A year ago, many conversations about AI for CPA firms still started with the same basic question: What can we actually do with this? That question is quickly being replaced by harder ones. At recent industry events and in conversations across the profession, firm leaders are talking less about whether AI matters and more about which workflows are worth changing, what still requires professional judgment and how to put the right guardrails around its use.
The data reflects that shift. Thomson Reuters reports that 81% of tax and audit professionals now use AI tools at least several times a week. But individual use does not necessarily mean a firm has figured out how AI fits into the way work gets done.
For CPA firms already dealing with capacity constraints, reviewer bottlenecks and rising client expectations, that distinction matters. The real promise of AI is not simply doing the same work faster. It is creating more room for professionals to spend time on review, judgment and client work.
Where CPA Firms Are Getting Practical Value From AI
Some of the most useful AI applications in accounting are not particularly flashy. They take work that is repetitive, research-heavy or time-consuming and make it easier for professionals to get to the part that requires judgment.
Recent real-world examples from CPA firms show AI already being used across tax, assurance, bookkeeping, workflow improvement and client service. Several areas stand out:
- Tax and research: AI can help professionals search authoritative materials, summarize findings, organize source documents, identify missing information and prepare first drafts of client explanations.
- Audit and assurance: Teams can use AI to organize and analyze documents, support research, identify anomalies or areas requiring additional attention and assist with workpaper preparation.
- CAS and advisory: AI can assist with variance analysis, reporting commentary, forecasts, scenario development, KPI analysis and client meeting preparation.
- Firm operations: Meeting summaries, internal knowledge searches, routine communications, document creation, client intake and workflow routing can reduce the administrative work surrounding client service.
The value is not the task itself. It is what happens when a research question takes less time, a reviewer gets a cleaner first pass or professionals spend less of the day chasing documents and preparing information.
AI is also moving beyond one-off prompts as more applications embed it directly into existing workflows and begin automating multiple connected steps. That does not mean every firm should race toward autonomous processes. It means the question is shifting from “What can this tool do?” to “Where should this fit into our work?”
The Harder Part Isn’t Buying Another AI Tool
Giving employees access to AI is easy compared with changing how work actually gets done. That is one reason AI access does not automatically translate into AI adoption. An employee might use AI to summarize a meeting, draft an email or conduct research without the firm changing a single meaningful workflow. The bigger gains start with the process.
Where is work getting stuck? Which tasks consume time without requiring much judgment? What gets repeated consistently enough that AI could improve it? Who owns the process today? What would a better result actually look like? A useful test is simple: If two people cannot describe the workflow the same way, it probably is not ready to be automated with AI.
That matters because technology adoption and integration have become significant business issues for accounting firms. In AICPA’s 2026 survey, firms with 101 to 500 professionals ranked technology adoption and integration as their No. 1 current issue, while managing change related to technology and AI ranked as the leading five-year issue across firm sizes. And even a well-defined workflow can run into another problem: the data and permissions underneath it.
If information lives across disconnected systems, permissions are inconsistent or employees cannot reliably find the right documents, AI does not make those problems disappear. Microsoft notes that Copilot works within the access permissions users already have. In practical terms, AI may follow your permission structure perfectly while exposing the fact that the permission structure needs work. That is why data readiness, access and workflow consistency matter alongside the AI tool itself.
Where AI Stops and Professional Judgment Starts
AI can analyze a document quickly. It cannot assume a CPA’s professional responsibility for what happens next. Recent AICPA guidance on the use of technology output reinforces that professionals still need to determine whether technology-generated output is appropriate for its intended use.
For CPA firms, the distinction is straightforward: AI can help collect, organize, analyze, summarize and draft. The professional remains responsible for deciding whether the result is accurate, sufficiently supported and appropriate for the engagement. That matters when AI touches technical research, tax positions, audit evidence, client recommendations or other work requiring professional judgment.
Firms also need clear expectations for how AI is used, particularly when client or firm data is involved. Approved platforms, data-handling rules, appropriate access and required review should all be part of the conversation.
As adoption expands, those decisions increasingly become part of the firm’s broader approach to security, quality and risk management. For a deeper look at what happens when AI use starts moving faster than those controls, read our perspective on AI governance.
What a Good AI Pilot Looks Like
There is no universal AI roadmap that works for every accounting firm. But a strong pilot usually has a few things in common.
- It starts with the workflow, not the tool. Look for repetitive work, bottlenecks or processes where professionals spend too much time preparing information before applying judgment.
- It solves a bounded problem. “Use more AI” is not a project. Improving tax research, speeding document intake or reducing time spent preparing for client meetings is.
- The rules are clear before the pilot scales. Employees should know which platforms are approved, what data can be used and when another person needs to review the output.
- Professionals stay accountable. AI should make it easier to prepare, analyze and evaluate the work, not create ambiguity about who owns the result.
- The firm measures what changed. The number of AI licenses a firm owns is not an AI success metric. Neither is the number of prompts employees write. Better measures are whether the work takes less time, requires less rework, moves through review faster or creates capacity the firm can put somewhere more valuable.
Saving 30 minutes on a task is not the end goal. The value comes from what the firm does with that time – faster turnaround, less overtime, more review capacity or more time with clients.
AI adoption in CPA firms is becoming less about access to the newest technology and more about where it genuinely improves the work. The strongest opportunities are often practical: faster research, less repetitive work, better access to information and more capacity for work requiring experience and judgment.
Getting there requires sound workflows, reliable data, clear expectations for responsible use and professionals who understand where AI fits and where it does not.
As firms move beyond experimentation, governance becomes part of that equation as well. Netgain’s AI Governance Service gives CPA firms a more structured approach to managing AI risk as adoption grows.