AI provides the edge that insurers need in general liability

General liability insurers need means at their disposal to minimize the impact of claims when they occur.

Artificial intelligence can help insurance companies minimize the cost of general liability injury claims by streamlining various processes and providing valuable insights to support decision-making. (Credit: Who is Danny/Adobe Stock)

General liability claims in commercial insurance can arise from accidents or incidents during business operations. These claims are typically filed against businesses for injuries or damages sustained by third parties, such as customers, clients or other individuals.

In this article, we will focus on liability claims resulting from injuries suffered by these third parties and how insurers can utilize artificial intelligence to help minimize the amount paid per claim while treating the injured party fairly.

Injury claim types

Many incidents can lead to bodily injury claims. Some common types of incidents that generate general liability claims include:

Many of these hazards and the injuries they cause seem preventable, but eliminating them can be cost-prohibitive. General liability insurers and companies that self-insure need means at their disposal to minimize the impact of claims when they occur. In today’s world, increasingly, the answer to this problem is artificial intelligence (AI).

AI Is the edge insurance carriers, captives need

Artificial intelligence can help insurance companies minimize the cost of general liability injury claims by streamlining various processes and providing valuable insights to support decision-making. Some ways AI can be utilized in this context include:

  1. Predictive claim analytics: AI can analyze historical claims data to identify trends and patterns suggesting an increased likelihood of slip-and-fall accidents in certain locations or under specific conditions. This information can help insurers proactively address potential hazards, reducing the frequency and severity of claims.
  2. Claims processing automation: AI can streamline the claims process by automating tasks such as document review, data extraction, and initial claim assessment. This can help insurers reduce administrative costs, shorten the claims resolution time, and provide a better experience for policyholders.
  3. Fraud detection: AI algorithms can analyze large data sets and identify patterns or anomalies that may indicate fraudulent claims. Insurance companies can reduce costs associated with paying out undeserved settlements by detecting and preventing fraudulent slip-and-fall injury claims.
  4. Litigation strategy and support: AI can assist insurers in identifying claims with a higher likelihood of litigation and help them develop effective defense strategies by analyzing historical case data, legal outcomes, and trends.
  5. Loss prevention recommendations: AI can help insurers develop targeted loss prevention strategies by identifying risk factors associated with slip-and-fall accidents, such as poor lighting, slippery surfaces, or cluttered walkways. Insurers can then provide policyholders with actionable recommendations to mitigate these risks and reduce the likelihood of future claims.

Reducing claim frequency and severity

Several artificial intelligence techniques can assist carriers, self-insureds and TPAs in reducing the frequency of claims. AI can identify patterns of incidents that lead to injury claims, enabling insureds to take proactive preventative actions to reduce injuries. It can monitor data in near real time to detect if these patterns change because of corrective action. And AI can highlight emerging patterns of new sources of injuries that may prove troublesome going forward. Among the wide range of AI techniques available, these are well-suited to these tasks:

  1. Machine learning (ML): ML algorithms learn from data, enabling them to identify patterns, make predictions, and improve their performance over time. There are several types of machine learning techniques, such as supervised learning, unsupervised learning, and reinforcement learning, which can be used to identify patterns in data depending on the problem at hand.
  2. Natural language processing (NLP): NLP focuses on enabling computers to understand, interpret and generate human language. NLP techniques can identify patterns in textual data, such as sentiment analysis, topic modeling, or named entity recognition.
  3. Anomaly detection: This technique is used to identify unusual or unexpected patterns in data that deviate from the norm. Anomaly detection algorithms can be statistical, machine learning-based or deep learning-based, depending on the specific application and data set.
  4. Clustering: Clustering is an unsupervised learning technique that groups data points based on similarity, allowing for pattern identification within the data. Standard clustering algorithms include K-means, DBSCAN, and hierarchical clustering.
  5. Time series analysis: Time series analysis involves the study of data points collected over time to identify trends, seasonality and other patterns. Techniques like autoregressive integrated moving averages (ARIMA), exponential smoothing, and deep learning methods like LSTMs can be employed to model and forecast time series data.
  6. Generative AI: An emerging technology under the umbrella of AI, GAI applies what it learns from creating a language model of the subject at hand, in this case casualty claims, and guides adjusters to take specific actions, supported by explanatory text statements. One way to think of GAI is that it integrates insights from the other AI techniques mentioned and communicates them to adjusters in an easy-to-understand manner.

