Bevaya, Formerly Roots Automation: AI Digital Coworkers for Insurance Operations, From Pilot to Production
Insurance runs on paperwork. Every policy, claim, and renewal generates a blizzard of documents: ACORD forms, loss runs, medical bills, legal demands, certificates, endorsements, and endless correspondence, most of it unstructured and much of it read, keyed, and routed by hand. The industry employs armies of skilled people to move information from one form into another system. It is expensive, slow, and exactly the kind of work AI is now built to absorb.
A company now called Bevaya, which you may know as Roots Automation, has spent years building AI to do this work directly. Its platform deploys what it calls Digital Coworkers, AI agents that read insurance documents, extract and reason over the data, and carry operational workflows across underwriting, claims, and policy servicing. This is a deep look at what Bevaya does, the results it reports, and, just as importantly, what it actually takes to move from an impressive pilot to production automation running across a carrier's real operations. That last step is where insurance AI succeeds or fails.
A quick note on the name
The company was founded in 2018 as Roots Automation by Chaz Perera and John Cottongim, and built its reputation, along with $43.9 million in funding including a $22.2 million Series B backed by Liberty Mutual Strategic Ventures, under that name. In 2026 it rebranded to Bevaya. Same team, same mission, expanded into a full insurance automation platform. Throughout this piece, Bevaya refers to the company formerly known as Roots Automation.
The problem Bevaya set out to solve
Insurance operations are a document-processing bottleneck. A submission arrives as a stack of PDFs and emails; a claim opens with a first notice, medical records, and correspondence; a renewal requires re-reading and re-keying. Each step depends on a skilled person extracting the right information from messy, varied documents and moving it into a policy or claims system. It scales linearly with volume, it is error-prone, and it burns out experienced staff on work well below their capability.
Bevaya's answer is to give carriers a digital workforce rather than another tool. Its Digital Coworkers come pre-equipped with insurance knowledge and are powered by InsurGPT, a proprietary large language model fine-tuned specifically for insurance documents. Rather than a general model guessing at an ACORD form, InsurGPT is trained to read the industry's actual paperwork, structured and unstructured, and extract it accurately. The idea is not to speed up a human keying data, but to have the agent do the reading and the workflow, with a human reviewing.
What Bevaya actually does
Bevaya is a platform of AI agents that execute insurance operations end to end. According to the Bevaya platform, it spans three core functions:
- Underwriting automation: submission intake, loss-run processing, ACORD extraction, renewal handling, and exposure-schedule review, at 98 percent-plus accuracy.
- Claims automation: first-notice setup, claim indexing, legal-demand extraction, medical-bill processing, and claim-file summarization, reaching as high as 99 percent straight-through processing.
- Policy servicing: endorsement processing, certificate-of-insurance generation, and premium-audit automation.
Underneath sit a few important pieces: InsurGPT, the insurance-tuned model; a Workflow Canvas for building custom automations; document intelligence that turns unstructured documents into structured data; and, crucially, human-in-the-loop review with confidence scoring and grounded explainability, so uncertain items are flagged and every decision can be traced to its evidence. In a regulated, audited industry, that explainability is what makes carriers willing to let agents touch real work.
Who benefits, and the proof in the numbers
The beneficiaries are clear. Operations staff stop keying documents and move to judgment and exceptions. Carriers get faster cycle times and more throughput without adding headcount. Policyholders get quicker claims and quotes.
The evidence is in production usage, not demos. Bevaya reports more than 115 workflows in production, that three of the five largest U.S. property-and-casualty carriers use the platform, a 74 Net Promoter Score, and, in aggregate under the Roots name, over $100 million in realized customer value across 115 deployments. One customer is cited achieving a 246 percent ROI in six months with 99 percent straight-through processing. Named customers include Eastern Alliance and a range of Fortune 500 carriers, specialty and workers-compensation insurers, and top national brokers. That major carriers run this in production is the strongest possible signal in a conservative industry.
Why this matters beyond insurance operations
Bevaya is a strong example of the pattern defining enterprise AI's winners. They are not general-purpose assistants. They go deep into one high-volume, document-heavy domain, build a proprietary model tuned to that domain's real inputs, and keep humans in control through explainability and confidence scoring. Insurance operations are a near-perfect target: enormous document volume, expensive skilled labor, and clear economic value in every point of automation.
It also puts a point on the real bottleneck in enterprise AI, which is not the model. Even a purpose-built model like InsurGPT only creates value once it is deployed into a carrier's actual workflows, systems, and teams, and running reliably enough to trust. That is the same shift reshaping how work gets delivered everywhere: away from renting hours and toward outcome-based delivery that is measured and proven, which is why so many enterprises now treat this as the defining question of the future of work.
