Insights · FP&A & Strategy

Board Pack Automation: Agents for the Numbers and the Narrative

Agents can assemble board and management packs from governed data and draft evidenced commentary. The CFO edits, judges and signs off instead of rebuilding slides after every late adjustment.

7 Nov 202610 min readFP&A & Strategy
Board Pack Automation: Agents for the Numbers and the Narrative

Board pack automation means using software agents to assemble the board reporting pack and management accounts from governed data, and to draft the commentary that sits alongside the numbers. Most finance teams already automate parts of the first job and none of the second, which is why the pack still consumes the week after close. This article explains how agents pull figures, tie them to the ledger, build the pages and write first-draft narrative for the CFO to edit. It also covers the controls that make the output safe to put in front of a board, and the parts of the pack that should stay manual.

1. What is board pack automation, and what does it cover?

A board pack is a fixed set of documents with a fixed audience. It usually holds an executive summary, the income statement, balance sheet and cash flow against budget and prior year, KPIs, a forecast update, risk and treasury pages, and the papers that need a decision. The management accounts pack is its monthly sibling: more detailed, and read by the executive team and business unit heads. Both rely on the same closed ledger, the same consolidation and the same definitions. Yet in many organisations they are built separately, by different people, in different workbooks.

Board pack automation covers two distinct jobs. The first is assembly: extracting balances, applying mappings and eliminations, calculating KPIs, populating tables and charts, and checking that every figure ties back to the trial balance. The second is narrative: explaining what moved, why, and what management intends to do about it. Assembly is deterministic and should be fully automated once definitions are agreed. Narrative needs judgement, so the realistic goal is a well-evidenced first draft that a finance business partner or the CFO edits. Text that goes to the board untouched is the wrong target.

2. Why board and management packs take so long to produce

The delay rarely sits in the close itself. It sits in the handoffs after it. Take a distribution group with entities in Dubai and Riyadh and a holding company in the UK. Once the ledgers are closed, an analyst exports trial balances, pastes them into a consolidation workbook, refreshes pivot tables, copies charts into slides and emails business unit heads for explanations. Each step is small. Together they create a chain where one late journal or one changed mapping forces the whole pack to be rebuilt. Nobody is then fully sure which version of the numbers the commentary was written against.

Commentary is the bigger bottleneck. Business unit heads reply late and in different styles, and they often explain the movement they expected rather than the one that happened. Finance then rewrites everything into a consistent voice the night before the deadline. The common mistake is to treat this as a formatting problem and buy a slide tool. The real issue is structural. Numbers and narrative live in separate places with no link between them, so every edit to one silently breaks the other, and the review meeting turns into a hunt for the latest file.

3. How agents assemble the numbers from governed data

Automated assembly depends on governed financial data held in one place. That means a model where chart of accounts mappings, entity hierarchies, intercompany eliminations, FX rates and KPI definitions are held once and versioned, rather than recreated in each workbook. A governed data layer such as Microsoft Fabric fits this role well. Data from SAP, Oracle, Dynamics 365 or NetSuite lands in one model, and the pack reads from that model through defined measures. If your group reports from several ERPs, this foundation matters more than any agent you put on top of it.

On top of that model, an assembly agent runs a defined sequence. It confirms the period is locked in each ledger, checks that the consolidated trial balance balances and that eliminations net to zero, refreshes the pack measures, and compares each figure with the prior version and the budget. Anything that fails a check is raised to a named owner before the pages are built. Typical examples are an entity with an unposted accrual or a KPI that moved outside its usual range. The agent does not correct the data. It stops, explains the break and waits for a person to resolve it.

The output is a set of pages generated from templates, with each number carrying a reference to the query and data version behind it. For a multi-entity group, this is what a governed financial intelligence platform looks like in practice. Board pages, management accounts and business unit packs are all drawn from the same measures, so the CEO's slide and the divisional controller's report cannot disagree. When a late adjustment arrives, the pack is regenerated from source rather than patched by hand, and the change is visible in the version history.

4. How AI agents draft board pack commentary for the CFO to edit

A commentary agent works from evidence, not general knowledge. For each line that crosses a materiality threshold, it gathers the drivers: volume and price splits, the journals that caused the movement, prior-month explanations, forecast changes and notes from business partners. It then writes a short explanation in the house style and cites the source of each claim. The technique is the same one used for drafting variance commentary at month end. Here it is applied to a wider set of pages, a more senior audience and a tighter word count.

Good AI-generated CFO commentary has a few recognisable traits. It states the movement, attributes it to a cause the data supports, separates one-off items from trend, and says what management is doing about it. Weak drafts restate the table in words or speculate about causes that are not in the data. One rule is worth enforcing from the first cycle. Any sentence the agent cannot evidence is flagged as a question for the business owner instead of being written as fact. That single control removes most of the risk of confident but wrong narrative.

The CFO's role changes rather than disappears. Instead of rewriting every page, the CFO edits a draft, sharpens the message for the board, and decides what to emphasise or hold back. That editorial judgement on tone, on what directors need to hear and on what remains uncertain should stay human. The agent's job is to make sure every draft sentence arrives with its evidence attached, so editing is quick and safe. Business partners benefit too, because they confirm or correct an explanation instead of writing one from a blank page.

