There's a solo CPA in the Midwest whose practice has been quietly outgrowing everyone's expectations. He brings in $2.5M in annual revenue with no partners, no associates, and no employees except for a part-time bookkeeper and a seasonal tax assistant.
That shouldn't be possible. Capacity constraints would suggest that a solo CPA is capped at roughly $1M-$1.5M in annual revenue before needing to either hire staff or cap new clients. Yet this practitioner is at $2.5M and still taking new clients.
The answer isn't genius business strategy or cutting corners on service quality. It's agentic AI handling the work he'd normally delegate.
When I asked him to walk through his workflow, the story that emerged is instructive because it shows what's possible with AI today (not theoretical AI, actual 2026 AI) and how it changes the economics of small professional service firms.
The Problem That Agentic AI Solved
Let's start with the constraint. A solo CPA's time is limited. He can bill roughly 2,000 hours per year (assuming typical vacation, professional development, business development time). At $300/hour average billing rate, that's $600K in potential annual revenue.
But CPAs don't work alone. Every client needs support work: preliminary research, preliminary return preparation, client communication, documentation review, internal QA, and follow-up. A typical junior CPA (one to three years of experience) handles 30-40% of the billable work on a client engagement, with the senior CPA handling 60-70%. That work takes capacity.
So a solo CPA operating at full capacity typically dedicates 40-50% of their billable hours to direct client work and 50-60% to delegated work that still needs their supervision. That's a natural break point. You either:
- Cap your practice at sustainable capacity (roughly $1M revenue)
- Hire a junior staff member to handle delegated work (requires profit to support salary, increases operational complexity, creates management burden)
- Find another way
This CPA chose option 3.
The Implementation: Agentic AI as Junior Staff
He didn't use a single tool. He built a system using multiple AI agents, each trained on specific workflows that junior staff typically handle.
Here's what he built:
Agent 1: Preliminary Return Preparation
The work: Junior CPAs spend 5-10 hours per small business return gathering information, building out the tax return framework, preparing schedules, and doing preliminary calculations.
What the agent does: It ingests client data (prior year return, current year financial statements, new transaction summaries), identifies what information is missing, prepares the initial return structure, populates schedules with preliminary numbers, and flags items needing clarification.
The implementation: He trained the agent on 50 completed tax returns from his practice. The agent learned his firm's standards, which items typically need follow-up, how he structures schedules, and the common pitfalls in his client base.
The result: Junior staff would spend 8 hours on preliminary preparation. The agent handles it in 30 minutes of machine time and creates an output the CPA spends 1-1.5 hours reviewing and refining. Net time savings: 6-7 hours per return.
Applied to his client base: Assuming 150 small business returns per year, that's 900-1,050 hours saved annually. At $85/hour junior staff cost (loaded), that's equivalent to $76,500-$89,250 in labor cost eliminated.
Agent 2: Research and Preliminary Analysis
The work: Junior staff spend 2-5 hours per client on tax research, analyzing financial position, identifying planning opportunities, and preparing analysis.
What the agent does: It takes a client's financial picture and tax situation, identifies relevant tax codes and regulations, researches recent guidance, analyzes their specific situation against those rules, and prepares a preliminary analysis.
The implementation: He connected the agent to his research databases (so it has access to primary sources, not just summaries), built prompts around the specific research questions that junior staff typically investigate, and trained it on the analysis framework his firm uses.
The result: What takes a junior CPA 3-4 hours takes the agent 10-15 minutes of processing time and the CPA 30 minutes of review. Net savings: 2.5-3 hours per engagement.
Applied to his client base: Assuming 250+ annual engagements with research components, that's 625-750 additional hours freed up.
Agent 3: Client Communication and Documentation
The work: Junior staff handle routine client emails, prepare engagement summaries, track follow-up items, and maintain client communication logs.
What the agent does: It monitors incoming client emails, drafts responses for routine inquiries, prepares engagement summaries, tracks outstanding items, and alerts the CPA when human judgment is needed.
The implementation: Connected to his email and case management system, trained on his firm's voice and communication standards, configured to handle specific categories of inquiries.
The result: Eliminates roughly 50% of routine client communication work.
Agent 4: Compliance Checklist and Deadline Management
The work: Tracking filing deadlines, preparing compliance checklists, following up on missing documents, managing engagement timelines.
