On the train home from BIBA 2017 I came across yet another article predicting job losses to automation, AI and machine learning. The piece was written following Warren Buffett’s annual Berkshire Hathaway meeting, where he apparently stated his belief that
AI will endanger some jobs. It’s not the first one I’ve seen – in January,
a study by PWC found that 30% of jobs are at risk by the 2030s and not wanting to be left behind, even the insurance industry is getting in on the act; earlier this year, Japanese life insurer
Fukoku Mutual replaced 34 claims staff with the IBM “Watson” platform. Could the same happen to the insurance industry? Are we about to witness a rise of the machines? And could it be that it’s been happening already?
“Automation anxiety” (as some people would call it) is nothing new; indeed, it has been studied in academic circles for some time, with MIT economists Daron Acemoglu and David Autor being just two of the better known (ie published) researchers in the field. Both have written for a while on the subject and most of their papers are actually quite readable; an example is Autor’s 2015
Journal of Economic Perspectives paper titled “Why are there still so many jobs?”, one of many research articles on the topic.
My initial “instinct” was that AI (and technological change) in insurance would lead (or have already led) to job losses, especially when you consider longer time periods of two decades or more. After all, we’ve all heard of the “man from the Pru” but that market has moved on (for non UK audiences – a reference to the old practice of selling life insurance on the doorstep). With that in mind, I decided to try and find some figures, and in doing so I realised yet again why relying on “instinct” is generally a bad idea. I decided to look at three of the largest insurance markets in the world – US, UK, and Germany. My decision to omit countries with large number of insurance industry employees (Switzerland, South Africa, Canada, China etc) stemmed from my desire to have a readable piece with a narrower geographic focus rather than the other way around.
UK
The Association of British Insurers (“ABI”) has publishes a year “Key Facts” document that estimates the number of employees in UK insurance; the first one I could find was 2012, the latest 2016. The data seemed to consist of both estimates and official stats so I decided to go to the source and download the relevant official time series from the UK Office of National Statistics website. For anyone interested, the relevant time series is called
“Jobs03” and it breaks down industry classifications into subgroubs, ie breaks insurance and financial services into: 1) financial services, 2) insurance/reinsurance/pensions and 3) auxiliary services.
Using that data (going back to 1978) and using Tableau for visualisation, I have overlaid two separate trends – insurance, reinsurance and pensions jobs against auxiliary jobs to insurance and financial services. The y-axis represents the number of jobs in thousands.
Take-away: clear reduction in the number of people directly employed, offset by a boom in auxiliary jobs, probably caused by regulatory developments such as Solvency II. There is one caveat – the auxiliary jobs are auxiliary to all financial services, so the number of auxiliary jobs dedicated specifically to insurance is much harder to estimate. The reduction in direct employment could have many reasons; one of them might be the move away from selling insurance via agents and call centres to selling it via price comparison websites.
Germany
Germany’s insurance industry association (“GDV”) publishes a yearly statistical factbook that includes estimates of the number of people employed in the insurance industry, which I have reproduced below. Please note that “Employees” in my table refers to what the GDV calls “employees subject to social security contributions in primary insurance and reinsurance companies and insurance intermediation firms as at 30 June of the respective year”, whilst “Agents” is my shorthand for GDV’s definition of “self-employed insurance intermediaries/advisers”. Blank cells indicate no data.
Take-away: although not hugely detailed, the size of the German insurance workforce has actually increased since 1990 (by about 16%) although the number of agents has reduced, certainly in the last 7 years, both in absolute terms as well as a percentage of the total workforce. In fact, the “employee” numbers have stayed in the 290,000 – 300,000 range since 2000.
USA
The US Department of Labor produces very detailed time series but thankfully the Insurance Information Institute publishes some very good insurance industry “employment trends” reports, with the latest available being March 2017. The report is highly granular, breaking down employment trends by insurance industry sub-sector (direct P/C, life, medical, reinsurance, agencies & brokerage, independent claim adjusting and third party administration of insurance funds). I’ve only chosen to reproduce two slides and full credit to the III and in particular thanks to Dr Steven Weisbart, Chief Economist, for allowing me to reproduce them.
