Flood risk AI is now a reliable tool- is there a desire to put it to work?


Flood insurance has been a wallflower at the coverage dance- an eager participant but not able to find a suitable partner.  Innovation efforts have found suitable risk prediction partners for carriers- FloodMapp, Hazard Hub, and Previsico among others- but is the flood insurance market ready?

Politics, inertia, customer preferences and regulation might keep the music from playing.

Patrick Kelahan is a CX, engineering & insurance professional, working with Insurers, Attorneys & Owners. He also serves the insurance and Fintech world as the ‘Insurance Elephant’.

Flooding causes damage globally in tens of billions of dollars/euros per year, and the bulk of that amount is uninsured.  Choose your source for insured/uninsured percentages- the conclusion is the same- flooding is a risk with global effects, and flooding is a risk that is globally accepted as difficult to underwrite.  Flooding is hard to predict, has regional effects involving huge numbers of properties, and subject to moral hazard perspective by property owners.  Consider these indicative data from McKinsey :

ins gap

Hard to predict is a real problem-if the probability of risk cannot be reasonably calculated, and the extent of possible risk is elusive, that sure makes an actuary’s job difficult. In the U.S. those problems were finally acknowledged as nonviable issues for the private insurance industry and a government program was established in the early 1970’s to address the risk, the National Flood Insurance Program (NFIP).  The underpinning of the NFIP was flood cover that was limited, was driven by a property’s location within flood maps drawn by risk exposure over time, and by the relative elevation of a property to risk (creeks, streams, rivers, wave and surge, etc.)

Fast forwarding to today that program morphed into a moral hazard, subsidized, under-funded monster that even in high risk areas had a low take up rate among property owners.  Mapping data became political footballs (no one wants a property that is in a designated Special Flood Hazard Area, SFHA), premiums became unbearable due to the effects of moral hazard/adverse selection, and outside of mortgagee requirements there is little societal or economic coercive pressure for a property owner to obtain the cover.  SFHA- a one in a hundred probability of a damaging flood?  I’ll take those odds and my chances, most property owners said.  And over time?  Persons who obtained the cover and suffered losses often simply repaired and waited for the next event, being indemnified multiple times without any underwriting consequence other than premium creep.  Those who had no cover either self-repaired or waited for government support in the form of grants or low-interest loans.

Human nature prevailed after 2005’s Hurricane Katrina, when flood/surge losses exceeded $100 Bn, and the number of US flood policies peaked the following few years to 5.7 million, only to decline more than 10% since then (see below):


And here’s a well-kept secret of the NFIP- claims can be adjusted only by flood-certified adjusters due to policy differences from private plans, and of course the bureaucracy’s zeal to restrict how flood funds are spent.  Subsidize the heck out of the program, but restrict efficiency in handling when indemnity is needed to jump-start recovery.

What of  countries other than the US?

There are few government subsidized flood programs elsewhere; most flood programs that are available are tied to property policies as extensions of cover (hello again, adverse selection issues) or bundled cover with more traditional policies.

An exception to lack of subsidized cover is the UK’s Flood Re initiative that cedes risk in excess of premium calculations to a not-for-profit organized as a backstop for carriers that is in the end supported by all insurers by market share.

Parametric options are becoming more accepted as alternatives to the complex nature of flood indemnity predictions.

So, it’s agreed that flood damage is a growing problem, and options for carriers and customers have been few.  However- if the probability of flood risk can be assessed and ‘tightened’, does that not move flood risk into a more traditional risk role?  The ongoing issues with US flood maps are that not only are the maps supported primarily by property elevation data that may have been influenced by politics, but that the mapping results are static until the next official surveys, and the mapping data are narrow in their basis- elevation, proximity to the water risk, and calculation of risk probability that is always backwards looking.