5. Controls, sign-off and the audit trail behind every figure

A board pack is a governance document, so the automation behind it needs governing too. Each agent should have a defined role, explicit decision rights and an escalation path. The assembly agent may refresh and validate but not post journals. The commentary agent may draft but never publish. Version control matters as much as accuracy. Directors need to know that the figures they approve are the figures that were reviewed, and auditors will want to see how a number in the pack traces back to the ledger and who changed what along the way.

In practice this means a sign-off workflow. The controller certifies the numbers, business owners confirm their explanations, and the CFO approves the final narrative, with every change logged. This is how AITHENTIC builds its domain-trained finance AI agents: each one carries a full audit trail, and people approve the exceptions. Whatever platform you choose, ask to see the trail for a single figure, from the board slide back to the source transaction, before you trust it with a live pack. If the vendor cannot show it, the governance is not there yet.

6. Where to start with management reporting automation, and what to leave manual

Start with the management accounts pack rather than the board pack. It is produced monthly, so the team gets twelve cycles a year to refine templates and thresholds, and the cost of an awkward draft is lower. Agree KPI definitions and materiality thresholds first. Most stalled projects stall on a disagreement about what 'gross margin' means in each entity, not on technology. Then run the agents in parallel with the existing process for a few cycles and compare the outputs line by line before anyone relies on them.

Some parts of the pack are not worth automating. Strategy papers, acquisition updates, sensitive people matters and forward-looking guidance for listed companies should be written by people, though agents can still supply the supporting figures. Nor is it worth automating a pack whose structure changes every quarter at the chair's request; stabilise the template first. Where the conditions are right, board pack automation turns the week after close into a review exercise. Finance time moves to judgement and conversations with the business, and away from reconciling slides against spreadsheets.

Key challenges

  • Numbers and commentary are produced in separate files with no link, so late adjustments silently break the narrative.
  • KPI definitions and account mappings differ between entities and between the board and management packs.
  • Business unit explanations arrive late, in inconsistent styles and often unsupported by the data.
  • Weak version control makes it unclear which figures directors actually reviewed and approved.
  • Manual copying from consolidation workbooks into slides introduces errors that are hard to trace.

The framework

A workable operating model has three layers. The data layer holds the closed ledgers, mappings, eliminations and KPI definitions in one governed model. The agent layer runs assembly, validation and commentary drafting with defined decision rights, raising breaks to named owners instead of fixing them. The human layer certifies, edits and approves: the controller signs off the numbers, business owners confirm explanations, and the CFO owns the final narrative. Each layer hands over through a recorded checkpoint, so the pack that reaches the board can be traced from slide to source transaction and every edit has a named author.

How to implement it

  1. Inventory every page in the current board and management packs, with its data source, owner and definitions.
  2. Agree KPI definitions, mappings and materiality thresholds and hold them once in the governed data model.
  3. Build pack templates that read directly from governed measures, with each figure referenced to its data version.
  4. Configure an assembly agent to validate period locks, balances and eliminations before any page is generated.
  5. Add a commentary agent that drafts evidenced explanations above threshold and flags unsupported claims as questions.
  6. Run in parallel for several cycles, then move to a sign-off workflow with logged controller, owner and CFO approvals.

Frequently asked questions

What is board pack automation?

Board pack automation is the use of software agents to build board and management reporting packs from governed financial data, including tables, charts and KPIs, and to draft the commentary that explains them. The numbers are generated and validated automatically against the ledger. The narrative is produced as a first draft with evidence attached, which finance business partners and the CFO edit and approve before the pack is issued.

Can AI write board report commentary?

Yes, AI can write a reliable first draft of board report commentary, provided it works from the underlying data rather than general knowledge. A well-designed agent explains movements above a materiality threshold, cites the journals and drivers behind each claim, and flags anything it cannot support. The CFO should still edit the draft, because tone, emphasis and what the board needs to hear are matters of judgement.

How long does it take to automate a management accounts pack?

It depends mainly on data readiness rather than technology. If ledgers close cleanly and KPI definitions are agreed, an automated management accounts pack can be running in parallel within a few close cycles. Groups with several ERPs, inconsistent mappings or undocumented spreadsheet logic should expect to spend most of the early effort on the data model and definitions before agents add much value.

Is board pack automation safe for audit and governance?

Board pack automation is safe when each agent has defined decision rights, no ability to post journals or publish, and a full audit trail. Every figure should trace from the board slide back to the source transaction, and every edit to the commentary should be logged with its author. Controllers certify the numbers and the CFO approves the narrative, so accountability for what directors see stays with people.

Which tools are used to automate board reporting?

Most automated board reporting combines a governed data layer, a reporting tool and agents. A typical stack takes ledger data from SAP, Oracle, Dynamics 365 or NetSuite into a model such as Microsoft Fabric, generates pages through Power BI or templated documents, and uses agents to validate figures and draft commentary. The specific tools matter less than holding definitions once and linking each number to its source.

Want these insights applied to your finance function?

Book a working session or take the 5-minute assessment to see where value is leaking.