A Stanford and MIT study published in August 2025 tracked 79 small and mid-size accounting firms after AI adoption. The firms that used AI well reallocated 8.5% of their time -- roughly 3.5 hours per week -- from data entry to client-facing work. Client support interactions jumped 55%.

But here's the part most people skip over: senior accountants got significantly larger gains than juniors. The experienced professionals treated AI as a collaborator and caught its mistakes. Junior staff accepted AI output at face value and saw smaller improvements.

That gap is the entire story of AI in accounting right now. The tool is fast. Whether the output is good depends entirely on how you prompt it and what you do with the result.

The Advisory Letter Problem

Client advisory letters eat time because they require two things that don't naturally go together: precision about the numbers and clarity about what the numbers mean.

Most accountants can pull the data in minutes. The writing takes an hour or more. You're translating ratio analysis, variance explanations, and trend observations into language a business owner actually understands and acts on.

This is exactly where prompting shines. Not because AI writes better than a CPA -- it doesn't. Because AI produces a structured first draft that a CPA can edit in ten minutes instead of writing from scratch in sixty.

Prompt Structure That Works for Advisory Letters

The mistake most accountants make is dumping raw financials into a chat window and asking for "an advisory letter." That produces generic output full of hedging language that sounds like it was written by nobody for nobody.

Good advisory prompts have three components: the financial context, the client context, and the output constraints.

The Prompt:

You are a CPA preparing a quarterly advisory letter for a client. Here is the context:

CLIENT PROFILE:
- Industry: Regional restaurant group (4 locations)
- Annual revenue: $3.2M
- Owner's priority this quarter: deciding whether to open a 5th location

FINANCIAL DATA (Q4 2025 vs Q3 2025):
- Revenue: $840K → $790K (-5.9%)
- COGS: 31.2% → 34.8% of revenue
- Labor: 28.1% → 30.5% of revenue
- Net margin: 8.4% → 4.2%
- Cash reserves: $180K → $142K
- Accounts payable aging: 22 days → 38 days

TASK:
Write a 400-word advisory letter that:
1. Opens with the one thing the owner needs to hear first
2. Explains the margin compression in plain language -- no jargon
3. Connects the financial trends directly to the expansion decision
4. Recommends 2-3 specific actions before any expansion discussion
5. Closes with what you'd want to review together in a meeting

Tone: Direct but not alarming. This owner respects straight talk.
Do not use filler phrases like "as you know" or "it's worth noting."

Why This Works: The prompt gives the AI a specific client with a specific decision to make. The financial data is structured so the model can calculate relationships, not just parrot numbers. The output constraints prevent the generic advisory-speak that makes these letters useless. And the tone instruction keeps it from defaulting to corporate hedging.

Expected Output:

Hi Maria,

Your margins need attention before we talk expansion. Here's what Q4 is telling us.

Revenue dipped about 6% from Q3, which isn't unusual for seasonal patterns. The bigger concern is where your costs went. Food costs jumped from 31% to nearly 35% of revenue -- that's an extra $28K per quarter if it holds. Labor climbed to 30.5%. Together, those two shifts cut your net margin in half, from 8.4% down to 4.2%.

In dollar terms, you kept $66K on $840K in Q3. This quarter you kept $33K on $790K. Your cash reserves dropped $38K, and your payables are taking almost twice as long to clear. That's not a crisis, but it's a trend that gets worse fast with a fifth location adding fixed costs.

Before we discuss the new location, I'd recommend three things:

First, audit your supplier contracts across all four locations. A 3.6-point COGS swing usually means at least one vendor relationship needs renegotiation or replacement.

Second, review staffing schedules against actual covers per shift. A 2.4-point labor increase across four locations suggests scheduling drift, not a wage problem.

Third, hold expansion conversations until we see Q1 numbers with these adjustments in place. Opening a fifth location on 4.2% margins means you're funding the buildout from an already-thinning cash position.

I'd like to walk through the location-by-location breakdown when we meet. Tuesday or Thursday work for you?

That's a draft an accountant can review, adjust for things the AI doesn't know (maybe Maria already renegotiated supplier contracts last week), and send. Ten minutes of editing versus an hour of writing.

Where Accountants Go Wrong With These Prompts

Skipping the client context. Without the industry, the owner's priorities, and the relationship tone, the AI writes a template. Templates don't build advisory relationships.

Feeding unstructured data. Pasting a full trial balance and asking for insights produces vague observations. Pre-selecting the relevant metrics and trends gives the AI something to work with.

Not specifying what to exclude. AI defaults to hedge phrases, disclaimers, and filler. Tell it what NOT to write and the output gets sharper.

The Judgment Layer You Can't Automate

Karbon's 2026 report found that 98% of accounting firms now use AI in some capacity, and 59% of professionals use it specifically for composing emails and client communications. The tools are everywhere. The differentiation is in what happens between the AI draft and the send button.

The Stanford study found that senior accountants got bigger gains because they knew what to correct. They caught when the AI overstated a trend, missed a contextual factor, or recommended something that didn't fit the client's situation.

This is exactly the risk the Morgan & Morgan case illustrates from the legal world. In February 2025, three attorneys at a major law firm were sanctioned for filing motions containing eight fabricated case citations generated by their in-house AI tool. They used prompts like "add Federal Case law from Wyoming" and filed without reading the output. Total fines: $5,000 and a professional credibility hit that costs far more.

The accounting version of this isn't fabricated case law -- it's a confidently wrong variance explanation, a misattributed trend, or an advisory recommendation that ignores something the AI couldn't know. The AICPA hasn't sanctioned anyone for AI-generated advisory letters yet. But the reputational cost of sending a client a letter with a wrong recommendation is real today.

Making This Part of Your Workflow

The accountants getting real results aren't using AI as a magic button. They're building prompt templates for their recurring deliverables -- quarterly advisory letters, management letters, year-end summaries -- with placeholders for the financial data and client context that change each period.

One practical approach: build a prompt library with 3-5 templates covering your most common advisory scenarios. Each template should have slots for the client profile, the relevant financial data, and the specific decision or situation driving the letter. Reuse the structure, swap the specifics.

The firms reporting the biggest gains -- the ones in the Stanford study reallocating 3.5 hours per week to client work -- aren't prompting differently every time. They've standardized the prompting and spend their time on the judgment layer: reviewing, adjusting, and adding the context that only a human advisor has.

Want hands-on prompt engineering training for your accounting team? Connect with Kief Studio on Discord or schedule a session. We build prompt libraries for professional services firms that turn repetitive writing into repeatable workflows.

client-communication

How Accountants Are Using AI to Draft Client Advisory Letters in Half the Time