← Back to Insights

Lanzko Insights

Practical notes on claims innovation and AI trends—built for claims leaders.

Come on people - we are better than this. Our industry is slow to adopt despite all the money being poured into AI solutions and this is one that should be an easy fix!

The Trillion-Dollar Industry Still Auditing Claims in Spreadsheets

We review a handful of files, type what we find into cells, and call it oversight. That worked when it was the only option. It isn’t the only option anymore.

Come on people - we are better than this. Our industry is slow to adopt despite all the money being poured into AI solutions. The Reality is even the most basic functions today are still using technology from the 1980’s

U.S. property and casualty insurers wrote more than $1.1 trillion in direct premiums in 2025. They incurred about $638 billion in losses and loss adjustment expense, according to the NAIC.

Thanks for reading Lanzko Insights! Subscribe for free to receive new posts and support my work.

Claims still requires judgement and insight and what technology has done is made claim handlers have to handle more information in more files. The job is still essentially the same and someone has to evaluate coverage, investigate the loss and set a timely and accurate reserve. Overseeing this function, whether in-house or outsourced, requires a review. Unfortunately the method for conducting these reviews has not changed and relies mostly on spreadsheets.

How it works now

I’ve done a lot of claims audits for reinsurers, carriers, and TPAs throughout my 25 plus years in this business and I can say unequivocally say that the routine hasn’t changed much in twenty years. In fact, I had a custom

You pull a small sample. It’s “random” or maybe its what I call “targeted random” so a set sample doesn’t exclude key claim types or ones with only limited activity. You open a spreadsheet with 30 to 80 question, contact within 24 hours? Coverage confirmed? Reserve adequate? you read the notes and score it. If you are lucky it rolls into a pivot table automatically, more likely there is time spent combining spreadsheets, ensuring data is correct and then developing analysis. From there someone reads the data and formulates opinions and drafts a report of findings.

I’m not knocking the people who do this work, its noble work. The spreadsheet was the best tool we had. Its also frought with inefficiency, inconsistent and time consuming. A 50 file audit will spend a week or two just getting the audit set up and assessed before a report is even drafted.

What the spreadsheet can’t hold

A spreadsheet stores answers, not evidence. When you mark “Reserve adequate” as No, the workbook doesn’t know which medical report got ignored or how long the file sat low. That lives in the auditor’s head.

Four problems follow.

Nothing connects. Every audit is its own spreadsheet of its own questions. Comparing years, TPAs, or adjusters means rebuilding the data by hand. And thats only true if you have a clean system to store and catalogue reviews. The reality is not nobody does.

You can’t trace a score. When a regulator, reinsurer, or plaintiff’s lawyer asks why a file was scored a certain way, it means pulling the underlying sheets and reviewing what made up the score. Its certainly doable but not the best way to approach a finding.

It looks backward. While it is true that audits are retrospective, the reality is since the review time and compilation add time to know what is going on. There is no real time feedback and by the time the deck is presented, reserves have developed and the leakage has been paid.

Leakage is commonly estimated at 5 to 10 percent of claim payments. Those figures mostly come from vendors, so treat them with caution. But even the low end, applied to $552 billion in net losses, is past $25 billion a year. Most of it sits in files nobody opened.

What AI actually changes

AI is very good at reading information and processing a lot of it very quickly. It is also very good at analyzing data to surface trends and changes. And if built correctly with the right context and governance, can review more files faster and more completely than any other manual method can.

Most of the AI talk in claims is about processing faster. Those gains are real, but they also create another problem. When handling speeds up, decisions pile up faster than a sampled audit can keep up. The oversight gap gets wider.

Specifically, AI can read everything, - that means every note, letter, and medical and legal report across the book. It doesn’t just skim the note but digests them and with context looks to the answer your reviewer does based on the same method they do.

Standards apply the same way every time. Audits, adjusters, and TPAs finally become comparable. Patterns like under-reserved injury types or high-cost defense firms start to show up.

Every finding points to proof. “Reserve inadequate” becomes “IME dated March 4 recommended surgery, reserve unchanged for 97 days.”

Review becomes continuous. It happens while there’s still time to change how the file ends.

The AI review also writes more and because of this not only finds the issue but shows specifically the reasoning and evidence why is chose to score a claim the way it did. In manual reviews the ability to thoughtfully synthesise when there is a positive finding is not always a luxury a reviewer always has. So they focus on the negative findings and justifying what went wrong. This is a product of the manual review not the intention of the reiviewer. AI changes this.

What it doesn’t change

AI doesn’t replace an auditor’s judgment. The risk I see isn’t that carriers will ignore AI., it’s that they’ll adopt it without the proper context set up and the evidence to support the findings with no human oversight signing off. That swaps one blind spot for another.

A defensible process does three things. The AI handles breadth by reading everything and applying your standards consistently. Qualified people handle depth by validating material findings and signing off. And you keep a record of what the AI surfaced, what the reviewer accepted or overrode, and why.

That’s the line between an AI-assisted audit and an AI-generated opinion.

The cost of standing still

If you keep reviewing claims the old way, it will cost you 90% more and you will be able to run limited reviews over time. I have a client that can only afford to review about 30-40 files on 11 TPAs a year. The AI version will amount to 100 files on those same 11 TPAs but twice a year. While their competitors are finding and fixing claim issues annually, the AI forward company is seeing insights 13x faster. Competitors who can show portfolio-wide quality with evidence will win the programs and the capacity. Regulators and reinsurers will expect oversight that matches AI-driven handling, and “we sampled 40 files in Excel” won’t cut it.

I’ve sat on both sides of this. As an auditor, what drove me nuts was notes that were cut and pasted and said nothing. As an executive, I watched excess carriers and reinsurers fail to find what they needed in our files and make it everyone’s problem. The spreadsheet served neither side.

Where I’d start

You don’t need to tear up your audit program.

  1. Look hard at your last audit. What share of the book did it touch? Could you back every score with a document? Can you trend those reviews over time? Is trending on annual basis really trending anything?

  2. Turn your standards into structured, versioned criteria.

  3. Pilot AI-assisted review on one line of business, with evidence links and human sign-off on anything material.

  4. Count what the pilot found that the sample missed.

That last number changes the conversation inside your company.

Claim reviews deserve better than a spreadsheet

I built The Audit Portal to solve this very issue. A system of record for handling any kind of review you might have. Re-use templates against different TPAs and get a consistent cross evaluations that can be tracked. Tired of waiting for results, review more files more often and see results in real time. Send me note and I can show you how.

Sources

Thanks for reading Lanzko Insights! Subscribe for free to receive new posts and support my work.