Verisk Just Plugged Claude Into Restoration. Here’s What That Actually Means.

If your social caught fire last week with everyone basing info off 1 new drop or even just speculation, you probably saw the headline. Let’s cut through it. What landed, what didn’t, and what an operator should do about it.

What actually shipped

On May 5, 2026, Verisk announced two new MCP connectors for Anthropic’s Claude. Per Verisk’s official press release, the two are:

• Verisk Underwriting Intelligence (ISO Indications). Aimed at carrier underwriters and actuaries. Surfaces loss cost trends, experience insights, and filing signals from Insurance Services Office through conversational queries inside Claude.
• Verisk XactRestore. Aimed at restoration contractors. Surfaces Verisk’s researched pricing and estimating intelligence through conversational queries inside Claude.

For shops in the restoration trade, only the second one applies. Verisk’s stated number for the XactRestore connector: experienced contractors can expect to save between 30 minutes and 2 hours per estimate.

From Lee Shavel, Verisk’s CEO, in the announcement: “Trust is the foundation of insurance, and that doesn’t change as new technologies emerge.”

In plain English: Claude can now call Verisk’s pricing database in real time. Same data, same rules, same governance. Claude doesn’t memorize anything. It just asks the question, gets the answer, and moves on.

That last paragraph is the one most takes have skipped over.

The protocol is the real story

MCP is short for Model Context Protocol. It’s a standard, not a product. Anthropic published it, is building it out, and is signing up vendors across the enterprise software landscape to ship connectors against it. Verisk just shipped two. Plenty more are coming, from other vendors, in other categories.

For a restoration shop, this changes the question.

The question stops being “does this one connector fit my workflow today?” The question becomes: what happens when every tool in your stack (pricing, job management, invoicing, photos) can answer questions in plain English?

Picture asking Claude (or whatever AI agent you run) something like this in 12 months. “Pull the three biggest water jobs from Q1, give me the average cycle time, the supplement hit rate, which carriers paid the fastest, and where my labor margin diverged from plan.” That query touches your job management system, your QuickBooks, your AR tracker, your photo log, and your CRM.

Without MCP, that’s a half-day of pivot tables for whoever you ask to build it. With MCP, that’s a 30-second conversation.

The Verisk-Claude connector is the first proof point in this trade. But it won’t be the last.

Worth saying out loud: Verisk has historically been behind the 8-ball on tech adoption. Updates ship slow, pricing data lags real-world cost shifts, and the platform feels like it was built in a different era because a lot of it was. The MCP play is the first time in a while where Verisk is moving with the curve, not three years behind it. Credit where it’s due.

Operator read: The connector is today. MCP is the next five years. Most of the “hot” takes you’re seeing only see today.

Quick sidebar before we get into the math. Anthony Nelson’s piece in C&R Magazine on this same announcement is a good read. He came at it from a different angle than I did. Worth the few minutes.

Your data stays yours

Worth being clear on what does and doesn’t flow through these connectors.

The two MCP connectors Verisk shipped (Underwriting Intelligence for carriers, XactRestore for contractors) are governed separately. They don’t share data with each other. What you do inside the XactRestore connector stays on your side of the fence.

Your estimates, your pricing, your job data. That’s all yours. The carrier can’t reach through the connector to use any of it against you in negotiations. The connector reads Verisk’s pricing data into Claude. It does not push your data the other direction.

There’s a fear circulating that using the connector will feed your contractor pricing back into Xactimate’s national pricing database and further suppress what carriers will pay. That’s not how this works. The XactRestore connector and Xactimate’s pricing intelligence are two different data streams. Using one does not contaminate the other.

The operator math

Set the protocol question aside for a minute. Run the math on the connector as it ships today.

Verisk’s stated time savings is 30 minutes to 2 hours per estimate. Take the conservative end. Half an hour.

A mid-size restoration shop writes between 8 and 12 estimates per week across its estimators. Call it 10. Half an hour each is 5 hours per week. At a loaded estimator cost between $75 and $100 per hour, that’s $375 to $500 per week, or roughly $19,500 to $26,000 per year on the conservative read.

That’s the upside. The downside is the cost of capturing it:

• The XactRestore Pro subscription, $199 per user/month ($2,388 per user/year)
• A Claude Max subscription, starting at $200 per user/month
• Setup time and the cost of bringing the team up to speed on a new tool
• The drag of any new tool entering a production workflow

At those numbers, a mid-size shop with 3 estimators is paying around $14,400 a year for both subscriptions against $19,500 to $26,000 a year in upside. Net positive, but the margin is tighter than the headline claims suggest.

For shops already running XactRestore, the only incremental cost is Claude Max (around $7,200 a year for 3 estimators), and the math works cleanly. For shops on Xactimate without XactRestore, you’re adding both subscriptions, and the margin narrows. Pilot before you commit.

The bigger point: the math is workflow-specific. If pricing lookups are your biggest bottleneck, the connector solves something real. If they’re not, it’s solving something you don’t particularly need solved. And in most shops I’ve looked at, pricing lookups aren’t the biggest bottleneck.

The chart above is what most owner/operators recognize from their own week. The Verisk connector touches a slice of the top bar. Not the whole bar. Definitely not the other two.

What this quarter should actually look like

Three moves. In order:

1. Get fluent in MCP vocabulary. Read Anthropic’s documentation at modelcontextprotocol.io. Not because you have to build anything. Because in 18 months you’ll be picking tools partly on whether they support MCP, and the operators who know that vocabulary now will negotiate from a better seat.
2. Pilot the connector on a closed estimate. If you’re already on XactRestore, run the new connector against an estimate you completed in the last 60 days. Compare the AI-drafted scope to what you actually wrote and submitted. That delta is your benchmark. Verisk’s marketing number isn’t.
3. Stop optimizing the wrong workflow. Look at your week. Where are the hours actually leaking? For most shops it’s some mix of estimate writing, carrier follow-up, and AR chasing. Solve the biggest leak first. Layer in tools like the new connector to refine the smaller leaks once the larger ones are stopped.

The shops paying attention to the operations, not the press releases, are the ones who’ll pull ahead of the competition.

Bottom line

The Verisk-Claude announcement doesn’t reshape your daily estimating workflow. It reshapes what your tech stack will look like 18 months from now. The protocol layer that Verisk and Anthropic just laid down will get more connectors, from more vendors, across more categories. The operators who recognize that and start positioning their stack accordingly will pull away from the operators who waited for AI to “be ready.”

Operations is the play. Everything else is timing. Bet accordingly.

— Jake

Sources & further reading

• Verisk official announcement (May 5, 2026). Primary source for the press release language, the Lee Shavel quote, and the time-savings claim.
• Reinsurance News coverage of the announcement.
• “Claude and Xactimate: The Headline vs. The Reality” by Anthony Nelson in C&R Magazine. A different read from this one, focused on the Xactimate vs. XactRestore distinction. Worth comparing.
• Anthropic’s Model Context Protocol documentation.