How Does AI Detect and Prevent Benefits Fraud?
Artificial intelligence is very good at spotting that random face in the crowd
When you get right down to it, almost all benefits fraud we’ve seen over the years relies on two basic strategies: disguise and blending in with the crowd.
Every person who generates a fraudulent claim will do everything they can to make that claim look just like every other claim – to not stand out, or raise any red flags. Second, those who work to defraud the system rely on their bogus claims getting lost among the huge numbers of legitimate claims that are submitted daily. In other words, don’t stand out and you won’t be noticed.
Now it seems like everyone is talking about AI – artificial intelligence – and all the things it can do, might do and is already doing. And one use for AI that I’m hearing and reading a lot about is in detecting and preventing benefits fraud.
I think we need to start by taking a step back in time to understand what AI can do for us in seeing through the disguises and picking out the fraud that’s trying to hide in the crowd.
There was a time, back in the day, when a benefits claim was an actual piece of paper, one copy of which would land, a few day later, in the “in” tray of the person adjudicating claims.
The adjudicator would then quickly review the claim – yes, that person is a member of the plan, yes that procedure is covered, yes all the right spaces on the form are properly filled out. He or she would then sign or stamp the claim and put it into an “out” tray for processing and payment.
Of course, that adjudicator would see many different claim forms. Dozens in a week, hundreds and hundreds over the course of a year. And he or she would get very familiar with every possible type of claim, every exception, every error, every ineligible claim and every variation or combination of factors on a claim.
Now here’s the interesting part. Once in a while, that adjudicator would pick up a claim form, scan it quickly and then stop for a second. Stare at the form. Maybe lean back.
And then say something like, “That’s funny…”
It looks like a perfectly ordinary claim. Properly filled out, signed, dated. And yet the adjudicator is saying, “Hold on a second here…”
Maybe it’s a claim for something he or she normally sees a couple of times a year, and now here’s the sixth one in the last month.
Maybe it’s that, normally, about half the people who go in for a certain procedure also get a second, unrelated service, and now suddenly everyone who goes to this one clinic for the first procedure is getting two or three of the second.
Maybe it’s a combination of procedures or services that don’t usually go together. Maybe it’s a spike in one type of claim, or sudden increase in claims from one department, or from one service provider.
It could be any of a very large number of slightly out-of-the-ordinary little things that seem perfectly normal by themselves, but make an experienced professional stop and take a second look when his or her finely tuned instincts raise a small red flag.
The next step was a trip to the filing cabinet, to pull every claim for the last three months. Or longer. And then to sort through them and make one pile over here, and another pile over there until the picture – if there is one – becomes clear.
That was, as I said, back in the day. Way back.
Now everything is automated and online. And now we have AI doing the same basic thing. But there are some big differences.
– When AI is monitoring benefits claims, it does not rely on a person to spot “something funny.” It scans all claims, all the time, routinely. It doesn’t rely on a person having the time to look at and think about every claim.
– AI does not rely on one person’s experience. It pools the knowledge and history of past claims – and past frauds – from across the industry for many years.
– AI can learn as it goes along. It identifies patterns – known patterns that are inherently suspicious, and many other patterns that are merely unusual. When a human operator, scanning the reports from the AI program flags some of the new, unusual patterns for further analysis, the AI sifts through them and adds the concerning patterns to its list of known patterns.
– AI works on a far larger scale than the claims adjudicator we discussed above. It can scan thousands and thousands of claims from different plan sponsors and pull up patterns that an individual employer or benefits provider would never flag, since there may be only one or two instances in any given workplace.
– In fact, AI is now looking at claims across the entire industry. Currently, the Canadian Life and Health Insurance Association (CLHIA) is coordinating efforts, through data pooling, to give individual plan providers access to data on patterns from across the industry. (Pensions and Benefits Monitor)
Of course, spotting a pattern of unusual activity does not mean much by itself. If a larger than normal number of plan members all start going to the same clinic for treatment, for example, it may well just mean that that clinic gives very good service, and the word is getting around.
That’s why the program will flag patterns for human operators to investigate further. I should also mention that the kind of data sharing the industry is talking about does not include any personal data on plan members – they are very careful to protect privacy.
Finally, it should also be noted that the kind of data on patterns of benefits claims can be very useful in areas that do not necessarily rise to the level of fraud. Someone may merely be stretching or attempting to stretch the rules. Or applying an iffy definition to the rules. And this can be addressed by clarifying the rules and updating the definitions to shut that particular door.
Benefits fraud has been with us as long as there have been benefits, but AI is giving the industry one more tool to prevent fraud – and, in doing so, protect the plan sponsors and plan members we all ultimately work for.
***
I really appreciate comments, ideas, suggestions or just observations about the blog or any other topics in benefits management. I always look forward to hearing from readers. If there’s anything you want to share, please email me at bill@penmorebenefits.com.
Copyright Notice
© Penmore Benefits 2023. All rights reserved. All of the content herein is the sole property of the Penmore Benefits, and may not be reproduced, transmitted, or stored in a retrieval system – in whole or in part – without the written permission of the Penmore Benefits. Links to the originating article are permitted.
The Buzz Bits
60% of employees say they’d leave current job for one with more inclusive benefits
Checking up on the dental plan
Are employees staying with their employers or moving on?
Just 13% of employees exercising to improve mental health
EAPs increasing virtual delivery