Which Loss Cost Version Is Behind the Number the Assistant Quoted — and How Do You Check It Before You Repeat It?

Updated Nov 1, 2026 — added the free loss cost vintage quick card, vintage check pack, and version verifier bot book downloads to this guide.

Short answer: you can’t tell from the chat answer itself. Since Verisk’s October 28, 2026 announcement put ISO Loss Costs inside Amazon Quick, Microsoft Copilot and Claude for authorized insurers, an underwriter can surface a loss cost in seconds — but the number arrives in conversation, not attached to the filing it came from. Loss costs are versioned data: they are updated in filings, on effective dates, state by state. This is the deep dive on one of the four questions we raised in what changes for brokers when Verisk loss costs live inside Copilot, Claude and Amazon Quick. So before you repeat any chat-surfaced figure to a client, you need three things the chat window doesn’t hand you: which version, effective when, filed where. This piece is the verification habit that gets you all three.

Why does a loss cost have a version at all?

Because it isn’t a permanent fact — it’s a snapshot that gets replaced. ISO loss costs are advisory expected-loss figures built from pooled industry data for a class and territory, and ISO updates them through filings that take effect on specific dates. The loss cost in force for a class in one state this quarter may be the new vintage, or last year’s, depending on whether the updated filing has taken effect there yet and whether the carrier has adopted it on its own schedule.

That was manageable when the number came out of a manual lookup, because the lookup lived next to the filing information. In a chat answer, the number travels alone. Verisk’s release describes the integrations as access to the same content under the same entitlements, permissions and governance — it says nothing about a conversational answer carrying its version and effective date with it into your file. Assume it doesn’t. The vintage is yours to ask for.

What three things pin a chat-quoted number down?

The version and its effective date. “The loss cost” is not a complete description. The complete version is: this class, this territory, this version, effective this date. If the underwriter can’t name the effective date behind the figure, you don’t have a number yet — you have a draft of one.

The state it was filed in. Loss cost updates don’t land everywhere at once. A version effective in one state may still be pending in the next. The figure behind an indication has to be the one filed and effective for the state where the risk sits, not whichever vintage the assistant surfaced first.

The policy period it has to cover. The right version is the one in effect for the account’s policy period — the one the carrier will actually rate on. A number pulled mid-year for a renewal effective next quarter may be quoted off the current vintage and rated off the next one. That gap is where indications move between the conversation and the binder.

What exactly do you ask the underwriter?

Four questions, in plain language, cover it:

1. “Which loss cost is that — and what’s its effective date?” Not the class description alone; the vintage. Get the date, not just the number.
2. “Is that the version in effect for this policy period?” This separates “the version I can see in the assistant today” from “the version your rating system will use on this account.”
3. “Is that the filed version for this state?” Especially on multi-state accounts, where vintages can differ by location.
4. “Can you put the basis in writing?” A chat answer is not a quote, and it isn’t file documentation either. Ask for the indication with its rating basis — class, loss cost version, effective date — in an email or on the quote document itself. That’s the version of the number you can defend later.

None of this is confrontational. Underwriters who work with this data respect the question, because the vintage problem is real on their side of the desk too.

How do you check it against what the carrier actually filed?

Three checks, none of which take long. First, compare the effective date to the policy period. If the quoted version takes effect after the policy incepts, or the policy straddles a version change, ask which vintage governs the rating — and get the answer before the client hears a premium. Second, look for the rating basis on the carrier’s own documents. When the formal quote or indication arrives, the class and basis should reconcile with what you were told in chat. If the documents and the conversation disagree, the documents are the conversation to have — with the underwriter, before binding. Third, watch the moving versions you already track. Verisk’s data updates have a way of surfacing downstream in brokers’ daily work — our piece on Verisk’s October 2026 price list update and what it does to claims settlements covers the claims-side version of the same habit: when the reference data moves, every number built on it deserves a second look.

What if the number turns out to be the wrong vintage?

Then you’ve caught, cheaply, the exact error this habit exists for. A stale-vintage figure usually moves at the worst moments — at binding, when the rating engine applies the current version, or later, when an audit re-reads the account against the data that should have applied. Caught at the indication stage, it’s a corrected quote and a short explanation. Caught after you’ve repeated it to the insured as firm, it’s a credibility problem and possibly a file problem.

The fix is procedural: re-quote on the correct version, tell the client the number moved and why — “the first figure was pulled from the prior loss cost version; the carrier rates this period on the version effective [date]” is a complete, professional explanation — and document the correction the same day. Brokers who name the vintage when they quote rarely have to have this conversation at all, because the client heard the condition attached to the number the first time.

What goes in the file?

The same documentation discipline that has always applied to numbers a client acts on, extended to where the number came from. Record the figure, the class, the loss cost version and effective date you were given, who gave it, when, and what you checked it against before repeating it. If the underwriter confirmed the basis in writing, that confirmation goes in too.

This is the policy check principle applied one step earlier in the process. Our piece on the policy check as the E&O front line makes the case for issued policies; AI-surfaced indications deserve the same treatment at the quoting stage, because the reliance is the same — if you repeated it, the file should show what you relied on and what you did to verify it. And it helps to remember what the underwriter is working from when they answer: our breakdown of what underwriters actually see when they look at your submissions is useful context for why a version question gets a precise answer instead of a shrug.

FAQ

Can’t I just ask the assistant which version it’s quoting?
You’re not the assistant’s user — under this announcement, access runs to eligible, authorized insurers under their own entitlements, not to brokers. Your channel is the underwriter. Ask them, and ask for the basis in writing.

Does the chat answer show an effective date automatically?
Don’t count on it. Verisk’s release describes the integrations as the same content under existing entitlements and governance, delivered through natural-language queries; it makes no promise that a conversational answer carries version metadata with it. Treat the effective date as something you request, not something you’re shown.

Why does the version matter if the carrier applies its own multipliers anyway?
Because the loss cost is the base everything else multiplies. A wrong-vintage base produces a wrong indication no matter how carefully the carrier’s own factors are applied on top of it — and the correction lands after you’ve quoted.

The version effective date falls in the middle of the policy period. Which one applies?
That’s the carrier’s rating rule, not a universal one — which is exactly why you ask before quoting rather than assuming. Get the underwriter’s answer in writing and file it with the quote.

What’s the one habit to build out of all this?
Never repeat a loss cost figure you can’t name the vintage of. Number, class, version, effective date — if one of the four is missing, the number isn’t ready to leave your desk.

Source: Verisk Analytics, Inc., “Verisk Brings Its Proprietary Insurance Intelligence to Amazon and Microsoft AI Platforms, Expands Offerings within Claude,” GlobeNewswire, October 28, 2026. Platform capabilities and eligibility are Verisk’s, as stated in that release. The verification practices described are this desk’s broker-practice framing.

Informational only. This article explains an industry announcement and general verification practices for insurance professionals; it is not legal, compliance or coverage advice, and it does not describe any specific carrier’s systems, filings or pricing.