· I'mBoard Team · governance · 9 min read
Anthropic Just Shipped 10 Finance Agents. Here's What Changes for Board Governance.
Anthropic's finance agents automate the analyst's workflow—pitchbooks, earnings analysis, KYC, ledger reconciliation. Boards face the same shift, with a higher bar: when an agent produces a board number, the new question is "where did this come from?" Here is what provenance-first governance looks like.
Anthropic just put a fleet of finance agents into the hands of analysts. Its agents for financial services release ships ten ready-to-run agent templates—plus a library of pre-built Agent Skills—that take on the grunt work of the finance desk: building pitchbooks, reviewing earnings, screening know-your-customer (KYC) files, reconciling the general ledger, and running month-end close. Ten distinct workflows, each one a task a junior analyst used to spend a week on.
The reaction in finance has been predictable: excitement, anxiety, and a scramble to figure out which roles this touches. But the more interesting question is the one almost nobody is asking. Board governance is built on exactly the same shape of work the finance agents just automated—and boards have a higher bar to clear.
This post walks through what Anthropic actually shipped, why the board pack is the same kind of artifact as a pitchbook, and why the thing that protects a board in an agent-assisted world is not better prompts. It is provenance: the ability to say, for every number in front of the directors, exactly where it came from and which definition it follows.
What Anthropic Actually Shipped
The finance agents are not a chatbot bolted onto a spreadsheet. They are task-scoped agents that read source material and produce a finished analyst artifact. A few of the templates Anthropic shipped:
- Pitch builder — assembles target lists, runs comparables, and drafts the pitchbook.
- Earnings reviewer — reads transcripts and filings, updates the model, and flags what changed versus the thesis.
- General ledger reconciler — reconciles ledger accounts and runs net-asset-value calculations.
- Month-end closer — runs the close checklist and produces the close report.
- KYC screener — assembles entity files, reviews source documents, and packages escalations for compliance.
Behind the templates sits a library of pre-built Agent Skills—discounted cash flow models, comparable-company analysis, earnings analysis, due-diligence data packs—that the agents draw on. The common thread: each one ingests a pile of source documents, applies a methodology, and emits a synthesized artifact that a human then reviews. That is the whole pattern—synthesis over source material, governed by a methodology.
Hold that sentence in your head, because it is also the exact job description of the people who assemble a board pack.
The Board Pack Is a Pitchbook in a Different Suit
Walk through what happens before a quarterly board meeting. A chief of staff or corporate secretary pulls financials from the accounting system, growth metrics from the data warehouse, pipeline from the CRM, and narrative updates from each function head. They reconcile the numbers, format them into a consistent deck, write a summary of what changed, and ship it to the directors as the board pack.
Different buyer. Different artifact. Same shape. An analyst building a pitchbook and a corporate secretary building a board pack are running the identical loop: ingest source material, apply a methodology, synthesize a reviewed artifact. If an agent can do the first, it can do the second—and the economic pressure to let it is going to be just as strong.
We have written before about why agent-ready board infrastructure is becoming table stakes rather than a novelty. The Anthropic launch is the clearest signal yet that the analyst’s version of this future arrived first, and the board’s version is close behind.
But there is a catch that makes the board case harder than the analyst case, and it is worth being precise about.
The Higher Bar: A Board Number Is a Fiduciary Number
When an analyst’s DCF is wrong, the deal gets repriced or the trade loses money. Painful, but bounded. When a board number is wrong, the consequences route through fiduciary duty: directors made a decision—on a financing, a hire, a pivot, an option grant—on the basis of a figure that did not mean what they thought it meant.
And board numbers are unusually easy to get wrong, because the definitions are contested. Ask three SaaS finance leaders to define ARR and you will get three answers. Does it include the recurring portion of a multi-year contract divided by term, or the first-year value? Does it count contracts that are signed but not yet live? Net revenue retention has the same problem, and so does almost every metric a board actually steers on. In our experience the figure that trips up a board is rarely miscalculated—it is calculated correctly against the wrong definition. Definitional drift, not arithmetic error, is what quietly poisons board reporting.
