Claims
Analytical Impact

• John Czuba
• September 2018
• 

New tools and greater availability of third-party data are changing how insurers evaluate claims, screen for fraud and settle cases faster, increasingly ahead of schedule.

During the A.M. Best webinar, How Smart Data Is Remaking Insurance Claims, a panel of claims professionals examined what tools are being used, what types and sources of data are making the greatest impact, and the issues insurers face as they rely more heavily on data-based claims automation.

New technology is helping insurers to react to claims faster to reduce fraud.

“I think the No. 1 thing that carriers are looking for is quick referrals. They ideally would like to be in real time. The way that the majority of the fraud framework platforms work is it's an instantaneous alert to the desktop of the handling adjuster, which then triggers a referral to SIU [special investigative unit]. The insurance industry is looking to get on top of that before it goes out the door. The quicker they can jump on that, the better. Timing in these types of investigations is everything,” said Jeffrey Rapattoni, the shareholder and co-chair of the Fraud and Special Investigation Practice Group with the law firm of Marshall Dennehey Warner Coleman and Goggin.

Following is an edited excerpt of the webinar's transcript, which also included Philip Beneventano, CPA and, also, senior accountant with the company RGL Forensics. Pamela Woodside, a partner with the law firm of Thompson, Coe, Cousins, and Irons and Kelly Lippincott, a member with the Carr Maloney PC law firm.

A.M. Best's Senior Associate Editor John Weber and John Czuba, managing editor of Best Insurance Professionals and Claims Resources served as moderators of the panel.

How has the claims environment changed subsequent to analytical implementation?

Rapattoni: What you're seeing is the modern claims environment becoming self-aware. They're more aware of what's in their shop.

As a result, they're able to become much more efficient than they were in the past. They're able to siphon through that data and find out what claim can be fast-tracked, what claim is going to need special counsel. Is it a cat loss? Is it low impact?

They're going to be able to identify fraud, and they're going to be able to move on that fraud much quicker than they did in the past. As a result, they're going to be able to get SIU involved and get those claims the specific attention that they need right away.

It's about the claims environment becoming more self-aware of what's going on in its own house, as opposed to being aware three days, four days, five days later. The data integration in the claims environment is immediate, instantaneous. It's 365, never stops, never shuts, always updating in the background, always working.

What types of wearable technology are being employed to manage workers' comp claims?

Woodside: There are different types of technology. You see technology, what I would call the hard technology, out in the construction field. Soft technology you might see in an office environment.

We're talking more about the hard technology. There's things that have been referred to. There's a very interesting helmet. It's called the “Daqri helmet.” It actually measures things, like distance to equipment that will tell a worker if they're in a danger zone. If there is something approaching, it'll set off an alarm.

If there's a risk of impact, there are various vests that are being tested now that have sensors in them that will tell a worker if they're bending or twisting in an improper manner.

When you're talking about workers' compensation claims, this is somewhat critical because you're wanting to reduce the possibility of repetitive injury from different bends and from different twists.

There is a thing that's called a proactive helmet. A lot of times, workers in the field don't want to wear like a hard, hard hat. It looks like a baseball cap. It has more of a flexible lining inside that's soft, but if there's an impact, it becomes harder to protect for impact.

What are the risks to insurers in using third-party data to evaluate claims?

Lippincott: The risk is for anybody who's retaining a significant amount of data. Personal information, security issues, privacy issues are all the risks that are facing the insurer. There are things lying in wait for the insurers that mishandle this information.

For instance, there's a watch group called Electronic Privacy Information Center known as EPIC, which I'll discuss in more detail later. They are out there looking and watching for how people are using this data, and making sure that it is being used appropriately and being protected appropriately.

There's also regulations that are ever-evolving. This can be a real risk to insurers that are using data as part of their claims-handling process.

How do the high-performing insurance carriers use data analytics to fight fraud?

Rapattoni: I think the No. 1 thing that carriers are looking for is quick referrals. They ideally would like to be in real time. The way that the majority of the fraud framework platforms work is it's an instantaneous alert to the desktop of the handling adjuster, which then triggers a referral to SIU.

