What Insurers Lose While Fraud Waits in the Queue 

SIU backlog is a loss-ratio problem. Every day a suspect claim waits in the queue raises the chance it gets paid before fraud is confirmed. Insurers reduce it by ranking referrals by fraud risk, automating case preparation and moving fraud-ring detection before payment. That protects the loss ratio and gets legitimate claims paid faster. 

We sat down with Abhishek Dwivedi to talk about how insurers are handling the growing queue of suspect claims, and why he thinks most of them are looking at the problem from the wrong angle. Most carriers treat SIU backlog as a staffing issue inside the investigations team. Abhishek sees it differently. We asked him why backlog belongs on the board agenda, who should own it, and what leaders can do about it now. 

Abhishek leads Xebia’s global insurance business across P&C, life, group insurance and reinsurance. He works with carriers on digital transformation, P&L performance and AI-led change, including the claims and fraud problems that move the loss ratio. 

The cost of the queue

Abhishek, as we built Xebia Agentic Fraud Intelligence, we kept hearing backlog described as an SIU problem. But the cost seems to show up all over the business. What does SIU backlog actually cost an insurer?

You’re right that it spreads. I’d put it in three places: claims leakage, loss adjustment expense and customer retention. Most carriers only track the first one, and they usually track it late.

Leakage is the obvious one. The Coalition Against Insurance Fraud estimates property and casualty fraud at about $45 billion a year, and the industry generally assumes around 10% of P&C losses involve fraud. So every week a suspect claim sits in the queue, there’s a real chance it gets paid before anyone confirms what it is. When we model that against industry averages, it comes out to more than $1.5 million in fraud that could be stopped before payment for every 1,000 claims.

The expense side is quieter. Investigators are some of the most expensive people in claims. One claims leader put it to us pretty bluntly: “A lot of what reaches my team isn’t fraud.” So you’re paying specialist rates to sort ordinary claims.

And retention never shows up in the SIU budget at all. An honest customer stuck behind a fraud check just sees a slow insurer, and they remember that at renewal.

That $1.5 million number is going to get attention. Would a CFO believe it?

Not on its own, and honestly, I wouldn’t want them to. It’s a model built on industry averages. My advice to any CFO is to rerun it with your own claim volumes and average severity before you take it to the board. That’s when it becomes your number.

If so much of the cost lands outside SIU, it sounds like this is bigger than one team. Why is SIU backlog a board-level issue, and who should own it?

That’s exactly why. Most of the cost lands in the loss ratio, the expense ratio, renewals and regulatory exposure. SIU sees maybe one of those. So backlog ends up on an SIU slide and almost never on a board slide.

I’d give ownership to the Chief Claims Officer, with the CFO holding the scorecard. The CCO runs the process where the queue forms, from intake all the way to payment. The CFO can put a monthly number on it, and boards respond to numbers. The COO has a role too, because the fix touches operations and systems. But if you split ownership three ways, you’ll be having this same conversation next year.

When fraud waits, everyone waits 

One thing that surprised us early on was how often fraud is confirmed only after the money’s gone. What happens to fraud losses when detection arrives after payment? 

It turns into pay-and-chase. You’re trying to recover money from people who planned for you to chase them. In my experience, recovery is slow and usually partial, and it eats legal and investigative time that would have cost far less before the payment went out. 

A claims leader summed it up for us really well: “By the time we’re confident, we’ve already paid.” The data was there. It just showed up after the payment run. 

Fraud rings are the worst case. Each claim in a ring looks clean on its own, so none of them gets referred. And at most carriers, link analysis only runs once someone already suspects a claim. By the time a ring surfaces, dozens of claims may already be paid, sometimes across several insurers. 

Have you actually seen that change in practice? 

I have. At a mid-tier US P&C carrier, a large property claim was held before payment because its bank account and phone number matched other suspicious customers. Under the old process, nothing about that claim on its own would have triggered a referral. The link would have come out after the money was gone. 

There’s another side to this we heard a lot about: the honest customers caught in the same queue. How does SIU backlog hold up legitimate claims, and what does that do to retention? 

When you can’t tell suspect claims from honest ones, everybody waits behind the same checks. 

In straightforward lines, customers can switch insurers in an afternoon. A claim is often the first time they find out what they actually bought. A delay with no explanation is what they’ll remember at renewal. 

So you need a positive reason to pay, not just the absence of a red flag. A US life and accident insurer we worked with released a genuine claim the same day, because the checks found no risk indicators and no shared bank account, phone, device or address with any other customer. The score, the reasoning and the approver went into the audit log right as the claim was released. 

Fixing it

So let’s get practical. How should insurers reduce SIU backlog? 

I’d focus on three things: rank every referral by fraud risk, automate case preparation, and run fraud-ring detection on every claim before payment, not after a referral. 

Ranking matters most, in my view. One carrier told us, “We bought the tools. We still don’t get one answer.” Four tools, four opinions, and an adjuster stuck in the middle. When you bring identity, documents, images, behavior and network links into one score, with the reasons ranked, the claims that matter reach investigators first. 

Automation is really about where you spend your expensive people. Pulling a case file together should happen the moment a claim is referred, without an investigator doing it by hand. 

And the ring check has to move earlier. Run it on every claim, before payment, and organized fraud shows up without anyone needing to suspect it first. 

A lot of leaders’ first instinct is headcount. Why not just hire more investigators? 

Because you’d be adding people to a queue that isn’t sorted. They’d just work through more claims that were never fraud. 

When we talked to carriers about this, the next question was almost always about their core systems. Can carriers fix SIU backlog without replacing them? 

Yes, and that’s often what decides whether the project gets funded at all. Xebia Agentic Fraud Intelligence reads the policy admin, claims and payment systems a carrier already runs, through API and data connectors. Intake looks the same as it does today, and it isn’t tied to one cloud or one AI model. Core modernization takes years, and fraud losses aren’t going to wait for it. 

Accountability and action 

Regulation came up in nearly every conversation we had while building this. How does the NAIC AI bulletin change accountability for AI-assisted fraud decisions? 

The NAIC Model Bulletin on the Use of AI Systems by Insurers makes the carrier accountable for AI-assisted decisions that affect consumers, including claims decisions, even when a vendor built the model. As of October 8, 2026, 29 jurisdictions, including Washington, D.C. and Puerto Rico, had adopted it. 

In practice, that means an examiner can ask why a claim was held, or why another one was paid in a day, and who signed it off. “The vendor’s model said so” won’t get you very far. 

So you want a reason code, a human approver and a timestamp on every decision, recorded when the decision is made. Rebuilding that record months later for an audit is painful, and it’s weaker evidence. And people stay in the loop the whole way: the adjuster still decides whether to pay, ask for more evidence or refer, and SIU and compliance still review what gets referred. 

Honestly, I think the carriers who get this right will end up moving faster. If you can explain every decision, you can afford to automate more of them. 

Last question. If a CEO or CFO wants to start on this right away, what should they ask their claims team on Monday? 

I’d start with five questions, and none of them needs a new system to answer: 

If the team can’t answer the first three within a week, the backlog is still being managed as a headcount problem. 

Want to see what your SIU queue is costing you, and how much of it could be stopped before payment?  

Talk to Abhishek or book a demo of Xebia Agentic Fraud Intelligence. 

Sources