Training a CFO on AI Tools: The 90-Minute Starter

When a founder of a 140-person services firm told me her CFO was "too senior for the AI thing," I asked what he did Mondays. Three hours of variance commentary. We started there. Ninety minutes later he had a first draft generator running.

Why CFOs are a special training case

Your CFO is paid to be skeptical. That's a feature, not a bug. The mistake I see SMB owners make is sending the CFO to a generic "AI for executives" workshop, where the examples are marketing copy and customer chatbots. None of that maps to what a CFO actually owns.

A CFO's day is variance commentary, AR review, vendor evaluations, board memos, audit prep, covenant tracking, and the occasional fire-drill (failed payment, surprise tax notice, missed forecast). Every one of those is a high-context, high-stakes text task — which is exactly where modern LLMs help most when used correctly.

Definition: Augment, don't replace — the AI drafts; the human signs. The CFO never publishes raw model output to the board.

The 90-minute starter below assumes nothing technical and produces one shippable artefact by minute 90. It is not theory.

What does a 90-minute CFO starter look like?

The session has three blocks of 30 minutes each. You sit shoulder-to-shoulder with the CFO; one of you drives the keyboard, the other narrates.

Block 1 (0-30 min): the variance commentary draft

Pull last month's actuals vs budget at the GL-account level. Paste the table into the chat tool with this prompt:

You are an FP&A analyst writing variance commentary for a [INDUSTRY]
company with [X] FTEs and ~$[Y]M revenue. Below is actuals vs budget
for [MONTH]. For each line >5% variance, write 1-2 sentences in the
voice of a CFO memo: state the variance, the most likely 2 drivers
based on the data, and what we should ask the owner of that GL.
Do NOT fabricate causes that aren't in the data. If a driver is
uncertain, write "needs owner input" and propose the question.
Output as a markdown table: GL | Variance % | Commentary | Question.

Two things happen here. First, the CFO sees the model produce 80% of a memo he already writes — which collapses skepticism faster than any pitch deck. Second, the "needs owner input" guardrail teaches the model is a draft tool, not an oracle.

Tool tip (Course for Business): Our 6-week program runs CFOs and their FP&A teams through this exact exercise on day 1, then layers in close acceleration, AR triage, and board-pack drafts. The framing is Augment, don't replace — the CFO never ships raw output, but starts every artefact from a 70%-done draft. Pairing one AI Champion (1:15-20) inside finance is what makes adoption stick after the workshop. → https://course.aiadvisoryboard.me/business

Block 2 (30-60 min): AR aging triage

Export AR aging. Have the CFO ask the model to cluster customers by behavior — chronic late payers, recently-slipped, healthy — and draft a tiered email sequence. The CFO immediately sees the value isn't in writing each email; it's in the segmentation logic he'd otherwise eyeball.

A $1B logistics company famously took its invoice-processing function from 7 to 2 FTEs after a similar redesign — that's the magnitude of upside, but only after the CFO understands the workflow personally. (See disclosure note.)

Block 3 (60-90 min): vendor evaluation

Pick a real RFP response sitting in the CFO's inbox. Ask the model to summarize, flag risks, propose a comparison rubric, and draft questions for the vendor. By minute 90, the CFO has used the same tool three different ways on three real artefacts.

How do you keep momentum after the 90 minutes?

Three rules:

1. The CFO does not learn alone. Pair them with one FP&A analyst as the finance team's AI Champion, ratio 1:15-20 inside the function.
2. Pick one weekly artefact (variance, board memo, cash forecast) and commit to drafting it AI-first for 6 weeks.
3. Schedule a 30-minute Shoulder-to-Shoulder session each week where the CFO and Champion review what worked, what didn't, what shipped.

Definition: Shoulder-to-Shoulder — a working session where two people share one screen, one keyboard, and one real artefact. Not a demo, not a webinar.

