What Accountants and Bookkeepers Discover When They First Use AI for Budgeting and Financial Planning

The first time most practitioners use AI on a budgeting engagement, they reach for it at the wrong stage. They build the numbers, then ask AI to “write the commentary.” The output is fine, but it’s not the most valuable thing AI can do here.

The genuinely useful pattern is to use AI three times across the engagement: once to prepare the planning meeting, once to interrogate the assumptions, and once to write the deliverable. Used that way, the budgeting workflow stops being a four-day process and becomes a two-day process where the conversations with the client are the slowest part, which is exactly how it should be.

Where AI saves time isn’t where you’d guess

Budgeting in a small firm has three time sinks. First, the planning meeting (preparing the agenda, listing the questions, anticipating where the client will struggle to give specifics). Second, building the model itself in the spreadsheet or the dedicated tool. Third, writing the commentary and the client-facing memo.

The middle one, building the model, is where most practitioners assume AI helps most. It doesn’t, much. Spotlight Reporting, Fathom, or a well-built Excel template handle the actual budget mechanics faster and more accurately than asking ChatGPT or Claude to project lines from a CSV. AI is genuinely good at the bookends, the planning and the writing.

The planning meeting prompt that doubles the value of one client hour

A bad budget meeting is one where the accountant asks “any changes for next year?” and the client says “I don’t think so.” A good one is structured to draw out the assumptions the client doesn’t know they’re making. AI can build that structure for you in under two minutes:

“I am preparing for an annual budget planning meeting with [CLIENT NAME], a [INDUSTRY] business in [LOCATION], with annual revenue around [REVENUE], [STAFF COUNT] staff, and [BUSINESS MODEL DETAILS such as service mix, channels, seasonality]. Last year’s actuals showed [TOP-LINE FIGURES, ANY UNUSUAL ITEMS]. Write a 50 minute meeting agenda with: (1) two opening questions to understand how the owner is feeling about the year ahead and any non-financial pressures shaping their plans, (2) ten specific assumption-gathering questions covering revenue drivers, pricing, staffing, capex, supplier costs, overheads, and any sector-specific risks for a [INDUSTRY] business, (3) two questions designed to reveal scenarios the client is privately worried about but hasn’t said out loud, (4) a final ‘what would success look like’ question. Each question should be open-ended and conversational. After the agenda, list the documents I should ask the client to bring to the meeting.”

That prompt produces something better than what most of us write from scratch under time pressure. It also catches the questions experienced practitioners stop asking after a few years of routine, because the prompt doesn’t know what’s “obvious.”

The assumptions register: the unsung deliverable

Here’s the discovery practitioners make on their second or third AI-assisted budget. The single most useful artefact in the whole engagement isn’t the budget. It’s the assumptions register, the document that says “we’re projecting 12% revenue growth because the client is launching service line X in Q2 with a hiring plan of Y.”

When margins move six months later, the assumptions register tells you exactly which assumption broke. Without it, you’re guessing. Most practices don’t produce one because writing it from notes is tedious. AI removes that excuse:

“Below are my notes from the budget planning meeting with [CLIENT NAME] [PASTE NOTES]. Convert these into a structured assumptions register with the following columns: assumption (one sentence), category (revenue / pricing / staffing / cost / capex / risk), source (client statement, market data, our judgment, etc.), confidence (high / medium / low), and ‘how we’ll know it’s wrong’ (the specific signal that would tell us this assumption needs revisiting). Output as a clean table. After the table, add a separate ‘open questions’ list of anything I asked but didn’t get a clear answer on, that we need to chase up before finalising.”

That table goes straight into the engagement file. It’s the thing that makes the budget defensible if the client ever asks “why did you assume that?” six months later, and it’s the thing that makes mid-year reviews fast because you already know which assumptions to test.

The commentary prompt that earns its keep

Once the model is built, the deliverable the client actually reads is the commentary memo. Two paragraphs of clear, plain English that explain what the budget says and what it implies. Here’s the prompt:

“Write a 350 word budget commentary for [CLIENT NAME], a [INDUSTRY] business with current revenue of [CURRENT REVENUE]. Inputs: [PASTE the budget summary table, the assumptions register, and the prior-year actuals]. Structure: (1) one-sentence headline of what next year is expected to look like, (2) a ‘what’s driving the numbers’ paragraph naming the two or three biggest movers, (3) a ‘what we’re watching’ paragraph naming two specific risks and the indicators we’ll watch for, (4) a closing paragraph framing the year ahead realistically (not falsely optimistic, not gloomy). Tone: warm, advisory, confident. Audience: a small business owner with no accounting background. No corporate buzzwords. Banned phrases: ‘in today’s economic climate’, ‘going forward’. If any number in the inputs looks inconsistent, flag it rather than smoothing over it.”

The “flag rather than smooth over” instruction matters. It catches typos and the occasional model error in your inputs, and it produces a memo you can put your name on without an embarrassing edit four hours later.

The failure mode: outsourcing the judgment, not the typing

The mistake practitioners regret most is using AI to make the call, rather than to draft the explanation of the call. If you ask AI “what should the revenue assumption be for [CLIENT]?” you’ll get a confident answer that’s averaged from the entire internet’s worth of small-business writing. That answer has nothing to do with this client.

The discipline that matters is: humans set the assumptions, AI writes them up. Humans interpret the variances, AI drafts the commentary. Humans decide what to flag for the client meeting, AI structures the agenda. Reverse that flow and you’ll produce technically polished documents that your judgment hasn’t actually touched, which is the worst of both worlds.

The counterintuitive part: budget commentary is the cheapest skill to upgrade in your firm

The AICPA and CPA.com 2025 AI in Accounting Report found 85% of accountants are excited or intrigued by AI, but only 37% of firms invest in training. The training payoff is concentrated in places like this one. Budgeting work has historically been senior-only because the writing required experience. With a tight prompt template and a one-hour team training, a competent associate can produce a draft commentary that a senior reviews in 15 minutes instead of writing in 90.

That single workflow shift, replacing two hours of senior writing with 15 minutes of senior reviewing, recovers more partner time than any other AI use case in a typical small firm. It’s not glamorous. It’s not the AI demo that goes viral. It’s the boring win that pays for itself in the first quarter.