Accounting is a good fit for AI in some ways and a bad fit in others. The work is document-heavy and repetitive, which helps. The stakes for accuracy and client confidentiality are high, which means mistakes cost more than they would in most businesses. A good implementation respects both.
The adoption gap inside firms
Recent surveys paint a picture of interest running ahead of use. A CPA Practice Advisor article summarizing 2026 research reports that 85% of accounting teams call AI a strategic priority, while only 10% use it extensively (FloQast), and that only 18% of professionals say their organizations track AI return on investment (Thomson Reuters).[1] It also cites a CPA.com and Blue J survey of more than 1,000 tax professionals in which 60% use AI-powered tax research at least weekly, up from 33% the year before.[1] These are secondhand figures from different surveys with different samples, so compare them cautiously. The direction is consistent: firms are experimenting widely and measuring little.
The author argues that the barriers are mostly behavioral: staff worry about job security, feel AI cheapens professional judgment, are unsure what use is permitted and are rarely measured on real changes in their work.[1] That matches what we see in small firms. The tool is usually fine. The rollout is the problem.
Workflows that fit
Document chasing
Every busy season, staff spend hours asking clients for the same missing items. An automated workflow can track which documents each client owes, send reminders by email or text on a schedule and flag clients who have gone quiet. Gate: a staff accountant approves any message that mentions deadlines, penalties or extensions. Measure: days from engagement start to complete document set; staff hours spent chasing.
Intake and organization
AI can classify incoming PDFs and photos (W-2s, 1099s, bank statements, receipts), extract the key fields and file them in the right client folder. Gate: spot-check extracted amounts against the source, especially for tax documents. Measure: minutes per client to reach a "ready for prep" state.
Workpaper and review prep
Tie-outs, variance flags and first-pass reconciliations are rule-driven work. AI can assemble comparisons and highlight unusual items for the reviewer. Gate: the preparer and reviewer remain responsible for the conclusions. Measure: review notes per file; time from prep to review.
Client communication drafts
Draft status updates, engagement letters from your templates and explanations of routine notices. Gate: every client-facing message is reviewed before it goes out. Measure: time to respond to client questions.
Research
AI-assisted research is widely used. It also needs the most skepticism: confirm any citation, ruling or code section against the primary source before you rely on it. A plausible-looking wrong cite is worse than no answer.
Confidentiality comes first
Client financial data is among the most sensitive information a small business holds. IBM's 2025 breach report found that a high level of unmanaged "shadow AI" added about $670,000 to the average breach cost.[2] That is a global average, not a figure for accounting firms. The practical point is that staff pasting client data into personal AI accounts is a real exposure. Before any tool touches client information:
- Check whether the vendor trains on your inputs and whether you can opt out in writing.
- Understand retention and deletion terms.
- Review the rules that apply to tax return information. Preparers face confidentiality and consent requirements (including those under Internal Revenue Code Section 7216), so talk to your professional liability carrier and counsel about how a given tool handles client data.
- Update engagement letters if your use of AI changes how client data is handled.
- Adopt a written AI policy covering approved tools, data categories and review. See our guide to a one-page policy.
A sensible rollout plan
- Pick one workflow with clear rules and low risk, such as document chasing.
- Measure the baseline for a busy-season month.
- Run it with two or three volunteers, not the whole firm.
- Set the gate and name who approves what.
- Review results after 30 and 90 days, and decide whether to expand.
- Talk openly about staffing. If staff think efficiency means layoffs, they will resist or hide their use. Say what you intend, and mean it.
Firms that treat AI as a way to take drudgery off their people, and then measure whether it did, tend to get farther than firms that buy a tool and wait.
Sources
- CPA Practice Advisor, "Why CPA Firms Need a Behavioral AI Adoption Playbook" (Sept. 2026)
- IBM, "2025 Cost of a Data Breach Report: Navigating the AI rush"
Figures are quoted from the sources above and were reviewed in October 2026. Where a source is a secondary summary of other research, we say so in the text. Survey results reflect each survey's own sample and definitions. Examples labeled as illustrations use made-up numbers to show the method. Nothing here is a forecast or a guarantee of results.
Want this worked out for your business?
Tell us what is wasting your time.