Bevaya Lands Insurance AI Agents Inside Teams and Outlook

Roots Automation has retired its own name. On May 28, the seven-year-old insurtech relaunched as Bevaya, a new AI agent platform built for insurance underwriting, claims, and policy servicing, and folded its old Roots product into the new brand. The platform runs on InsurGPT, a set of models the company says it trained on more than 300 million proprietary insurance documents.

The rebrand is only part of it. Bevaya’s agents are designed to run inside the software insurance teams already keep open every day, including Outlook, Microsoft Teams, and core systems such as Guidewire. That placement, more than the new logo, is what ties a small insurtech launch to a much wider shift in how enterprise AI reaches the people who use it.

What the Roots Rebrand Puts on the Table

Roots Automation was founded in 2018 by two former AIG executives and came to market in April 2019. Bevaya now becomes the company’s primary brand going forward, with the previous Roots platform absorbed into it. For existing customers, the practical change is the platform itself rather than the name on the login screen.

The platform ships with a library of pre-built agents for the three workflows that eat most of an insurer’s labor: underwriting, claims, and policy servicing. These agents read documents, analyze them, and recommend next steps across tasks like triage, clearance, rating, and coverage analysis. A built-in assistant lets staff create and manage their own agents using plain-language prompts instead of code, and the company has wrapped the whole thing in dashboards, human review tools, and governance controls.

That last layer matters more than it sounds. Insurance is a regulated business where a wrong extraction can mean a mispriced policy or a denied claim, so the appeal of any agent here rests on whether a human can see what it did and why. Bevaya’s review mode highlights the source text an answer was drawn from, which is the kind of feature compliance teams ask about first.

Insurance people deserve better tools. Bevaya is ours.

That line comes from Chaz Perera, chief executive and co-founder, in the company’s flagship launch announcement. His co-founder is John Cottongim, the company’s co-founder and chief technology officer, who spent more than 15 years on global transformation programs at AIG and Mars before the two started the business.

InsurGPT and the 300 Million Document Moat

The engine under Bevaya is InsurGPT, an ensemble of insurance-specific AI models rather than a single general chatbot. Its selling point is the training data: loss runs, supplemental applications, demand letters, and claim forms that the company says general AI labs simply cannot get their hands on.

Specialized Models, Not One General Brain

Instead of one model doing everything, InsurGPT splits the work. A loss-run model reads claims history across carrier formats. An ACORD model handles the standardized forms that run U.S. insurance. A separate OCR model, which the company calls GutenOCR, reads logos, checkboxes, and stamps on scanned and unstructured pages while keeping track of where on the page each value sat. A grounding model traces every answer back to its source document.

This is the multimodal piece the launch emphasizes: the system can analyze images and scanned material, not just clean digital text, which is most of what actually lands in an insurer’s inbox. The full InsurGPT model breakdown lists six specialized components handling tasks from page separation to structured-form extraction.

Where the Accuracy Numbers Come From

The company publishes benchmark comparisons against general-purpose models, and the gaps are wide on the tasks where insurance precision matters. On claims indexing, it claims 99% accuracy against 62% for the strongest general model tested. On underwriting extraction, it cites 93% against the 80 to 84% range it reports for GPT-5.0 and Gemini 3.0 Pro.

These are vendor figures, not independent test results, so the right way to read them is as a claim about where a narrow, domain-trained model beats a broad one. The argument is plausible. A model that has seen 300 million property and casualty (P&C) documents should outperform a general model on a loss run, the same way a general model still wins at drafting an email.

Where Bevaya Plugs Into Microsoft 365

For a productivity audience, the integration list is the most interesting part of the announcement. Bevaya does not try to become the place insurance staff spend their day. It connects to the systems they already live in and pushes its output there.