The insurance industry is looking to get on top of that before it goes out the door. The quicker they can jump on that, the better. Timing in these types of investigations is everything.

Sometimes people can't be found, they stage losses, they leave the country, so the more proactive you can be, the better you're going to be able to save the company money upfront as opposed to coming out the door.

It's really all about getting your time back. Do we need a higher vendor in another state, another jurisdiction? Do we need another lawyer in another state or jurisdiction to take an examination under oath?

All of these decisions, if they can be made quicker, they're ultimately going to save the carrier big dollars in the end. It's really all about the analytics framework getting the insurance company quicker and more efficient in the SIU realm.

How much of an impact does analytics have on litigation?

Rapattoni: Huge. It's amazing. When I first started out, I really didn't see a lot of the carriers coming to us with data. Particularly in my practice with building major case or recovery actions or RICO actions, the data that the carrier can put in front of a lawyer is staggering.

You talk about a medical provider recovery action, and they're able to create huge data runs of CPT [current procedural terminology] codes collated by dates they would perform. How long those particular modalities took. The minuscule detail that they're able to put in the hands of the lawyers now is fantastic.

The byproduct of that is that, as a lawyer, we're able to do our job much more efficiently. I'm able to craft an affirmative litigation pleading so much more specific, so much more specific, to the judiciary, so the judge understands exactly what I'm going for and that I'm not on a witch hunt.

What are some of the technological tools being used to combat workers' comp fraud?

Woodside: At least one is social media. Everybody likes to post what they're doing, where they are. I think that's been a real big tool for combating workers' comp fraud, the ability to see what a person is doing.

If someone says that they are completely unable to do something, with the ability to go and monitor their social media accounts, you can see if they are actually incapacitated or if they have been skiing that weekend.

It's been just a really good thing for combating workers' comp fraud. Similarly, I think a lot of people are wearing wearable devices.

It's a similar thing. People may not think about it, but a lot of people will automatically post their data or their data is automatically uploaded to their wearable device or platform.

People don't really read the privacy rights or exactly who can get access to that data, but it's actually fairly broad-based who can get access to that data. You can actually get a fair amount of information as to what a person's activities are and that will tell you also what a person has been doing.

Whether they have been sedentary or whether they went on a three-mile run or hike, etc. That has also been something that has combated workers' comp fraud.

There are a lot more companies that are using surveillance video to be able to see whether or not an employee was actually injured on the job, when they were injured, if the injury was as severe as the employee's saying.

A lot of times an employee may say, 'I slipped.' Then when you look at the video, it's the severity of the injury or the way they describe the injury, it didn't occur the way the employee describes the way the injury occurring.

Or, they are exaggerating the severity of the incidents of the injury. I think those are the main ways. Also, one of the ways that's new is predictive analysis.

Companies are using predictive analysis to be able to aggregate data in various files to possibly flag potential risk files or fraudulent files and to highlight it for special investigation as maybe a potential fraud.

If it's a person who either gets injured on a Friday, late on a Friday or early on a Monday, or has a lot of workers' comp claims filed and/or has a history of certain things in their past, that may get flagged, but the information may be located across different files.

Now with predictive analysis and analytics, you can aggregate all that information and really bring it together. I think you have that ability now that you didn't have right before.

What areas of advancement in workers' compensation technology have been most valuable as far as you're concerned?

Woodside: Social media has been by far the most invaluable.

I think people today just constantly, always, always posting their whereabouts, what they're doing, where they are, who they're with. I think that's one of the main things that really has been one of the main advancements that has really helped.

Again also, probably the wearable technology. A lot more companies are having employee wellness programs. As part of their employee wellness programs they are giving people wearable tech devices.

The hard versus soft prevention of workers' comp injuries, wearable technology is one of those things that can prevent soft workers' comp injuries.

Wearable technology—given the amount of data that you can see that's going on in a person's body, heart rate, how they sleep, things of that nature—they have been found to be predictors of potential stress later on in life.

That has been one of the things that has been a great advent in preventing workers' comp injuries.

It's been a little bit questionable, whether that has been something you really want people to know about, but it has been definitely a boon for preventing workers' comp injuries, to the extent that, based on some wearable technology, that someone may be developing a stress-related injury or could develop a stress-related injury, and therefore, you may need to take steps to get them out of a situation.

