deftly distinguish between the two young adults that both have a diagnosis of attention-deficit hyperactivity disorder.
We know three things about type 2 underwriters:
- They account for, by far, the largest % of current life underwriters.
- The art of underwriting eludes them.
- AI poses a clear and (soon to be) present danger to their future.
Type 3 Underwriters
Type 3 underwriters have mastered the science of underwriting and are fortuitously endowed with the intuitive skills needed to integrate this with the art of underwriting.
Like savvy detectives, they see both the straight line and the more elusive dotted line relationships between components of a case so they can self-confidently reconcile scenarios where tree vs. forest perspectives differ.
They are not reined in by their manual, which they recognize as an aggregation of guidelines subject to modification based on all the evidence at hand.
For these reasons, they can readily sort out the differences in insurability status between the two aforementioned applicants with ADHD.
We know 3 things about Type 3 underwriters:
- They cannot be distinguished from their peers on the basis of formal education, years of experience or other garden-variety objective criteria. Their advantage is being able to bring intuitive skills to bear when assessing risks.
- They represent somewhere between 15% and 25% of all life underwriters.
- AI poses no direct threat to them…that is, so long as they work in a new business operation where the art of underwriting is recognized and highly valued.
Prior to the mid-1990s, extended mentoring of new underwriters for as long as 5+ years was commonplace.
Twenty years later, protracted mentoring is all but extinct and we have paid a steep price for pulling the plug on this practice.
That price is reflected in comments by chief underwriters at our study groups regarding their candid perception that current underwriters with 10-15 years’ experience are notably less skilled than those with the same experience level were back in the 1980s and 1990s.
The good news is that there is presently a sizeable share of type 2 underwriters with the latent (but as yet unrealized) potential to become type 3 underwriters.
The best advice one can offer type 2 underwriters is to muster all the energy they can and channel it into becoming type 3 underwriters.
Chief underwriters can also play a key role here by expanding the use of mentoring as well as making greater use of the case clinics; that is, if conducted solely by individuals with type 3 underwriter skills.
So it goes.
* This phrase was shamelessly hijacked from the brilliant English film Educating Rita, which chronicles the encounters between a working class homemaker with an unquenchable thirst for learning and an empathetic professor with a rather florid alcohol problem.