Applying AI to the problem

When it comes to reducing the severity of claims, many of these same techniques can be used to build models that triage claims at the first notice of loss (FNOL), predict which claims will be litigated, and discover fresh insights from relationships in unstructured data contained in medical bills, doctors’ reports, and the plethora of other documents associated with injury claims.

Triaging a claim is essential in predicting how complex and severe a claim will likely become over time. Today, triage models built from a carrier’s data can assist in assigning the claim to the appropriate adjuster. By working with an AI software as a service (SaaS) provider, a carrier can combine their data with anonymized data from other carriers and captives to make their models more robust. The best of these models work not only with FNOL data but also update their scores and alerts every day that new information appears on the claim file.

Similarly, models focused on predicting the probability of litigation by the injured party work with FNOL and all subsequent data. Once an attorney represents a claim, the best litigation models score plaintiff attorneys on the outcomes they have achieved on similar claims. They can even score defense counsel, in-house or out-of-house, to find which attorneys perform best from a defense perspective against the attorney a claimant has chosen to represent them.

Another type of model that leverages natural language processing of scanned and digital documents to uncover valuable items and relationships among them in unstructured data is now emerging. The best among these are revolutionary and work as a “second set of eyes” on the myriad paper and digital forms and other documents that healthcare providers, pharmacies, hospitals and clinics generate. Like triage and litigation models, the best performing models of this type act in the background and process new information when it becomes part of the claim file.

What insurers need to compete

While the benefits of building the capabilities described above are significant, building their capabilities is no easy task. Here are the required elements of a successful AI strategy:

  1. Data: Artificial intelligence strategies start with data related to the problems being solved. An abundance of claims data is generally not enough. For data to be valuable, it must be high quality: error-free, well-defined and heavily populated (i.e., no blank or garbage fields). Having claim systems with lots of historical data is not enough. The data must be extractable into a repository where data scientists can access and work with it.
  2. Cross-functional engagement: This is a critical but often overlooked element of a successful strategy. Data science alone cannot solve the totality of building, deploying and using artificial intelligence to manage claim outcomes. At a minimum, claims and technology must be equal partners in the process. It is highly recommended that the front-line people who will deploy and use the tools be involved from the beginning. This helps ensure that problems are correctly defined, solutions are feasible, and adoption occurs.
  3. Senior leadership support: AI projects of the scope and scale described here are hard to keep “stealth.” AI is also a topic much senior management is interested in already; at least, they’ve heard a lot of the hype surrounding it. Though senior support isn’t as critical as the other components described here, I believe having a couple of executives supporting and defending the initiative is much better.
  4. Adaptable technology: Legacy systems are a fact of life in insurance. And they were not designed with AI in mind. That said, creative technologists can find ways around this. Many are approaching AI with a SaaS frame of mind. As long as data can flow to AI models and their output can flow back into a decision system, all should be well. It is essential not to make assumptions in this area, though. See step 2: Cross-functional engagement.
  5. Change management: Change happens, but it is most effective when well managed. For many insurance companies, AI is something new. And a lot of insurance company cultures abhor change. Making change management a crucial part of your strategy from the get-go will help overcome the barriers the culture will erect, sometimes without even knowing it.
  6. Measurement: Finally, a multidimensional measurement process needs to be established. It’s hard to prove causality when a model is implemented and results (hopefully) improve. But that doesn’t mean you shouldn’t try. Having predefined metrics that should improve with the introduction of a model is a first start. A few months, at a minimum, of measurement is needed before all constituents can begin weighing the evidence on whether the models work or not. And don’t forget about qualitative input. Start collecting this from the time the model becomes usable. It can help in a “weight of the evidence” evaluation of the strategy. Still, it can also help identify parts of the process implementation that need to be altered to improve utilization and proper use of the model and its output.

While AI processes can be built in-house, it is often best to get started with outside help. If a vendor is experienced and has many successful implementations, this can be a more cost-effective approach to your AI strategy. As mentioned before, there are some software as a service providers of insurance AI that have established and successful track records. Some vendors can even get you up and running with AI within three months. So, choose wisely!

Looking to the future

General liability coverage is fraught with all sorts of claims that can be very costly for insurance companies to settle. Artificial intelligence, and the automation it enables, is a weapon carriers and self-insureds can use to combat lengthy, costly claims. To leverage artificial intelligence, several components of a succinct strategy need to be in place. Carriers can try to develop these capabilities in-house or can rely on third-party providers who can often get them up and running faster.

Tom Warden is a research fellow at CLARA Analytics, is an experienced leader in using data, analytics and AI to solve complex business problems. Any opinions expressed here are the author’s own.

See also:

Recommended Stories