AI for Accountants: Where to Start

Accounting practices have more AI opportunity than almost any service business, and more reason to be careful about where they use it. The work is structured, repeatable, and document-heavy, which is exactly what AI handles well. It also involves client data and regulatory exposure, which means some things stay human.

Here's where to start and what to skip.

Where AI actually helps

Client onboarding and intake

New client onboarding involves the same information gathered in a slightly different order every time. AI can draft the intake questionnaire, summarize what came back, flag missing information, and produce a first-pass client summary before the first meeting. This alone saves most practices an hour or two per new client.

Meeting prep and follow-up

Before a client meeting: AI can pull together a summary of open items, recent correspondence, and anything flagged since the last meeting. After: AI can turn meeting notes into a follow-up email with action items. Most accountants write these from scratch every time. They don't have to.

Proposal and engagement letter drafts

The structure of a proposal doesn't change much from client to client. The scope, the fee, the terms. AI produces a solid first draft from a brief intake. You edit it. The blank page problem disappears.

Internal documentation

Process documentation that nobody ever writes because there's no time to write it. AI can draft it from a voice note or a rough outline. Once it exists, training new staff gets faster.

Client communication

Routine emails: responses to common questions, deadline reminders, document request follow-ups. These follow patterns. AI handles the draft. You review and send.

Where to be careful

Any client-facing output that doesn't get reviewed

AI drafts. You review. That's the workflow. An AI-drafted email that goes out without review is a liability, not a time saver.

Tax advice and technical guidance

AI gets things wrong in ways that look right. Any technical output that goes to a client needs a human sign-off. This isn't AI being bad at accounting. It's AI being bad at knowing what it doesn't know.

Anything involving confidential client data in a tool your firm hasn't vetted

Check with your practice management software and your professional liability coverage before feeding client data into any AI tool. Most of the major tools have enterprise agreements that address this. Most smaller practices haven't checked.

Where to start

Pick one of these and build it properly before adding anything else:

Meeting follow-up emails. Low risk, high frequency, immediate time savings. Build a prompt that takes rough notes and produces a clean follow-up. Run it for a month. See what it saves.

Once that's running, the next step is usually intake summaries or proposal drafts.

Want help finding what fits your practice?

The AI Workflow Audit is built for exactly this. We look at how your practice actually runs, find the highest-leverage places, and give you a plan you can act on.