AI in pitchbook and board pack production: from a pitch builder template to your boardroom
Published 1 May 2026
The Anthropic launch shipped a pitch builder agent for investment banking teams that creates target lists, runs comparables, and drafts pitchbooks. The same pattern applies, with different inputs, to the work corporate finance teams do every month producing the management pack and every quarter producing the board pack.
I have built and reviewed board packs on both sides of the table. The work has not historically been the work it should be. AI-assisted pack production has the potential to change that, if the team designs for the change rather than letting the change design itself.
This post is the practitioner read on what AI in pack production actually does, where the time savings are real, and where the failure modes are hiding.
What the work has historically been
Most management packs and most board packs are produced under time pressure by a finance team that is also closing the books, dealing with audit queries, and responding to ad-hoc requests from the executive. The pack work gets done in the last week before the meeting, often by the most senior person on the team, often at the cost of the work they should be doing. Published research puts the average cost of preparing and reading board packs at around $4 million a year per company, with the largest spending more than $10 million. Sixty-three per cent of board members and governance professionals rate their packs as “weak” or “poor,” and up to 40% of pack pages go unread. (Source: Governance & Compliance Magazine, May 2025.)
What the pack actually consists of is three layers. The numbers, which come out of the management accounts. The narrative, which has to interpret the numbers in the language of the board. The supporting analysis, which gives the variance a story and the strategic decisions a context.
The numbers are the easiest part. The narrative is the hardest, and it is where the work concentrates. The supporting analysis is the part that gets cut when time runs out.
What an AI-assisted workflow does is move some of the work from the narrative layer back into the supporting analysis layer, which is where the value lives.
What the agent actually changes
A pitch builder agent on the banking side produces a first draft of the pitchbook from the underlying data. Goldman Sachs rolled its in-house GS AI Assistant out to more than 10,000 employees in 2025 and the published claim is around a 50% reduction in pitchbook production time. (Source: Training The Street, State of AI in Finance 2025.) The same architecture, pointed at a corporate finance use case, produces a first draft of the board pack from the management accounts.
What that looks like in practice is the team receives the close numbers on, say, the 6th working day. The agent produces a draft pack on the 7th. The team’s day is no longer “produce the first version of the pack.” It is “review and substantiate the draft the agent produced.” The director who used to be doing the production work is doing the supervisory and judgment work earlier in the cycle.
The single biggest gain is in the narrative drafting. The agent can write a competent variance commentary, a competent commentary on the trading picture, a competent commentary on the cashflow story. The agent’s narrative is not the narrative that goes to the board. It is the starting point.
The second biggest gain is in the consistency. The pack the agent drafts uses the same definitions period after period. The variance the team computed differently last quarter is computed the same way this quarter. The chart that the director rebuilt three times because the colours did not match is now rebuilt to template once. The trivial corrections that used to consume the back half of the cycle stop consuming it.
The third gain is in the supporting analysis. The time the senior used to spend on the narrative is now available for the analysis. The board pack with a thoughtful analysis of the customer cohort, the contract base, or the cash position is the board pack the chair actually engages with. That is the version of the pack that earns the finance function its seat at the table.
What the agent gets wrong
The narrative the agent produces will be fluent, plausible, and sometimes misleading. The categories of error matter.
The variance commentary that explains the variance correctly without flagging the irregularity behind it. The agent will explain a 12% increase in cost of goods on price effects, when the actual driver was a one-off rebate adjustment from a supplier. The numbers are right. The story is wrong. This is the same failure mode the month-end closer agent produces on the close.
The strategic interpretation the agent does not have the context to make. The agent does not know which of the board members has been worried about working capital since the November meeting. The agent does not know that the CFO’s commentary last quarter created an expectation about Q1 the board is now testing against. The strategic context lives in the people in the room. The agent is not in the room.
The narrative that defaults to safety. The agent’s drafting style, on the high-end models, is competent and conservative. The pack that needs a sharp commentary, the pack that has to tell the board a difficult truth, the pack that has to flag the issue before it lands in the audit, is the pack the agent will under-write. The drafting that defends against the obvious challenge will be there. The drafting that pre-empts the harder challenge will not.
The team that takes the agent’s draft and lifts it into the pack is the team whose pack reads like every other AI-assisted pack. The team that uses the draft as scaffolding and writes over it with the substantive judgement is the team whose pack is worth reading.
No high-profile published enforcement case has yet named AI-assisted board pack or pitchbook drafting as the cause of a regulatory failure. The most credible cautionary precedent is in legal practice, where Mata v Avianca in 2023 became the canonical “AI hallucinated cases the lawyer did not check” sanction, and where 200-plus AI-driven legal hallucination cases have been documented in the first eight months of 2025 alone. (Source: Jones Walker on AI legal failures.) Finance has not had its Mata moment in a public ruling. The fact that one does not exist yet is not evidence the risk is theoretical.
What changes about the team
The director’s job becomes more about judgment and less about production. That is the board pack post argument extended one step. The director who can write a sharper commentary in less time is the director the board values. The director who treats the agent’s draft as the answer is the director the board stops reading.
The senior accountant’s job becomes more about the supporting analysis. The work the senior used to spend on the production layer is now available for the analysis layer. The senior who develops the analytical muscles fastest will be the most useful person in the room within twelve months. The one who treats the freed time as slack will be the most exposed.
The junior’s job is the one that changes least. The junior is still building the workings, still gathering the inputs, still preparing the supporting evidence. The agent did not eliminate that work. The agent draft is only good when the inputs the junior built are good.
What does not change
The accountability for the pack does not change. The pack the board reads is the team’s pack. The fact that the first draft was written by an agent does not transfer the accountability to the agent. The director who signs the pack is signing on the team’s behalf, not the agent’s.
The data discipline does not change. The pack the agent produces is only as good as the management accounts it is drawing from. The data quality post is the long version. The shorter version is that an AI-assisted pack on top of a manual close that does not reconcile to the ledger will be a fast, plausible, wrong pack.
The board’s expectations do not change. The board does not want the pack faster. The board wants the pack better. The version of “better” they care about is the version where the pack illuminates the business. That is judgment, not production speed.
What I would do with this
If I were running a finance function this quarter and the agent template were available, I would pilot the pack drafting workflow on the management pack first, not the board pack. The cost of a mistake in the management pack is lower, the audience is internal, and the team’s learning is faster.
I would run two cycles in shadow mode. The team produces the pack manually. The agent produces a draft in parallel. The team compares them. The director identifies the categories of error the agent introduces on this team’s work, and the team adapts.
In cycle three, I would let the agent draft the pack and have the team review and edit. I would not skip the board’s first AI-assisted pack to a stage where the director was not deeply involved in the redraft. The board’s confidence in the pack is the function’s social capital. Spend it carefully.
I would also use the freed time to invest in the supporting analysis layer that used to get cut. The first pack with a thoughtful customer cohort analysis is the pack that demonstrates the value of the new workflow to the board. That is the conversation worth having.
Where this lands
AI in pack production is one of the use cases that will move fastest, and one of the use cases where the finance function has the most agency in deciding how it lands. The agent can produce a faster version of a pack the board does not value, or a sharper version of a pack the board actively engages with. The choice is in the team’s hands.
The boards that get the most from this generation of finance AI will be the ones whose finance team treated the agent as the first draft of the work, and treated the substantive analysis as the work the team kept doing.
Maebh Collins is a Fellow Chartered Accountant (FCA, ICAEW) with Big 4 training and twenty years of operational experience as a founder and senior finance leader.