Notes from the research frontier: characterising AI innovation clusters in the Insurance Industry
As insurance becomes increasingly digital, the issues of AI adoption become increasingly sector-wide. In October 2020, we published our first white paper exploring the role of start-ups in the insurance knowledge space. In this article, we provide a brief overview of our approach and what we found out.
Our knowledge gap: what we wanted to find out
In December 2018 UKRI decided to fund us studying the insurance industry. More specifically, they wanted to know how the insurance sector might transform through application of Artificial Intelligence (AI). This request came with a respectable research grant. So, you know… we jumped into the water, engaging with tier one consultancy companies, insurance incumbents that commissioned in depth reports to strategize their next steps, global government bodies, associations.
We identified a common thread amongst all these stakeholders. They were all in their different ways trying to figure out an answer to the following question “What are the common features linking together all the ongoing Insurtech AI innovations and activities?”
How we addressed this knowledge gap
Early in our research we realised that in this rapidly moving industry, our findings could be out of date by the time they are published. Start-ups come and go. Industry boundaries blur so much, to the point that we can no longer always clearly decide which companies are in fact insurance and which are not. There might be companies below our radar that possess the revolutionary technology and business model but right now simply don’t meet our current definition of ‘insurance’. Indeed, one can miss out on finding the next ‘Lemonade’.
Therefore, instead of chasing shadows, we decided to track and understand the knowledge possessed by different AI startups themselves. Not that such an approach made our lives much easier… as innovation studies scholars we that tracking knowledge is a challenge. Unlike tangibles that are easy to quantify, in stock or flow, intangible knowledge is difficult to quantify.
Some scholars go to patent data to track ‘spillovers’ or track innovations such as rapidly evolving technologies, transformation of industries, or how IP percolate to specific industries. However, the “patent data path” has its weaknesses.
First, patent data only selectively represents the technological frontier. Particularly when it is the adoption of an already existing technology, it might not be reflected in new patents. Moreover, not all innovations are patentable. For instance, unpatented open platforms, such as GPT-3, that offer free software libraries will have transformative effects in the use of AI-related technologies, yet we would miss out on them. Also, patent-based business models are not necessarily the common way to create a competitive advantage, particularly not in services like insurance.
Second, patent datasets impose a sectoral straitjacket, i.e., patents may not reflect the eventual application sector of a technology. Third, because there is a lag between patent data and the innovation they document, it is difficult to track the emergence of a technology in near real-time, but only analysing few years back data, such a framework is too wide in some cases. We resolved instead to use AI to study AI (to be more accurate, we used big data analytics to study AI with some modest algorithm use, but ‘using AI to study AI’ sounds too good not to use). More specifically, we created a ‘Knowledge Space’ a concept used in innovation studies, using near real-time data driven algorithms to generate insightful understanding from a large startup dataset (The Crunchbase database https://www.crunchbase.com/ ). Access to an extensive database such as the one we use, means that we do not necessarily have to rely on pre-structured technology or industry classifications. Instead, by using a tagging system, we can classify companies into any combination of industries and/or technologies and as the companies evolve, the classification evolves with them. Therefore, these databases not only address the issue of blurring boundaries between industries, they also provide a dynamic picture of where technology is being applied. Databases such as the startup database we use can also document changing business models and track their funding stages in near real-time.
What we found: visualising relationships amongst categories of knowledge
Some of our main findings is presented in Figure 1. AI is part of a cluster of Artificial Intelligence and related technologies. Moreover these are clearly related to a wider set of key General-Purpose Technologies (GPT), that together create a wider family of AI-Related Technologies.
There are six distinctive modules of technology in our InsurTech knowledge space, relating to HealthTech, Blockchain; AI and Machine learning; PropTech, Finance and Banking, the Internet of Things, and Auto Insurance. Within this ‘backbone’ of technologies three related “AI-type” categories play a major role. Categories of Artificial Intelligence, Machine learning and Big Data were found to be strongly related key nodes in the network.
Geographical and structural relationships
‘Our knowledge space” also provided additional insights. The digital transformation is already quite mature in insurance and even starting to consolidate. We see less start-ups entering the scene in recent years compared to three or four years back. We also see clear geographical patterns where distinctive types of start-ups are found. These geographical patterns correspond to theories in economic geography around regional specialization and sector coalescence. We found that the US pattern of metropolitan areas that has a significant population of InsurTech start-ups follows a polycentric structure, while in Europe it is London that dominates the scene. Although secondary centres in the EU do add up to a significant critical mass, its geography is much less spiky than in the US. Of the smaller centres, Tel Aviv stood out in the analysis as an exception to this pattern where the start-up population has remained vibrant in recent years.
Within this network, it is remarkable that there are limited direct linkages in the knowledge space with categories designated with the financial sector and the core AI segment of the AI cluster. Rather, most linkages are intermediated through “software as a service” (SaaS) which seems to act as a bridge.
However, the direct relationships between the AI technologies and HealthTech and Auto Insurance seem to be stronger than with Finance. This analysis again fits descriptions of dissimilarity between the Insurance sector and some of the technologies that it wants to integrate as part of the digital transformation. The role of SaaS suggests there might be some kind of “platformization”, where third party software platforms offer AI solutions to the financial sector in a modular kind of way.
With respect to investors and the role of accelerators, the most important finding regards the presence of accelerators as key “glue” in the investment pattern dataset. Thus, we believe that accelerators could play in important role in the strategic coupling of AI start-ups with the established insurance sector. We also found some evidence that investors in London would have a stronger propensity to invest in the Blockchain and PropTech modules and less in the HealthCare and Auto insurance sectors. San Francisco shows a somewhat reverse pattern.
We recommend that policy makers shift their perspective from an ‘Industries’ view to a ‘knowledge space’ view, focused on the connections and synergies between industries and sub-industries. Such a shift in viewpoint allows to identify skillsets and knowledge fields that are transferable between industries or act as bridges between them. Future research needs to be knowledge rather than industry focused and by identifying the right clusters, focus could be on smart specialization.
The role of incubators and accelerators programme is significant for InsurTech start-ups to scale ups alike. These institutions signal the market about start-up potential. Because start-ups have been through the drill of an incubator they gain the skills, experience, know who and know how to secure more funding possibilities. The evidence shows that being part of incubator programmes helps getting access to funding. Hence, stimulating incubation of start-ups could have outsize effects on bringing new elements of the knowledge space to maturity
Policy could also consider closing the knowledge space gap between investors and companies. Policy makers need to understand the gap between investors’ location and expertise versus the InsurTech companies’ knowledge space and their location and see how they can stimulate that more bridges are created. This could entail stimulating cross-industries knowledge flow. This might generate opportunities for companies that provide technological solutions in one industry to find opportunities in other industries. The cross-industry approach can expand the business proposition, increase streams of revenue for the tech companies, develop new stream-line channels and expand innovation opportunities for incumbents.
The knowledge space provides a dashboard how particular financial centres compare to its peers. Each knowledge space can be benchmarked with other InsurTech strong clusters in other places in the world. To keep those clusters’ competitive advantage there are three elements that will require inflows: Maintaining and developing the local skills and stimulate the local knowledge space; securing and encouraging funding opportunities; providing network opportunities within the industry and cross industries.
What comes next?
Since our report was first written, we have been expanding our work to cover other advanced services sectors: watch this space for updates!
If you are interested in our research please contact Dr. Tzameret H. Rubin T.firstname.lastname@example.org
Tzameret H. Rubin, Michiel van Meeteren