Your Insurance Claims Team Is Drowning in Busywork. A Computer Use AI Agent Can Fix That in Days, Not Years.
The insurance industry spent $25.7 billion last year on claims adjudication. Nearly $18 billion of that was, by one analysis, completely unnecessary. That's not a typo. That's $18 billion in wasted back-and-forth, manual review, re-submissions, and human beings copying data from one system into another, over and over, every single day. Meanwhile, the average auto claim now takes 22.3 days to close. Property claims sit at 23.9 days. And insurers keep telling us the system is working. It's not. It's a slow-motion disaster that the industry has normalized because nobody wanted to do the hard work of actually fixing it. That's changing right now, and the weapon of choice is the computer use AI agent.
The 'We Have Automation' Lie That Insurers Keep Telling
Ask any insurance executive if they've automated their claims process and they'll say yes. What they mean is they bought some RPA bots in 2019, bolted them onto legacy systems, and called it a day. RPA, which is the robotic process automation that companies like UiPath built their entire business on, is basically a fragile screen-scraping script. It breaks every time a UI changes. It can't handle exceptions. It can't read an unstructured PDF, make a judgment call, or navigate a system it wasn't explicitly programmed for. There's a reason the Reddit community for UiPath professionals is openly debating whether RPA is dead. It's not dead, but it's not enough either. The gap between 'we have some bots' and 'we have genuine end-to-end claims automation' is where billions of dollars go to die. Real computer use AI, the kind that actually sees a screen, understands context, and takes action like a human would, is a completely different category. Most insurance ops teams haven't touched it yet.
The Numbers That Should Make Every Claims VP Uncomfortable
- ●$25.7 billion: what providers spend annually just on claims adjudication, per a February 2025 Premier Inc. analysis
- ●$18 billion of that is flagged as 'potentially unnecessary' expense from claims that should have been paid on first submission
- ●22.3 days: average auto claim repair cycle time in 2025, UP from prior years despite supposed automation investments
- ●300%: how much longer medical reviews extend life insurance claim processing times
- ●60%: the reduction in manual processing time that AI-driven workflows are achieving in documented deployments
- ●50%: corresponding cost reduction per claim in those same implementations
- ●96% of employees at organizations using AI for claims reported measurable time savings, per Google Cloud case data
- ●Only 34% of insurance companies have meaningfully adopted AI agents, up from 8%, meaning 66% are still running on spreadsheets and hope
"Nearly $18 billion was potentially wasted arguing over claims that should have been paid at the time of submission." That's not an efficiency problem. That's a structural failure that the industry has decided to pass on to customers and providers.
The Dark Side: When Insurers Automate Wrong
Here's where it gets ugly. Big insurers didn't wait for good AI. They grabbed whatever was available and used it to deny claims faster. UnitedHealth's nH Predict algorithm got sued for having a reported 90% error rate on claim denials. Cigna's PxDx algorithm allegedly let reviewers spend an average of 1.2 seconds per claim before rejecting it, and a lawsuit claims they denied over 300,000 claims that way. These are not computer use AI agents making thoughtful decisions. These are blunt denial machines wearing an AI costume. The backlash is real, the lawsuits are piling up, and regulators are paying attention. This is what happens when you use AI to cut costs at the expense of accuracy. The right use of computer-using AI in insurance isn't to deny claims faster. It's to process them correctly, the first time, without the $18 billion in unnecessary back-and-forth. Those are completely different goals and the industry desperately needs to understand the difference.
What Actual Computer Use AI Looks Like in a Claims Workflow
A real computer use agent doesn't just read structured data. It opens the actual desktop applications your team uses. It navigates browser-based portals. It reads PDFs that were scanned sideways by a fax machine in 2003. It cross-references policy documents, pulls loss run reports, checks coverage limits, flags discrepancies, and fills out the adjudication form, all without a human touching a keyboard. This isn't theoretical. The technology exists right now. An AI agent can handle first notice of loss intake, triage by claim type and complexity, pull relevant policy data from your CMS, flag potential fraud indicators, and route to a human only when genuine judgment is needed. What used to take an adjuster 45 minutes of screen-switching and data entry takes the agent under 4 minutes. That's not an estimate. That's what operations teams are reporting in 2025 deployments. The bottleneck was never the adjuster's brain. It was the mechanical, repetitive computer work surrounding every single decision they made.
Why Coasty Is the Computer Use Agent Insurance Teams Are Actually Deploying
I'm going to be straight with you. There are a lot of AI tools claiming to automate claims workflows. Most of them are wrappers around API calls that can't touch a real desktop. Coasty is different in a way that matters specifically for insurance. It's a computer use agent that operates on real desktops, real browsers, and real terminals. It scored 82% on OSWorld, which is the benchmark that actually tests whether an AI can use a computer the way a human does. That's higher than every competitor right now, including Anthropic's Computer Use and OpenAI Operator. In insurance, that gap matters enormously. Your claims systems aren't all API-friendly. Your adjusters work across four different legacy portals, a shared drive, an Excel workbook from 2015, and a claims management system that hasn't had an API update since the Obama administration. Coasty doesn't care. It sees the screen, it understands the task, and it executes. You can run agent swarms for parallel claim processing, meaning 50 claims handled simultaneously instead of one adjuster working through a queue. There's a free tier to start, BYOK support if your compliance team needs it, and cloud VMs so you're not installing anything on sensitive infrastructure. The ROI math for insurance is almost embarrassingly obvious once you run it.
The insurance industry is at a genuine fork in the road. One path is more of the same: RPA bots that break, adjudicators buried in copy-paste work, $18 billion flushed annually on preventable back-and-forth, and the occasional class-action lawsuit from using blunt AI hammers on nuanced coverage decisions. The other path is actual computer use AI that handles the mechanical grind, frees your adjusters for the work that genuinely requires human judgment, and closes claims in days instead of weeks. This isn't a distant future thing. The teams winning right now are the ones who stopped waiting for their legacy vendors to catch up and started running real computer use agents on real workflows today. If you're still manually keying claim data in 2025, that's a choice, and it's an expensive one. Go see what Coasty can actually do at coasty.ai. The free tier exists for exactly this reason.