When would Benford's law be used in fraud analysis?

Beneventano: Benford's law may be useful to see if a data set has been fraudulently aggregated or added to. Now, most people would think, in a data set, that there's an equal probability of a number starting with a two, a five, a seven. But in a data set spanning many orders of magnitude, it's highly likely that the number will start with a one or two, compared to an eight or a nine.

If you're looking at money that's growing or compounding at 20% in a bank account, start with a 100, the next year it becomes 120, the year after that 144, 173. Right out of the gate, you see four instances starting with a one. Then, it becomes 207, 248, 299, so three instances starting with a 2, and then it bumps up, 358, 429, 516, all the way up.

The reason is, when you're taking $100 and trying to grow that into $200, you're really doubling that number, whereas if you're going from, maybe $700 to $800, a lot smaller percent increase, that could be done more quickly.

This is useful. We've seen this happen with transactions where they have, let's say, approval of $10,000 for a limit that requires a signature, a lot of times you'll see something where people will try to put in $7,000 or $8,000, try to get below that limit.

If that's the case, you would see a lot of instances starting with seven, eight, nine, and that really wouldn't fit Benford's law, and that would be a cause for alarm. Now, Benford's law is not always useful if we're looking at human height, human weight, so then that doesn't really span a good number of orders of magnitude.

If we're looking at sales transactions, expense transactions, things like household net worth, those would vary significantly over many orders of magnitude, and we would expect Benford's law to apply. Again, mostly numbers starting with one, two, and three, not as many starting with seven, eight, or nine.

What techniques can be used to identify the beginning of fraudulent activity?

Beneventano: Bayesian techniques allow for conditional probabilities. As additional knowledge is obtained, the model is updated. That really takes into account both type I and type II errors. We're looking at type I error as a false positive and type II error as a false negative. Obviously, both of those want to be avoided.

If I could use credit card fraud as an example, you wouldn't want to be constantly notified about transactions that were valid, and have them saying that it could be fraud, and you definitely don't want anything where fraud goes undetected and hits your account.

If you have enough months of buying habits, I think, for a normal person that uses a card consistently, several months would probably fit the bill. That's going to look at things like the timing of purchases, so if you're typically using it a few times a week, and all of a sudden, there's many instances for a given week or for a day, it's going to flag that.

The size. If you're making small purchases, groceries, things like that, that's $100, $150, and then you see something for $1,000. Based on the buying habits, that will be flagged. Again, the frequency. If you're not using it multiple times a day, and all of sudden you see it used five or six times a day, that would be flagged.

The timing, as well, so if you're buying things 30 minutes apart at the most frequent, and then there's a couple of purchases within two, three minutes of one another, that would get flagged.

Then, looking at the same location, so if you're at a mall, it might make sense that you might be using it in a couple of different stores, but typically, if you see it used at the same store four or five times in a day, that would get flagged.

Another example would be a consumer's buying habits after their payday. They may be more likely to spend immediately after the payday, and the Bayesian techniques would account for that, too.

Really, it's about starting with a prior probability, but updating that based on the individual or the company in question. As more data is aggregated, the model is updated, and it makes it easier to spot frauds without false positives or false negatives.

Is there a chance that AI could become the preeminent source of claims settlement?

Rapattoni: I really don't think you can get rid of human touch. I think you can enhance the human touch, but I don't think you can get rid of the human touch, or at least not with the technology that we're aware of today.

There has to be some semblance of being able to do a human function, a human element of looking at these claims. Sometimes it's a judgment call. Honestly, I'm not sure AI is advanced enough to make a judgment call. You get testimony, it's a live testimony, you're sitting down, you're interviewing somebody, it's human intuition on whether or not you believe it.

Can AI replace that? I think the answer to that is, no, although I'm certainly no expert on technology, but I would be willing to suggest that getting rid of that human touch is probably long after I've retired, certainly.

Again, it's fascinating. If you had asked me that question 10 years ago, I couldn't have even possibly imagined it. Now we know it's the beginning, it's a reality, and we're just starting off on that journey. To see where that is in 10 years is fascinating.