---
"title": "How do I roll out AI across an FP&A team?"
"description": "A step-by-step playbook for rolling out AI across a financial
  planning and analysis team, from choosing first workflows and setting data
  rules to training, controls, measurement and deciding what to automate."
"section": "Playbooks"
"sidebarTitle": "Roll out AI across FP&A"
"published": "2026-11-10T14:22:13.341Z"
"updated": "2026-11-10T14:22:13.341Z"
---

## How do I roll out AI across an FP&A team?

It depends on where your FP&A team loses time. If most of the cycle goes into writing and explaining, such as variance commentary, forecast narratives and board or management packs, a phased rollout of approved AI tools plus hands-on training usually shows results within one or two reporting cycles. If the time goes into pulling and reconciling data from several systems, start by fixing that flow, because AI on top of messy inputs mostly produces faster mistakes. Large teams should pilot with one sub-team first. Small teams can usually go together.

## Who this is for

CFOs, FP&A leads and finance operations managers who want their planning and analysis team using AI every month, not just experimenting, while keeping the numbers reviewable.

## The short version

- Map the planning and reporting cycle and pick three to five tasks to start with.
- Agree on approved tools and data rules before anyone uploads a forecast.
- Train on last cycle's real work, then use AI on the current cycle.
- Build review steps into the workflow so every number is traced to a source.
- Measure time and rework per task, and decide what to standardize or automate after two cycles.

## Step 1: Map the cycle

Write out one full planning or reporting cycle, task by task, with an owner, the inputs and roughly how long each step takes. For most FP&A teams the list includes:

- Pulling actuals and other inputs from the general ledger, payroll and operational systems
- Building or refreshing variance analysis against budget and forecast
- Writing variance commentary and explanations for leaders
- Updating forecasts and running scenarios
- Assembling the management or board pack
- Answering ad hoc questions from leaders between cycles

Mark which tasks are mostly writing and explaining, which are mostly calculation, and which are mostly moving data. Our guide on [finding which workflows to automate with AI](/docs/find-workflows-to-automate-with-ai) has a simple way to score them.

## Step 2: Choose where to start

Good first tasks are frequent, reading or writing heavy and easy to check: variance commentary drafts, first-pass summaries of department budget submissions, turning a forecast change into a plain-English explanation, and preparing questions for budget owners. Avoid starting with anything where AI does the arithmetic that goes straight into a report. Calculations should stay in your spreadsheet or planning system, where they can be audited.

## Step 3: Set tools and data rules

FP&A works with forecasts, compensation data and figures that may be material and non-public. Before rollout, decide:

- Which AI tools are approved, on business plans that do not train on your data
- Which data may go in, and which stays out, such as individual salaries
- Who reviews AI-assisted output before it reaches leaders, the board or lenders

If you have no policy yet, see [how to write an AI acceptable use policy](/docs/write-an-ai-acceptable-use-policy) and [how to keep client data safe with AI](/docs/keeping-client-data-safe-with-ai).

## Step 4: Train on last cycle's work

Run two short, hands-on sessions using real material from the most recent close or forecast. Have analysts draft last month's variance commentary with AI and compare it to what was actually sent. Have them ask an AI tool to explain a forecast movement and then check every figure it cites. This teaches the strengths and the failure points at the same time. Our page on [AI training for finance teams](/docs/ai-training-for-finance-teams) covers the content in more depth.

## Step 5: Use it on the live cycle

In the next cycle, each analyst uses AI on the chosen tasks, with a named reviewer. Hold a 30-minute check-in each week to share what worked, add good prompts to a shared library and drop what did not help. A team prompt library and simple standards keep quality consistent. See [creating prompting standards](/docs/create-prompting-standards).

## Step 6: Build in controls

Keep controls simple and visible:

- Every figure in commentary is traced to the model or ledger, not taken from the AI tool's text.
- The reviewer signs off as they do today. AI does not change who is accountable.
- Keep the prompt and source files used for anything that goes to the board, so the work can be reconstructed.
- Spot-check a sample of AI-assisted commentary each cycle for errors.

## Step 7: Measure and decide

Before starting, record how long each chosen task takes and how many review rounds it needs. After two cycles, compare. Expect gains to be uneven: commentary and summaries usually improve first, while tasks that depend on messy inputs may not move. Then decide, task by task, whether to standardize it, automate it with a connected workflow, or drop it. Our guide on [measuring whether AI training worked](/docs/measure-ai-training-results) explains what to track.

## When to move from training to engineering

If analysts still spend hours each cycle exporting, cleaning and matching data, that is a sign to build something connected rather than train harder. A custom workflow that pulls actuals, flags variances above a threshold and drafts commentary for review can remove that work, provided a person still approves the output. See [AI agents for month-end close](/docs/ai-agents-for-month-end-close) and [custom AI vs off-the-shelf tools](/docs/custom-ai-vs-off-the-shelf-tools).

## Common mistakes

- **Letting AI do the arithmetic.** Calculations belong in auditable models.
- **Rolling out in the middle of a busy close** without time to learn.
- **No baseline,** which makes it impossible to show the result to the CFO.
- **Skipping the reviewer** because the draft looks polished.
- **One training session and done.** Habits fade without weekly follow-up. See [why AI adoption stalls](/docs/why-ai-adoption-stalls).

## How Ruby helps

- **AI training:** hands-on sessions built on your own reporting cycle, with follow-up during the live cycle.
- **AI strategy and consulting:** mapping the cycle, choosing first tasks, setting data rules and measuring results.
- **AI engineering:** building connected workflows for data pulls, variance flags and draft commentary where training alone is not enough.

If you want a rollout plan built around your FP&A cycle, [book a call with Ruby](https://rubyadvisory.com/book-a-call).

## Related pages

- [How should a finance team learn to use AI?](/docs/ai-training-for-finance-teams)
- [How do I run an AI pilot that doesn't stall?](/docs/run-an-ai-pilot-that-does-not-stall)
- [How to roll out AI across fund operations](/docs/roll-out-ai-across-fund-operations)
- [What does an AI implementation roadmap look like?](/docs/ai-implementation-roadmap)
