An AI agent can take board meeting prep from a multi-day scramble down to a reviewable draft. It gathers the recurring inputs that go into every deck, financials, KPIs, pipeline, hiring, and product progress, from the systems that hold them; refreshes the standard sections; drafts the narrative and an agenda; pulls last meeting's action items forward with their current status; and flags anything missing or needing a decision. What it does not do is send the deck. A board deck is a high-stakes document, so the agent stops at a complete draft and hands it to a founder or executive for review and approval before anything goes out.

This post covers the board-prep job specifically: what the agent assembles, how it keeps the numbers honest, and where the human checkpoint sits. The same input-gathering pattern shows up in adjacent reporting work like the quarterly business review and weekly KPI reports, but a board deck carries more weight, so accuracy and approval matter more here than anywhere else.

What board prep actually takes
What board prep actually takes

What board prep actually takes

Board prep is repetitive in structure and exhausting in execution. The deck looks roughly the same every cycle: a financial summary, the headline KPIs, pipeline and bookings, hiring against plan, product and roadmap progress, and a forward-looking ask or set of decisions. The structure rarely changes. What changes is the data, and pulling fresh data from five or six different systems is where the days go.

For most founders and operators, the cycle looks like this. Someone exports the latest financials from the accounting tool. Someone else pulls KPI numbers from analytics or a data warehouse. Pipeline comes out of the CRM. Hiring status comes from the recruiting tracker or a spreadsheet. Product updates come from whoever runs the roadmap. Then all of it gets reconciled, formatted into the standard slides, wrapped in a narrative, and checked against last quarter so nothing contradicts what the board was told before. The mechanical part of this, gathering and refreshing, eats the hours. The judgment part, deciding what the numbers mean and what to ask the board, is where the founder's time is actually worth spending.

An agent takes the mechanical part. It does the gathering and the first-pass assembly so the founder starts from a populated draft instead of a blank deck, and spends their time on judgment and message rather than copy-paste.

Gathering the recurring inputs

The first job is collecting the inputs that appear in every board deck. The agent connects to each system of record and pulls the current figures. You define which inputs belong in your deck and which source each one comes from; the agent does not guess.

A typical board deck draws on:

Because every figure comes straight from its source, the draft is traceable. Each number carries a note of where it came from and when it was pulled, which matters enormously when a board member asks "is this the same revenue figure we saw last month" during the meeting. The agent never invents a number to fill a slide; if it cannot source a figure, it leaves the slot empty and flags it.

Refreshing the standard deck sections

With the inputs gathered, the agent refreshes the standard sections of your existing deck template. This is deliberately not a redesign. Boards expect continuity. The same slide structure, period over period, is a feature, not a limitation, because it lets directors compare cleanly across meetings. The agent updates the content inside that structure.

For each recurring section, the agent drops in the refreshed figures, recomputes the period-over-period and plan-versus-actual comparisons, and updates the charts that track trends over time. The financial slide shows this quarter against last quarter and against plan. The KPI slide shows the current values against their targets and the trajectory. The pipeline slide shows current coverage against the bookings goal. The hiring slide shows actual headcount against the plan. None of these require creative judgment; they require correct data placed correctly, which is exactly what an agent is good at.

Where the deck includes commentary that is purely descriptive, "revenue grew to X this quarter, up from Y," the agent drafts that text directly from the figures. Where the commentary requires interpretation or a strategic claim, the agent leaves a clearly marked placeholder for the founder rather than manufacturing a narrative the data does not support.

Drafting the narrative and agenda

A good board deck is not just slides; it is a story with an agenda that respects everyone's time. The agent drafts both, as a starting point for the founder to shape.

For the narrative, the agent assembles a draft of the standard framing: a one-paragraph summary of the period, the key wins, the areas that need attention, and the headline numbers in context. This draft is grounded entirely in the data it gathered, so it does not overstate or invent momentum. The founder then edits it to add the strategic message, the parts only they can write, while keeping the factual scaffolding the agent produced.

For the agenda, the agent proposes a structure based on the deck contents and the open items: a review of progress against last meeting's commitments, the standard reporting walkthrough, the specific decisions that need board input this cycle, and time for discussion. Putting the decision items up front in the agenda, rather than burying them after forty slides of reporting, is a small structural choice that makes meetings more productive. The founder adjusts the agenda and timing to fit how the board actually runs.

Tracking action items from last meeting

One of the most valuable and most-skipped parts of board prep is following up on what was committed last time. Boards remember what they asked for. Walking into a meeting unable to report on the prior cycle's action items erodes trust faster than a soft quarter does.

The agent reads the previous meeting's notes or action log, extracts each open item, and checks its current status. For an item like "close the VP of Sales hire by end of quarter," it checks the hiring tracker. For "reduce burn to under the agreed monthly target," it checks the financials. For "ship the enterprise tier," it checks the product roadmap. It then assembles a clean status view: what was committed, who owned it, and where it stands now, done, in progress, or slipped.

