Businesses today can readily access massive volumes of increasingly precise data from a widening variety of sources. For Property risk engineers, this profusion of data presents many significant opportunities – more data = greater understanding – as well as some substantial challenges; as the cliché goes, it’s like trying to drink from a fire hydrant.
Fortunately, new tools powered by artificial intelligence (AI) are helping to unlock the opportunities and lessen the challenges.
More productive and effective
One especially promising use of AI in Property risk engineering involves machine reading. Software engineers and social scientists have developed ever more powerful algorithms for determining the exact meaning of a word or phrase given the context, and also for recognizing the relationships between different facts or characteristics.
In practice, that means Property risk engineers can now use AI to review, analyze and summarize the realms of material that is available about a client’s, or interested prospect’s, properties.
Say, for example, AXA XL is asked to quote on a global Property program for a major multinational company with several hundred production facilities worldwide. The underwriter working on the quote turns to our risk consultants to assess the prospective client’s overall risk profile and highlight specific risks at particular facilities. One of the challenges, however, is that the documentation runs into the thousands of pages with individual reports covering, e.g., location, construction details, occupancy information, specifications for the protection systems, qualitative insight on management practices, data on exposures to natural catastrophes, and so on.
Previously, the risk consultants would structure the review based on assumptions about what is most relevant for the underwriter. In many cases, that meant prioritizing the client’s most high-value properties. It is often the case, however, that because of their high values, the protection measures and risk management programs implemented at these locations are already quite robust. And the story can be much different at smaller facilities where the protection systems aren’t as substantial, even though they are a critical part of the company’s supply chain.
That’s changing. A risk consultant today can start the process by using AI to conduct an initial review and analysis of that mountain of information. Based on the parameters we’ve built into the software, the AI tool we’ve created with an external partner returns a summary report highlighting the most relevant risks by location and by type. With that initial analysis, the risk engineer can then drill down on the most salient issues as flagged by the system. In other words, AI helps our risk consultants to be more focused and productive by providing informed guidance, based on a complete, contextual review of a realm of data, on where he/she should dig deeper, and what to look for.
Consistent with our “from payer to partner” approach, we’re also taking advantage of these new capabilities in our ongoing work with clients to help them better manage their exposures and reduce the potential for a loss. Our Property risk engineers, for instance, make thousands of site visits a year. These are extremely useful in developing a first-hand understanding of a client’s properties and in helping them to implement practical and cost-effective measures to protect their physical assets.
Here, too, contextual analysis of diverse data on a client’s operations and exposures is helping us to prioritize site visits so that we deploy our engineers to those places where an on-the-ground risk assessment will have the most value.
Another exciting opportunity for capitalizing on advances in AI is by adding a probability dimension to our loss estimates.
Currently, risk management and underwriting decisions are based on estimated maximum losses, e.g., MFLs (maximum foreseeable loss) or NLEs (normal loss expectencies). And these are usually huge numbers, especially with, for example, a mega-warehouse filled with high-value inventory. But the reality is that the likelihood of an event generating losses close to the estimated MFL or NLE levels is very low, especially considering the protection systems that are common in these facilities.
But what if we were able to produce, with a high degree of certainty, more precise estimates of the probability of different levels of average annual losses? Armed with this greater insight, CFOs and Risk Managers should be able to more closely tailor their risk management and mitigation strategies and budgets to the actual losses the company can expect. That’s a very different discussion, and mindset, compared to one focused on MFLs or NLEs.
Insurance is one of the very few industries in which the full production costs – in our case, that’s the losses we pay – aren’t known until some point in the future. That’s why we devote enormous resources to trying to anticipate the future. We’ll never have the perfect crystal ball, but with increasingly powerful computers coupled with progressively sophisticated AI, we can better support our clients by foreseeing different events with greater confidence and clarity.
Given the scope and complexity of many global supply chains today, pinpointing potential vulnerabilities and staying abreast of the changing fortunes of hundreds if not thousands of suppliers is extraordinarily complicated. Moreover, it’s not unusual today for virtually all of the companies in a particular sector to rely on the same two or three suppliers, especially for specialized components or materials. This means that if one goes offline, for whatever reason, the impacts can disrupt production across the entire industry.
One solution: AI-enabled software that continuously trawls the internet for news and public information about a company’s suppliers, including peripheral suppliers who may only provide a single part or ingredient.
The benefits of this approach are twofold. Firstly, AI can help a company identify an exposure it might not be aware of. An automotive company, for example, can hoover up all of the available information about its suppliers and discover that there are only two mills in the world producing the fibers used in seat belts and, perhaps even more importantly, all of its competitors also rely on these same sources.
Secondly, this capability also can be used to alert a company in advance about a potential disruption. In this case, the data feed on the company’s supply chain can include a real-time overlay of events in different locations. So when a major hurricane is about to strike, for instance, the company can quickly identify the OEMs in the area that could be affected and take immediate action to mitigate the impacts.