Industry

Your Insurance Claims Team Is Burning $40 Per Claim Manually. An AI Computer Use Agent Fixes That Today.

Sophia Martinez||7 min
Ctrl+P

An insurance claims adjuster spends roughly 40% of their day copying data between systems, chasing PDFs, and re-entering information that already exists somewhere else in the company. Not analyzing claims. Not talking to customers. Not making judgment calls. Copying and pasting. In 2025. And the industry is somehow surprised that auto repair claims now take an average of 23.1 days to close, more than double pre-pandemic processing times. The problem isn't the people. The problem is that insurers keep trying to solve a software problem with more humans, and a human problem with broken software. There's a third option most of them haven't seriously tried yet: a real AI computer use agent that actually operates the tools your team already uses.

The Numbers Are Genuinely Embarrassing

Let's just put the stats on the table and sit with them for a second. Manual claims processing costs between $40 and $60 per claim. Automated processing drops that below $20. That's not a marginal improvement. That's a 50-60% cost reduction sitting right there, available right now, and the majority of the industry is leaving it on the table. Meanwhile, 74% of insurers are still running on legacy technology, according to data published by Earnix in early 2025. Not old technology they're actively replacing. Legacy technology they've identified as a major challenge and still haven't fixed. McKinsey's 2025 insurance AI report noted a single insurer's AI-related losses exceeding £60 million in one year from operational inefficiency. The industry collectively processes billions of claims annually. The waste isn't a rounding error. It's the main event.

RPA Was Supposed to Fix This. It Didn't.

About ten years ago, every major insurer got sold on Robotic Process Automation. UiPath, Automation Anywhere, Blue Prism. The pitch was simple: build bots that follow rigid rules, automate repetitive tasks, save money. And it worked, sort of, for the simplest possible workflows. But insurance claims are not simple. They involve PDFs with inconsistent formatting, web portals that change their layouts, scanned documents with handwriting, legacy systems that don't have APIs, and exceptions that show up constantly. RPA bots break the moment anything changes. They can't read a messy FNOL document. They can't navigate a vendor portal they haven't been explicitly programmed for. They require armies of developers to maintain. The dirty secret of enterprise RPA is that many companies now employ more people maintaining their bots than the bots actually save. That's not automation. That's just a different kind of manual work with extra steps and a bigger licensing fee.

UnitedHealth is currently facing a class-action lawsuit that advanced in February 2025, alleging its AI system denied Medicare Advantage claims in seconds without proper medical review. CVS denied 22% of claims. Cigna denied 21%. The backlash is real, the lawsuits are real, and it proves one thing: using AI to blindly rubber-stamp denials is not automation done right. It's automation done recklessly. The goal was never to automate the outcome. It was to automate the process.

What 'Computer Use' Actually Means for Claims Teams

Here's where the conversation gets interesting. The new generation of AI isn't calling APIs or following pre-programmed decision trees. A proper computer use agent sees your screen, reads what's on it, and operates your actual software the same way a human would. It can open your claims management system, read an incoming FNOL email, extract the relevant fields, cross-reference a policy document, navigate to the right screen in your legacy platform, and populate the data. Without an API. Without a custom integration. Without a six-month IT project. This matters enormously for insurance because the industry runs on a patchwork of systems that were never designed to talk to each other. Guidewire, Majesco, Duck Creek, homegrown platforms from 2003, third-party vendor portals that haven't been updated since the Obama administration. A computer use agent doesn't care. It works with whatever is on the screen. That's the unlock that RPA never delivered.

The Legacy System Excuse Doesn't Work Anymore

The most common pushback from insurance ops leaders is some version of: 'We'd love to automate but our systems are too old and too complex.' This used to be a fair point. Integration-based automation genuinely does require modern APIs, clean data structures, and IT bandwidth that most insurers don't have. But computer use AI doesn't need any of that. It interacts with your software through the user interface, the same way your employees do. If a human can log into it and click around, an AI computer use agent can too. The 74% of insurers still on legacy systems aren't facing a technical barrier to automation anymore. They're facing a mental model barrier. They're still thinking about automation as something that requires replacing infrastructure. It doesn't. Not anymore. The adjuster's desktop is the integration layer. Full stop.

  • No API required: computer use agents operate through the UI just like a human employee would
  • Works across Guidewire, Duck Creek, Majesco, and any homegrown platform with a screen
  • Handles unstructured inputs: scanned docs, handwritten notes, inconsistent PDFs
  • Deploys in days, not quarters, no IT integration project needed
  • Processes claims in parallel across multiple virtual desktops simultaneously

Why Coasty Is the One Actually Worth Talking About

I've looked at a lot of tools in this space and most of them are impressive in demos and mediocre in production. The benchmark that actually matters for computer use AI is OSWorld, the standard test for how well an agent handles real-world computer tasks across browsers, desktops, and terminals. Coasty scores 82% on OSWorld. That's not a marketing number. That's the highest score in the category, higher than Anthropic's Computer Use, higher than OpenAI's Operator, higher than anything else currently available. For insurance specifically, that gap matters. Claims workflows involve multi-step tasks across multiple systems. An agent that scores 60% on OSWorld is failing 40% of the time. In a claims context, that means errors, rework, and compliance risk. Coasty's computer use agent runs on real desktops and cloud VMs, supports agent swarms for parallel execution across dozens of simultaneous claims, and doesn't require you to rip out your existing systems. It has a free tier if you want to test it without a procurement process, and it supports BYOK if your security team has opinions about API keys. The insurance teams I've seen use it aren't replacing adjusters. They're letting adjusters do the actual work, the judgment calls, the customer conversations, the complex exceptions, while the computer use agent handles the 40% of the day that was just copying data between screens.

The insurance industry has a choice to make right now. It can keep spending $40-60 per claim on manual processing, keep watching auto claims drag past 23 days, keep maintaining RPA bots that break every time a vendor updates their portal, and keep calling legacy systems an unsolvable problem. Or it can accept that computer use AI has genuinely changed what's possible, without ripping out existing infrastructure, without a two-year IT project, and without the ethical disaster of using AI to auto-deny claims at scale. The UnitedHealth lawsuits aren't an argument against AI in insurance. They're an argument against using AI badly. Used correctly, a computer use agent doesn't make the denial decision. It handles the data work so your human adjusters can make better decisions, faster, with less burnout. That's the version of insurance automation worth building. Start at coasty.ai and see what 82% on OSWorld actually looks like in a real workflow.

Want to see this in action?

View Case Studies
Try Coasty Free