Consider some of the risk assessment/prediction innovators:

  • FloodMapp, a real-time flood prediction software platform that “uses proprietary technology for large-scale, rapid predictive flood and hazard modeling.” Large scale and rapid, words that have not been typically found in the flood risk lexicon.  And here’s a real innovation- the real-time aspect of the tech can be leveraged by insurers to advise policyholders of flooding potential before it actually affects properties, allowing insureds time to mitigate potential losses, and while the company’s information does not touch on this aspect, it’s important- potential reduction in immediate danger to residents. FloodMapp has proven its concept with insurers in the US and globally, and will roll out its primary products, Predictive Mapping (flood modelling to predict an extent of flooding), and Hazard Mapping, (tool for underwriters to apply in determining the risk of a particular property).  As co-founder Juliette Murphy suggested, “if we can have an app that tracks when our pizza will arrive, with proper, detailed flood risk data we should be able to have an app that tells when a flood may arrive.”
  • Hazard Hub, a company that promises, “the most comprehensive risk data ever created.” Whether that promise can be proven, the company can prove that for multiple risks (air, fire, earth, water, and man made risks) its proprietary ‘scoring of properties for exposure to any or all of these risks is comprehensive and accessible.  Yes, the firm does want to be a carrier’s go-to tool to aid in underwriting, but it also wants to be a risk understanding tool for individual property owners through its com website, where the firm will provide a free risk analysis for dozens of risk category, assigning the respective property a score of ‘A’ (low or no risk) to ‘F”, substantial risk.  In terms of comparison with traditional flood maps, Hazard Hub’s analysis will narrow the risk measure to the subject property as opposed to a broad assessment, potentially advising that a property in an ‘X’ flood hazard area (low risk) may have elevated risk due to its immediate geographic placement.  It educates customers and agents to allow better insurance decisions.
  • Previsico, a newer entrant whose purpose is, “produc(ing) round the clock street-level flood risk predictions and analytics. Uniquely, these are continuously modelled and updated using a combination of different weather forecasts. This allows us to map the likelihood of short and long-term surface water events in real-time to the street level. Real-time data production, working to allow property owners to protect structures and residents. The company has a robust staff of data scientists and modeling experts; the firm plans rollout of the product offerings- mapping, individual risk assessments, training and ‘nowcasts’ in greater breadth during 2019 and beyond.

With the advent of availability of these and other risk prediction tools, will the insurance industry leverage the detailed analyses into more widely available, less costly and more desirable flood products?

Some considerations:

  • A generally available flood cover within the US could produce a $40 Bn premium ‘bump’ for carriers, if the risk is aggregated back into standard HO policies. ( Flood Insurance Gap Represents a $40 Billion New Market, Guy Carpenter)
  • Detailed flood risk information can be leveraged into parametric programs, including municipalities establishing immediate financial response programs if a flood trigger/index is met
  • New parametric programs such as that offered by FloodFlash will become more widely available due to the availability of risk data, and changes in market tendencies related to flood risk
  • Reliance on government subsidies may be reduced as less costly alternatives become available within the private flood market.
  • There will be a more affirmative approach to risk education as information becomes more robust.  As John Siegman of Hazard Hub noted, the biggest change in flood risk will be when there’s a recognition that agencies are best suited as trusted advisors re: flood risk for customers, and carriers are able to regularly communicate risk proactively to their customers based on empirical, topical risk data.
  • Will opt-out choices become integrated into policy applications, e.g., risk scoring supports flood cover and prices it, prospective customer chooses to not have it included? Or the carrier opts to not extend HO cover?
  • Flood programs will evolve from subsidized, backward looking programs to forward looking, risk-rated private plans.
  • Private flood insurance plans integrated into HO policies opens the door to a wider spectrum of adjusting/methods of settlement.

There’s plenty to discuss about flood insurance, its magnitude of insurance gap, subsidization of cover, and potential if the private market becomes the primary flood insurance vehicle.

The presence of innovative AI risk analysis, predictive programs, and reasonably inexpensive access to same by carriers and insureds just might prompt a new tune for flood insurance, which surely would be a welcome addition to the insurance dance card.

image source



Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.