An agent does not resolve this ambiguity. It inherits it. Point a board-pack agent at your systems and it will happily produce an ARR figure—using whatever definition is implicit in the data it found, with no flag that a different team three slides later used a different one. The agent makes the synthesis faster and the ambiguity invisible at the same time. That is the dangerous combination.
Provenance Is the Differentiator
The defense is not a better model. It is provenance—a durable answer to “where did this number come from?” attached to every figure a board sees. Concretely, that means three things travel with each metric:
- A canonical definition. ARR means one specific thing, written down, not whatever the source system happened to compute.
- A cited source for that definition. The standard the definition follows, named and linked, so it can be audited rather than trusted.
- A traceable derivation. The path from raw data to the displayed number, reproducible on demand.
This is exactly the layer we have been building at imboard, and it is why the Anthropic launch reads to us as validation rather than threat. Our public KPI catalog publishes a canonical definition for every board metric we track, each one carrying its source attribution. The ARR definition and the net revenue retention definition, for example, both cite the SaaS Metrics Standards Board verbatim—we do not paraphrase a published standard into our own words, we name it and link it. (imboard is not affiliated with or endorsed by the SaaS Metrics Standards Board, and nothing here implies Anthropic endorses imboard; we reference both as independent sources.) That is the difference between a number you can defend in front of an audit committee and a number you merely hope is right.
The plumbing that lets an agent consume those definitions is the imboard MCP server: instead of an agent guessing at what ARR means from the shape of your data, it reads the canonical definition and computes against it. The full derivation methodology lives in our ontology methodology documentation. The point of the architecture is simple: an agent should be fast and show its work. Speed without provenance is how a wrong number reaches a director faster than ever.
What Boards Should Do Now
You do not need to wait for a board-pack agent to land in your stack. The provenance discipline pays off the day you adopt it, agents or no agents:
- Write down your KPI definitions. One canonical definition per metric, owned by one function. If you cannot point to the definition, neither can an agent.
- Cite the standard. Where a published standard exists, adopt it by reference rather than reinventing it. It moves the metric from “trust us” to “here is the source.”
- Make derivation reproducible. A board number that cannot be regenerated from source is a number that cannot be audited.
- Decide your agent policy before the agent arrives. Boards that treat agent-generated material as a governance question now will not be caught flat-footed when the corporate secretary starts using one.
Anthropic’s finance agents are a preview, not an anomaly. The analyst’s workflow got automated first because the buyer was obvious and the artifact was standardized. The board’s workflow is next—and the boards that come through it intact will be the ones that made provenance a requirement before automation made it urgent.
FAQ
What did Anthropic ship with its finance agents?
Anthropic released ten ready-to-run agent templates for financial services—including a pitch builder, an earnings reviewer, a general ledger reconciler, a month-end closer, and a KYC screener—backed by a library of pre-built Agent Skills such as discounted cash flow models, comparable-company analysis, and due-diligence data packs. Each agent ingests source documents, applies a methodology, and produces a finished artifact for human review.
Why does this matter for board governance?
Assembling a board pack is the same kind of work an analyst does building a pitchbook: ingest source material, apply a methodology, synthesize a reviewed artifact. The agents that automate the analyst workflow can automate the board-prep workflow too. The difference is that board numbers carry fiduciary weight, so the accuracy and provenance bar is higher.
What is metric provenance and why is it the differentiator?
Provenance is a durable answer to “where did this number come from?” attached to every figure a board sees. It means each metric travels with a canonical definition, a cited source for that definition, and a reproducible derivation from raw data. When agents generate board material, provenance—not speed—is what makes a number defensible in front of an audit committee.
How does imboard support provenance-first board reporting?
imboard publishes a public KPI catalog with a canonical, source-cited definition for every board metric, an MCP server that lets agents compute against those canonical definitions instead of guessing, and documented derivation methodology. Standards such as the SaaS Metrics Standards Board are cited verbatim rather than paraphrased.
What should boards do before agents reach board prep?
Write down one canonical definition per KPI owned by a single function, cite a published standard wherever one exists, make every board number reproducible from source, and set an agent policy before a corporate secretary starts using one. The provenance discipline pays off immediately and protects the board when automation arrives.
Part of our Startup Governance Guide — Why every board KPI deserves source attribution, and how provenance protects directors in an agent-assisted world.