What the agent does: Maintains a calendar of deadlines for each client, generates automated checklists, flags missing items, sends reminders.
The implementation: Connected to his calendar and engagement tracking system, trained on his compliance requirements by practice area.
The result: Eliminates the need for a separate compliance tracking system and the administrative work that goes with it.
The Economics of the System
Here's where it gets interesting. Let's quantify the impact:
Time freed up:
- Preliminary return prep: 900-1,050 hours/year
- Research and analysis: 625-750 hours/year
- Client communication: 300+ hours/year
- Compliance management: 200+ hours/year
- Total: 2,025-2,300 hours per year
What he did with it:
- Took on 50 new clients (representing $500K+ in new annual revenue)
- Reduced overall stress and improved work-life balance
- Spent more time on high-value advisory and planning work rather than mechanical work
Cost of the system:
- Claude API usage: $200-300/month
- Integration and maintenance: roughly 2-3 hours per month (which he handles himself)
- Total annual cost: $3,000-$4,000
ROI: If he'd hired a junior CPA to handle delegated work ($50K-60K salary, $70K-85K loaded), he'd need to generate $150K+ in additional revenue just to break even. Instead, he's generating $500K+ in additional revenue with a $3,500 annual cost.
That's not just ROI. That's a completely different business model.
The Critical Details That Made It Work
Not every solo CPA reading this will immediately replicate his success. The implementation details matter enormously.
1. Training Data Matters
He spent significant time training each agent on his specific practice. Generic agents trained on public tax data wouldn't have worked. Agents trained on his 50-100 previous returns, his communication templates, his analysis frameworks, and his client profiles worked far better.
The lesson: You can't drop in a general-purpose AI agent. You have to train it on your specific work patterns, your standards, and your context.
2. Human Review Is Still Essential
He reviews every output before it goes to a client. The agent gets 90%+ of technical tax matters right, but the 10% that's wrong can be costly. His workflow is: agent prepares, he reviews, he modifies as needed, then it goes to the client.
The lesson: Agentic AI is a force multiplier, not a replacement. It handles the mechanical work; he handles judgment.
3. Integration Is Key
The real power came from connecting the agents to his existing systems: email, case management, document storage, research databases. Isolated AI tools would have created more work. Integrated systems eliminated work.
The lesson: Choose tools that integrate with what you already have or invest in integration. Standalone AI tools often fail because they create friction.
4. Starting Small Matters
He didn't roll out all four agents simultaneously. He built preliminary return preparation first, used it for two months, measured the impact, then added the next agent. This phased approach let him refine each agent's training and output quality before adding complexity.
The lesson: Implement one agent workflow, get it working perfectly, then expand. Don't try to automate everything at once.
What This Means for Other Solo and Small CPAs
Not every CPA practice will be willing or able to implement agentic AI at this level of sophistication. It requires technical comfort, time investment in training the agents, and willingness to let AI handle client-facing work (with review).
But for CPAs who can make that investment, the payoff is substantial. A solo CPA could plausibly:
- Increase annual revenue from $1M to $1.5M-$2M by freeing up capacity
- Reduce the need for staff hires until much larger revenue levels
- Shift more time toward higher-value advisory work
- Improve work-life balance by eliminating mechanical tasks
The catch: you need to be comfortable with AI. If you're still skeptical about AI's ability to handle professional work, this approach won't work. If you're willing to experiment and iterate, the opportunity is real.
The Broader Implication
This case study illustrates something important: agentic AI isn't just a tool for large enterprises. It's potentially most valuable for solo practitioners and small firms because the leverage is highest.
A large accounting firm hiring a junior staff member is replacing labor with labor. A solo CPA implementing agentic AI is replacing labor with capital (the cost of building and maintaining the system). The capital approach is dramatically cheaper.
Over the next 18-24 months, I expect we'll see significant consolidation among solo practitioners. Some will embrace agentic AI and scale dramatically. Others will remain small, cap their practices, or eventually exit. The gap between the two groups will widen.
If you're a solo CPA reading this, the question worth asking is which group you want to be in.
Subscribe to the Briefing
Continue accelerating your intelligence with unfiltered ROI tracking, tool benchmarks, and architectural implementation drops.