The job losses in reinsurance seem very dramatic when looking at the two slides together but we must remember the y-axis has very different ranges. I would have recalibrated but as I didn’t have access to the raw data I just used the III’s slides instead.
Another useful resource is Statista, who have an online visual covering an even longer time period (1960-present). I suspect their data is the same as the III’s.
Take-away: the US insurance industry has employed over 2 million employees since 1990 and between 1998 and 2013, it stayed in a relatively narrow band (2.3-2.4 million, according to the III). In 2014 it rose above that band and it seems to have now broken the 2.5 million barrier.
So why the continued creation of insurance jobs in the US? Even Germany has 16% more insurance industry jobs than it did in 1990. As much as I’d like to come up with a snappy conclusion, it’s difficult to do so, especially when there is no universal way of measuring the number of employees within the insurance industry, making analysis and predictions much harder. That being said, my personal view is that the “industry” tends to vastly over-estimate the pace of change. I don’t think overall insurance industry losses are going to be anywhere near the numbers speculated on, though I accept certain practices have now gone (door to door selling of life insurance, in the UK), with the resultant loss of some jobs.
Overall though, many of those positions have been replaced with a plethora of “knowledge workers”, driven by both regulatory changes (eg Solvency II in Europe, Sarbanes Oxley etc) as well as technological progress, especially with respect to the availability of data to make analytics decisions. I went to the Reinsurance Association of America Catastrophe Modelling conference in Orlando a couple of years ago and was surprised by the sheer number of vendors of weather data. Twenty years ago, most syndicates at Lloyds did not have an in-house chief actuary, and virtually all of them do now; in fact their actuarial and exposure management teams have begun to resemble small armies, much to the consternation of CFOs keen to keep costs down in a market of depressed rates.
My prediction for the future? Probably not rocket science, but I think the “knowledge economy” will continue to dominate job creation, and any potential job losses to AI will be more than supplemented (in the aggregate) by the creation of highly skilled positions requiring strong analytical and digital skills (regardless of how those jobs are “badged”). New hazards are coming online (excuse the pun!) all the time; witness the recent cyber attack that, among other things, affected the UK’s national health service. Some insurers are virtually “betting the house” on cyber but it’s not just digital hazards that need looking at. Copenhagen Business School published a report some time ago outlining the challenges and opportunities offered by Arctic shipping, suggesting that “Arctic liner shipping may become economically feasible by 2040 if the ice covers continues to diminish at the present rate”. It’s hard to say what hazards (physical or otherwise) will need to be insured two decades from now, but I would expect they will need to be analysed, most likely with a combination of human and computer input. It’s not unfeasible, for example, that underwriting jobs will become much more data driven – indeed, data entry, the staple of a trainee underwriter’s work, could well disappear over the next decade as electronic insurance placing platforms become the new normal. AI systems could replace parts of a job (data input and cleansing, for instance), but I would expect the analysis to still be human-led, if only because computers aren’t built for the political environment that most corporations have (and will continue to have).
This has important ramifications for educational policy and virtually every academic paper on the subjects accepts that, even if job numbers don’t change (in the aggregate), there will be polarisation between the people who benefit (overwhelmingly educated, with digital and analytical skills) and the “others”, especially if the “others” perform very routine jobs with little value added. Discussing what needs to be done to ensure workers have the right skills for 2020 and beyond is outside the scope of this article but I feel caution is needed when reading overly pessimistic reports predicting the demise of X number of jobs by year Y. Stanford University produced an excellent report on AI some time ago and suggested “AI will likely replace tasks rather than jobs in the near term, and will also create new kinds of jobs. But the new jobs that will emerge are harder to imagine in advance than the existing jobs that will likely be lost.” I feel much of this parallels the Hollywood movie industry – technology has been a boon to companies involved in animation and/or VFX. I watched “Guardians of the Galaxy Vol 2” a few days ago and it seemed like the credits rolled on forever.
Perhaps a decade from now, we will have more chatbots and automatic settlement of claims. It’s inevitable that some jobs will be lost, perhaps even affecting incomes and salaries as a higher supply of available workers competing for fewer jobs. But someone still needs to program that chat bot in the first place and the complexity of today’s digital world makes it likely that a whole team will be needed…