The demo automates one document. Production automates the operation
Watching a Digital Coworker read an ACORD form flawlessly is convincing. Turning that into hundreds of workflows running reliably across a carrier's live operations is a different undertaking. Getting there is a deployment project, and it has several distinct pieces, none of which the software does for you.
1. Workflow and business-rule configuration. A Digital Coworker is only useful when it reflects the carrier's actual process: its rules, its exceptions, its routing, its definitions of a good extraction. Building and tuning those workflows on the Workflow Canvas, and validating them against real cases until the output matches what the carrier's best staff would produce, is careful, expert work that determines whether the automation is trusted or overridden.
2. Integration into core systems. Value only appears when the agents connect to the systems the carrier runs on: the policy administration system, the claims platform, the document repositories, and the many inbound channels where work arrives. Wiring those together reliably and securely is substantial integration work and a frequent place timelines slip.
3. Human-in-the-loop and quality-control design. Straight-through processing is earned, not switched on. Someone has to design where the agent acts autonomously, where a human reviews, how confidence thresholds are set, and how exceptions are handled, then tune those thresholds as trust grows. Getting this right is what safely moves a workflow from 60 percent to 99 percent straight-through.
4. Operations change management. This is the piece that decides success. Experienced insurance staff are being asked to trust agents with work they have always done themselves, and to shift into reviewing and handling exceptions. If they do not trust it, they re-do the work and the gain evaporates. Building trust takes the explainability, a rollout sequenced to clear wins, redesigned roles, and a feedback loop so corrections improve the agents.
5. Measurement and continuous tuning. The ROI case has to be proven in the carrier's own numbers. Baseline cycle time, straight-through rate, accuracy, and cost before, measure the same after, workflow by workflow, and keep tuning. Without that loop, you cannot prove the value to the executives who funded it or justify expanding to the next process.
Why carriers stall, and it is not the software
None of those five require a better model. The bottleneck is capacity and know-how, not capability. Deploying a digital workforce well is a cross-functional effort spanning operations, IT, and each line of business, and most carriers do not have a spare team that understands both insurance workflows and AI deployment sitting idle to run it. A lean, fast-growing vendor keeps its own people focused on the product and the largest accounts. The carrier's staff are busy running operations. So the program lands as a few workflows in one department and never expands, and the enterprise-wide value never fully arrives.
This is not a Bevaya problem or an insurance problem. It is the defining problem of enterprise AI. Hiring a permanent deployment team for a one-time rollout is slow and expensive, and traditional staffing rents bodies by the month rather than delivering a working outcome.
Closing the gap faster
Carriers that win with operations AI treat deployment as the real project and resource it properly. That usually means a delivery layer that sits behind the vendor's product and the carrier's team and owns the last mile: configuring the workflows and business rules, integrating the core systems, designing the human-in-the-loop and quality controls, running the operations change management, and building the measurement loop that proves ROI.
This is exactly what a Virtual Delivery Center is built to do, and it works across regulated industries, insurance included. AiDOOS runs elastic delivery pods that plug in alongside a product like Bevaya and the carrier's own people, take on the configuration and adoption work, and get workflows live and running at high straight-through rates, process by process. The carrier reaches production without building a permanent team for a one-time push, and it pays for delivered outcomes, not effort. For a sense of how enterprises put this managed-delivery model to work, see how enterprises use AiDOOS as their VDC.
The sensible way to start is small and provable: pick one workflow, say first-notice-of-loss setup or loss-run processing, get it fully deployed, trusted, and running at a high straight-through rate, measure the before and after, and use that proof to expand across the operation.
The bottom line
Bevaya, the company formerly known as Roots Automation, has taken the document-heavy core of insurance operations and rebuilt it around AI digital coworkers powered by an insurance-tuned model that shows its work. Major carriers running 115-plus workflows in production, and results like 246 percent ROI in six months, say the approach is real. The technology is ready.
The question every carrier should ask is not whether the AI can read a document. It is whether the carrier will configure the workflows, integrate the systems, design the controls, earn the operations teams' trust, and prove the ROI, or whether the program will stall as a few workflows in one corner of the business. Answer that well, and the throughput and savings scale across the enterprise. Leave it to chance, and even a proven platform like Bevaya becomes another pilot that impressed and never spread. The model is no longer the hard part. Deployment is, and that is a solvable problem.
If you are deploying Bevaya, Roots Automation, or any enterprise AI, AiDOOS provides the delivery pods that get it to production and proving ROI. To explore the product itself, visit Bevaya. For more on where enterprise delivery is heading, read our North Atlantic Briefing.