This turns the opening of every board meeting from an improvised recollection into a documented follow-through report. It also carries forward anything still open into the new action list, so commitments do not quietly evaporate between meetings. The same closed-loop tracking pattern underlies the quarterly OKR tracking agent, where keeping commitments visible across cycles is the entire point.

Flagging gaps and decision items

An honest board draft is as much about what it cannot confirm as what it can. The agent surfaces three categories of flag rather than hiding them inside otherwise-finished slides.

This flagging is what makes the draft trustworthy. A deck that quietly fills every slot looks complete but hides its own weak points. A deck that says "this figure could not be sourced" and "this number jumped, please verify" is one a founder can actually rely on. Designing agents to surface uncertainty rather than mask it is a core safety practice, covered in more depth in AI agent safety and guardrails.

Why human review is non-negotiable

Board reporting is exactly the kind of work where the agent must stop short of acting on its own. The output goes to the people the company answers to, carries legal and fiduciary weight, and shapes decisions about funding, strategy, and leadership. The cost of an error, an overstated number, a missed risk, a misframed narrative, is high and hard to walk back.

So the workflow is deliberately gated. The agent produces a complete draft plus a change summary: what it refreshed, what it could not source, what anomalies it flagged, and what decisions it surfaced. A founder or executive reviews the draft against that summary, traces any figure they want to verify back to its source, edits the narrative and judgment calls, and approves the deck for distribution. Only then does it go out, and the agent does not send it; a person does. This is the human-in-the-loop checkpoint, and for board work it is mandatory rather than optional. The pattern, and how to build it into any agent workflow, is laid out in how to add a human in the loop to an agent.

If you are new to the underlying idea of an agent doing multi-step work like this and stopping for approval, what is an AI agent and the glossary explain how an agent reasons across systems and tools, which is what separates this from a static template or a single export.

How Gravity handles board meeting prep

Gravity is an AI agent platform. You describe your board deck in plain words: which sections it has, which system each number comes from, where last meeting's notes live, and who reviews before it ships. An expert-built agent handles the assembly. It connects to your finance system, analytics, CRM, recruiting tracker, and roadmap; refreshes the standard sections; drafts the narrative and agenda; pulls the prior action items forward with status; and flags every gap, anomaly, and decision item.

Then it stops. The finished draft and a change summary land with you for review. You verify the numbers, sharpen the message, and approve, and nothing reaches the board until you do. You do not build a pipeline, write scripts, or wire integrations yourself. Pay per use: $1 equals 1,000 credits, and you only pay when the agent runs.

If you want to stand up a workflow like this, setting up your first AI agent walks through the path from a plain-language description to a running agent. Board prep is a strong first use case precisely because the structure repeats every cycle: define it once, run it before each meeting, and review the draft instead of building the deck from scratch. The closely related investor update drafting agent follows the same gather-draft-review pattern for the monthly investor email, with the same human sign-off before anything sends.

FAQ

Can an AI agent prepare a board deck for me?

An AI agent can prepare the draft. It gathers the recurring inputs from their sources, refreshes the standard deck sections, drafts the narrative and an agenda, and pulls forward last meeting's action items with their status. It does not send anything. A board deck is a high-stakes document, so the agent stops at a complete draft and routes it to you for review, edits, and approval before distribution.

How does the agent make sure the numbers are accurate?

The agent pulls every figure directly from the source system you point it at: the accounting tool for financials, the analytics or data warehouse for KPIs, the CRM for pipeline, and so on. It does not invent or estimate numbers. Each metric in the draft carries a note of where it came from and the date it was pulled, so you can trace any figure back to its source during review. Where a source is stale or a number looks off against the prior period, the agent flags it rather than papering over it.

Does the agent need my approval before the deck goes to the board?

Yes, and that is by design. The agent produces a draft and a summary of what changed, what it could not source, and what needs a decision. A founder or executive reviews and edits it, then approves it for distribution. Nothing reaches the board until a human signs off. This human-in-the-loop checkpoint is non-negotiable for board reporting.

What inputs does a board-prep agent gather?

The recurring ones: financials such as revenue, burn, runway, and cash position; the KPIs your board tracks; sales pipeline and bookings; hiring progress against plan; and product or roadmap milestones. It pulls each from its system of record, refreshes the standard slides, and notes anything missing. You define exactly which inputs and which sources belong in your deck.

Can the agent track action items from the last board meeting?

Yes. The agent reads the previous meeting's notes or action log, pulls each open item forward, and checks its current status against the relevant system or owner. The deck then shows what was committed last time, what is done, what is in progress, and what slipped, so the board sees follow-through rather than starting each meeting from a blank page.