The named integrations cover both insurance core systems and everyday collaboration tools:

• Guidewire, the dominant policy and claims platform across P&C carriers
• Outlook, where most submissions and claim correspondence still arrive as email and attachments
• Microsoft Teams and Slack, used for the agents’ alert notifications and human handoffs
• SFTP and custom HTTP APIs, for the older internal systems every insurer still runs

That design choice puts Bevaya in the same lane as a growing list of third-party agents reaching workers through Microsoft’s surfaces rather than around them. It echoes the logic behind Perplexity’s Computer agent landing as add-ins across Word, Excel, PowerPoint, and Outlook, where the agent meets users inside the app instead of asking them to switch context.

Agents Moving Into the Tools People Already Use

Step back from insurance and the pattern is hard to miss. The interesting AI products of 2026 are not standalone chatbots. They are task-specific agents that embed into the apps where work already happens, and they are increasingly aimed at vertical industries with their own data and rules.

Gartner put a number on the shift. The research firm predicts that 40% of enterprise apps will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. The same Gartner forecast on task-specific enterprise agents frames 2026 as the year agentic AI moves past individual productivity into shared workflows, exactly the territory Bevaya is targeting.

Microsoft sits at the center of this because so much enterprise work passes through its tools. Its own assistant has been reworked around the same idea, as the Microsoft 365 Copilot redesign built around in-app adoption showed, reporting usage jumps inside Word, Excel, PowerPoint, and Outlook once the agent lived where people already worked. Third parties are following the marketplace route too, as with Denodo listing its Agora data service on the Microsoft Marketplace for one-click agent access.

Bevaya is the insurance entry in that catalog. It brings a vertical model and a stack of pre-built agents, then routes the results through Teams and Outlook so the AI shows up as an alert or a draft rather than another tab. For IT teams managing a Microsoft estate, that is the part worth tracking: the agent layer is arriving through tools they already administer, with their own governance and identity questions attached.

The Numbers Behind the Pitch

The launch leans on a track record, not just a roadmap. The company says its agents are already in production at scale, and the figures it puts forward are specific enough to weigh.

• 115 production deployments across insurers, brokers, and third-party administrators (TPAs, firms that process claims on behalf of insurers)
• Three of the top five P&C carriers, plus three of the top 10 brokers and three of the top 20 TPAs
• $100M+ in realized customer value claimed for 2025 alone
• 3 to 4 times capacity gains reported on the workflows the agents handle

Read with a cold eye, these are self-reported metrics from a company announcing a product, and “realized customer value” is a soft figure that bundles labor savings, faster cycle times, and avoided cost into one headline. None of it has been independently audited. What the deployment count does signal, though, is that this is not a launch-day prototype; the agents have been running inside large carriers under the old Roots name for a while.

If the embedded-agent thesis holds, the next year tells the story through adoption inside those Microsoft surfaces. If insurance staff actually act on agent outputs in Outlook and Teams instead of routing around them, Bevaya becomes a template other verticals copy. If the work stays parked in dashboards nobody opens, the rebrand will have changed the logo and little else.

Frequently Asked Questions

What is Bevaya?

Bevaya is an AI agent platform for insurance launched on May 28, 2026 by Roots Automation, the New York insurtech founded in 2018. It provides pre-built agents for underwriting, claims, and policy servicing, and it replaces and rebrands the company’s earlier Roots platform.

What is InsurGPT and how is it different from ChatGPT?

InsurGPT is the ensemble of insurance-specific models powering Bevaya, trained on more than 300 million proprietary insurance documents such as loss runs and ACORD forms. Unlike general models, it is built only for insurance document tasks, and the company reports it outperforms GPT-5.0 and Gemini 3.0 Pro on its insurance benchmarks.

Does Bevaya work with Microsoft Teams and Outlook?

Yes. Bevaya integrates with Outlook for email and attachments and sends alert notifications through Microsoft Teams and Slack. It also connects to Guidewire, SFTP, and custom HTTP APIs, so agent output reaches staff inside tools they already use rather than a separate application.

How many companies already use the platform?

The company reports more than 115 production deployments across insurers, brokers, and TPAs, including three of the top five property and casualty carriers. These figures are self-reported and have not been independently audited.