Ask anyone who works an insurance file what eats their day, and you'll hear some version of the same answer: finding things. The limit is in the policy, the exclusion is in an endorsement, the context is in an email, and the one detail that matters is in a note somebody wrote two years ago. Keyword search hands you a pile of documents and wishes you luck.
Vertafore thinks it has a better way. Its new Velocity Document Intelligence Agent, which runs inside ImageRight, lets agency, MGA and carrier staff ask a question in plain English and get an answer back with citations down to the page. In early testing, Vertafore says, document search and review time dropped by up to 90%.
We asked David Kallman, Vice President of Product Management at Vertafore, where that saved time actually goes, what happens when the AI is wrong, and which jobs look most different two years from now. His answers are below, lightly edited for format.
Q: Vertafore talks about AI reducing "distribution drag." What does that actually look like day-to-day for an underwriter or CSR who's used to keyword search?
Today, when an account manager, underwriter, claims manager, or a service rep needs to find a document or information in their system, they face a keyword search, often resulting in a pile of disconnected hits. You type a term, get a list of documents that happen to contain it, and then you have to open them one at a time and piece the answer together yourself. The answer to even a simple question is often spread across a policy, an endorsement, a few emails, and a note from two years ago, so the day-to-day work becomes opening file after file and reading until you find it.
That's literally a drag: tedious, frustrating, and not a particularly good use of a human's expertise.
The Document Intelligence Agent changes the interaction. Instead of guessing the right keywords, you ask the question the way you would ask a trusted colleague or a subject matter expert who has read everything in the file: "What are the limits and exclusions on this policy?" or "What's the status of this claim?" It works across all the ImageRight content you're authorized to see and answers in plain language, with citations back to the exact pages so you can verify it in seconds. It removes the friction of hunting and reading through a pile of documents. The judgment stays with the person.
Q: The early numbers show up to a 90% cut in document search time. Where does that saved time actually go?
Our early adopters have been telling us specifics. One reported that identifying missing renewal policies across an account with 30-plus policies, a task that used to take about an hour, took about two minutes with the Document Intelligence Agent.
Where that time goes depends on the role and the customer's priorities. For an underwriter it might be more submissions worked in a day. For a CSR it might be faster responses or more proactive service. For a claims team it might be getting to the parts of a file that need human judgment sooner.
What I'd point to is that the work that used to be "find and read" becomes "review and decide." That shifts capacity toward the work customers actually want their team doing to address what matters most to their business.
Q: Natural-language search sounds great until it's wrong. How do you handle the risk of a confident-sounding wrong answer in something like a coverage dispute?
It's the right concern, and it shaped how we built the agent. The starting point is that every answer is grounded in the documents and cited down to the page. Users can click through and see the exact source, so they aren't just taking the AI agent's word for anything. If the evidence isn't in the file, the agent is designed to say it can't tell rather than guess, because a confident wrong answer is worse than no answer.
We're also intentional about what the agent isn't doing. It summarizes what is in documents. It doesn't make underwriting, claims, legal, or coverage decisions, and it isn't meant to settle a dispute. A person stays in charge of the decision. The agent is also read-only, so it can't change anything in the file, and it only draws on content the user is already permitted to see by permissions set by their organization.
Beyond the design, we measure it. We test accuracy and citation quality against defined targets and collect thumbs up or thumbs down feedback from users to keep improving. In a coverage dispute, I'd want the citation to be the thing that's trusted, and the AI agent to be the fast path to it.
Q: You've framed this as a step toward the "Agentic Agency" and "Digital Underwriter." What's the next capability that has to exist before that vision is real?
This vision is already becoming a reality for us. We're actively delivering live, embedded AI in the solutions that independent agencies, MGAs, and carriers use today, and the Document Intelligence Agent is a continuation of that work. Getting to a fully agentic agency is the next stretch of the same journey, so here is what it takes.
Before an agent can act, it has to understand reliably. That's the foundation we're building now: deep, accurate understanding of the content in ImageRight, with the permissions, structure, and context that come with it.
The next steps build on that. Agents need to work across more of the picture (across files, policies, and claims, not just one), and they need to work alongside other agents and workflows in a governed way. Trust and control matter as much as capability, so each step has to earn its way in with accuracy, auditability, and a human in the loop where it counts.
Q: For an agency, MGA or carrier evaluating this, what's the actual implementation lift: data cleanup, system integration, staff retraining?
Our approach to AI gets at the heart of this question. Because we're embedding AI agents directly into the solutions our customers already use and trust, implementation is lighter than most people expect, especially compared to their other experiences with other AI tools.
There's no big project required to get started. The Document Intelligence Agent runs inside ImageRight on the content that's already there. Documents don't move, and the permissions you already have carry through, so there's no separate system for people to adopt or manage. And on retraining, asking a question in plain language is a small change for users. Most of the enablement is helping teams learn what to ask and how to use the citations to check the answers.
Q: Source citations down to the page level is a nice trust feature. What happens when the underlying documents are messy, mislabeled, or scanned images?
That is the normal state of insurance files, so we built for it. Scanned documents are processed with OCR so the text is searchable, and because the agent works from the content itself, a mislabeled document title is less of a barrier than it is for keyword search on names.
It's not perfect, and we don't pretend it is. Poor scans can reduce accuracy. When the agent can't find or read the evidence, it should say so rather than fill the gap, and the page-level citation lets the user look at the original image and confirm. We treat those gaps as something to surface, not hide.
But we're also hearing from users that Document Intelligence Agent can help when their information is less than perfect. Because the agent uses the meaning and context of the question, early testers found relevant results even in mislabeled documents. Those are results that would be easy for both a person and keyword searching to overlook.
Q: Two years from now, what job function inside an agency, carrier or MGA looks the most different because of tools like this?
My view is that the roles with the most document-heavy work change the most, and that's underwriting assistance, account management, claims handling, and policy servicing. The shape of the job moves from gathering and reading to reviewing and deciding, with more time for judgment, relationships, and complex cases.
I don't expect these roles to disappear. I expect the people in them to handle more, with better information in front of them, and the best teams to redesign the work around that.
David Kallman is Vice President of Product Management at Vertafore, where he leads the company's vision for an agentic AI future. You can read Vertafore's announcement of the Velocity Document Intelligence Agent here.