BCG's 2025 AI Radar found ~78% of orgs have deployed AI but only ~25% see meaningful value. The gap is almost entirely on the people-and-process side (BCG's 10-20-70 framing). For a CFO, "people-and-process" means you, them, and one Champion in a room every week for six weeks.

Team scan (what AI champions report after week 1)

A typical 30-500-employee finance team after week 1:

• Adoption: CFO + 1 FP&A Champion using the tool daily; 2-3 others experimenting
• Use case #1: variance commentary draft — saves ~60-90 min/month
• Use case #2: AR follow-up email tiering — saves ~2 hours/week
• Use case #3: vendor pack summarization — 30 min instead of 2 hours
• Use case #4: board memo bullet draft — 45 min instead of 3 hours
• Use case #5: policy / contract clause comparison
• Friction: "the model is confidently wrong on numbers" — fixed by pasting tables, not retyping
• Risk flag: confidential data uploads — 46% of employees have done this on public tools; finance must use approved-tier tools only
• Saved time: typically 4-6 hours/week per CFO once Champion is up
• Honest miss: close acceleration takes 6-8 weeks, not week 1

Tool tip (Course for Business): Most CFOs we train inside the 6-week program end week 1 with one shippable artefact and end week 6 with a full set: variance memo, board pack, AR sequence, vendor rubric, FP&A model commentary. The CFO becomes the audit-trail owner; the FP&A Champion becomes the workflow owner. Augment, don't replace is the default posture. → https://course.aiadvisoryboard.me/business

Micro-case (what changes after 7-14 days)

A typical 200-FTE professional-services firm trains its CFO and senior FP&A analyst together on day 1. By day 7, the variance commentary that used to consume 3 hours every Monday is a 45-minute review of an AI-generated draft. By day 14, the AR aging triage runs weekly instead of monthly, and DSO drops by 4-6 days because the chase emails actually go out. The CFO stops being the bottleneck on the board pack — drafts arrive Friday, signed Monday. Total CFO time saved: roughly 5 hours/week. Total FP&A time saved: roughly 8 hours/week. Most of that gets reinvested in scenario modeling that nobody had time for before.

Note on this case: This example is illustrative — based on typical patterns we observe with companies of 30-500 employees, not a single named client. Specific numbers are rounded approximations of common ranges, not guarantees.

FAQ

Should the CFO learn to "prompt engineer"? No. The CFO learns to delegate to the model the way they'd delegate to a junior analyst — with context, constraints, and a clear deliverable format. That's it. The deeper craft sits with the FP&A Champion.

What about confidential financial data? Use enterprise-tier tools with no-training data settings, or self-hosted options. 46% of employees have uploaded confidential data to public AI tools — your CFO is the right person to set the rule, but only after they've used the tools themselves and understand what's actually risky.

Will this replace headcount? Not in finance, not yet. The realistic pattern is the same FTEs doing more (scenario work, faster close, better vendor diligence). The $1B logistics 7→2 FTE example is invoice-processing — a specific, repetitive workflow, not the CFO's job.

How is this different from sending the CFO to a generic AI exec course? Generic courses optimize for awareness. This optimizes for one shippable artefact in 90 minutes. BCG found programs under ~5 hours produce no behavior change — but a 5-hour program built around the CFO's actual artefacts beats a 40-hour generic one.

What if the CFO refuses? Then start with the Controller or Head of FP&A and let results speak. CFOs who watch their own team ship in week 1 typically join in week 2-3. Forcing it doesn't work.

The takeaway

CFOs are not hard to train — they're badly trained. Ninety minutes against a real variance file beats a week of generic content. Pair the CFO with one Champion, run six Shoulder-to-Shoulder sessions, and finance becomes the most AI-mature function in the company faster than marketing. The hard part isn't the tool; it's choosing the artefact and showing up for six weeks.

Next step: pick the one artefact your CFO produces every week that they hate the most. That's the starter file.

If you want every employee — including your CFO — to ship their first AI automation in five days, book a 30-min call and we'll map your finance team's first week. → https://course.aiadvisoryboard.me/business