What to Keep Human
Tax judgments, planning recommendations, and anything that requires professional accountability stay with the licensed CPA. AI drafts; the professional signs. Client relationship management, particularly for high-net-worth or complex business clients, requires the trust of a human advisor, not an automated system. A $45,000 tax-and-advisory retainer client is paying for judgment, not throughput.
Anything that gets signed or stamped with a CPA's credentials must have that CPA's genuine review. Circular 230 does not recognize "the model wrote it" as a defense. Neither does state board discipline, professional liability insurance, or a client who receives an IRS notice tied to an AI-drafted position nobody actually checked.
Audit and attest engagements deserve particular care. PCAOB and AICPA auditing standards require documented evidence of professional judgment. AI can accelerate evidence gathering and workpaper preparation, but the judgment documentation remains the auditor's responsibility. Firms using AI in attest work should have written policies covering where AI is used, how outputs are reviewed, and how independence is maintained.
What the ROI Looks Like
Firms that implement AI document intake and communication workflows typically see administrative staff capacity increase by 25 to 40 percent during peak season. Engagement letter production time drops from two to three hours per week to under 30 minutes. Clients report faster response times and fewer "where are we" calls because status communication is more consistent.
Implementation costs for a focused rollout (document intake plus client communication automation) typically run $18,000 to $45,000 for a small-to-midsize firm. A full advisory memo and research summarization deployment at a regional firm with 80 to 150 staff usually sits between $65,000 and $160,000. Payback windows of 4 to 9 months are normal when the first deployment targets tax season workflows.
Beyond the per-task savings, the compounding benefit is capacity. Firms that integrated AI ahead of the 2026 filing season reported billable staff hours freed for advisory work, new client onboarding, and CPE that did not happen in prior seasons because everyone was drowning.
Compliance Considerations
Client data in accounting is highly sensitive. AI systems processing tax documents or client correspondence must comply with your data governance policies, IRS Publication 4557, state privacy laws (including California CCPA and New York SHIELD Act where applicable), and the terms of your professional liability insurance. Client data should not be processed through general-purpose consumer AI tools. Enterprise configurations with data privacy controls are required.
The AICPA's Statement on Standards for Tax Services addresses use of preparer technology, and the forthcoming 2026 ethics guidance adds specific language on AI oversight. Firms should document the AI tools used on each engagement, the review protocol, and the staff member responsible for the final output. This documentation lives with the workpapers and is available on request.
How to Evaluate Your Options
Start with the workflows that hurt most during peak season. For most firms, document intake, client status communication, and engagement letter administration are the top three by volume and frustration. Measure the current state honestly: hours spent, cycle time from request to response, error rates. These numbers form the baseline against which you will measure improvement.
Next, examine your practice management stack. CCH Axcess, Thomson Reuters UltraTax, ProSystem fx, Drake, and Canopy each expose different integration surfaces. A vendor promising "works with any tax software" usually means a manual export workflow. Direct integration with your practice management system is what makes the difference between a tool that saves hours and a tool that creates a new administrative burden.
Finally, press the vendor on data handling. Where is client data stored. Who has access. Is the data used for model training. Can you produce an audit trail showing every AI-generated output on a specific engagement. If any of those answers are fuzzy, the tool is not ready for a CPA practice. A polished website and brand on the vendor side does not substitute for a clear data processing addendum.
What Implementation Looks Like
Most accounting firm AI projects start with the highest-friction seasonal workflow: document intake, or client communication. Integration with your practice management system (CCH, Thomson Reuters, ProSystem fx, Canopy) shapes the technical approach. Initial setup and testing typically takes four to six weeks, with a go-live target ahead of the next heavy season. Staff training runs two to three weeks of parallel use so the team is comfortable before volume ramps.
Running Start Digital helps accounting firms identify the right starting points and implement systems that actually integrate with existing workflows, not bolted-on tools that create more management overhead. We pair AI integration services with the SEO, website design, and hosting and maintenance work that keep the client-facing side of a modern CPA practice running.
Frequently Asked Questions
Is it safe to use AI with sensitive tax data?
It depends entirely on how the AI is deployed. Consumer AI tools like general-purpose ChatGPT interfaces are not appropriate for client tax data. Enterprise AI systems can be deployed with data governance controls: private model access, no training on your data, audit logs, role-based access, signed data processing addendums. Those systems are compatible with professional data handling requirements and IRS Publication 4557 safeguards. The question to ask any vendor: where does our data go, who has access to it, and can you prove it.
Can AI handle the variety of document formats clients send in?
Yes, with some caveats. Modern AI handles PDFs, images of documents, Word files, Excel spreadsheets, and even reasonably clean phone photos. Low-quality scans, handwritten ledgers, and unusual foreign tax documents are more challenging. Most accounting firms find that 75 to 85 percent of client documents are clean enough for AI extraction, and the remaining 15 to 25 percent still require a human look. The net time saving is still substantial, and the workflow typically surfaces problem documents for review rather than pretending it extracted them correctly.
Will clients feel comfortable with AI handling their information?
Client perception depends on transparency and trust. Firms that communicate clearly ("we use technology to make your experience faster and more secure") rather than hiding AI involvement typically find clients receptive. The key is that AI is a backend tool enhancing your team's work, not a replacement for the advisor relationship clients chose your firm for. The rare client who is uncomfortable appreciates the option to opt out of AI-assisted workflows.
How does AI handle state tax complexity across multiple jurisdictions?
AI can assist with state tax research, summarization of state-specific requirements, and drafting state-specific disclosures, but multi-state tax planning and compliance require licensed tax professionals who understand the specific rules. AI reduces the research burden. It does not replace the judgment needed for nexus determinations, apportionment decisions, or complex pass-through entity tax elections. The practical value is faster research triage, not autonomous multi-state practice.
How does AI interact with professional liability and malpractice exposure?
Your professional liability coverage almost certainly applies to AI-assisted work the same way it applies to intern-assisted work: the licensed professional who signs the return owns the outcome. Most carriers have begun issuing guidance that expects firms to document AI use, maintain review workflows, and avoid using tools that train on client data. Several carriers offer premium credits to firms with documented AI governance policies. Talk to your broker before your next renewal.
What is the realistic timeline for a small firm to roll out AI?
A 10-to-30-person firm can typically deploy a focused AI workflow (document intake, client communication, or engagement letter automation) in 6 to 10 weeks from kickoff to production use. The work breaks down roughly as: two weeks of discovery and integration scoping, three to four weeks of build and configuration, one to two weeks of parallel testing, and one to two weeks of staff onboarding. Starting the project in August or September positions the firm to enter the next tax season with the new workflow live and tested.