When Your Accountant Is an AI: Building the Elite Finance Solopreneur Practice
How fractional CFOs and finance solopreneurs are building leaner practices, and why human judgement matters more now than it ever has.
In Session 12 of the Future of Finance Lab, Lance Rubin is joined by Carl Seidman, FP&A trainer, LinkedIn Learning instructor, and fractional CFO based in the US, for a no-nonsense conversation about what AI actually looks like in a serious finance practice.
Carl has built dozens of financial models over 15 years and now runs his entire business accounting function off Claude.
Together they dig into the real risks, the real workflows, and what it genuinely takes to build a premium practice when everyone around you has access to the same tools.
Future of Finance Lab Nugget
Carl Seidman is running his entire accounting and finance function off Claude, replacing hours of bookkeeper time every month for himself and his team.
This isn’t just about offloading admin.
A practitioner with real process knowledge can codify their frameworks into repeatable AI workflows at a precision that off-the-shelf tools can’t touch.
Stack domain expertise onto well-built skills and you get genuine efficiency.
That’s very different from plugging in a generic template and hoping for the best.
Key Insights and Frameworks
AI Is Not a Deterministic Tool. Stop Treating It Like One
The thing that gets lost in the AI noise: these tools are probabilistic, not deterministic.
A traditional RPA bot executes its rules exactly as written, every time.
Claude, ChatGPT, Gemini, all of them, make educated guesses. That’s how they’re built. You can’t prompt your way around it. A fancier model just makes the wrong answer look more convincing.
Lance had a client come in recently with a three-way cash flow model built with Claude.
It looked polished.
But there was no monthly granularity, no seasonality, no error checks, no proper working capital treatment, and none of the best-practice structure you’d expect from an experienced modeller.
The client’s accountant flagged it straight away.
The model wasn’t anywhere near ready for an investor IM. The client was engaged, the model was rebuilt properly, and the lesson cost them time and money.
Hallucination risk sits somewhere between 20% and 90% depending on context.
That’s a massive range, and it moves based on how much specific, useful context the model has been given.
Even with great context, it never reaches zero.
Lance also ran into a client who’d been actively using AI but had never heard the word “hallucination.”
That gap in basic understanding is its own risk, and it’s more common than people think.
It’s Not About the Prompt. It’s About the System
Here’s what most people get wrong: the prompt is the smallest part of a well-built AI workflow.
Carl spent a full day, not the 30 minutes people imagine, building a single skill for one type of model: a 13-week cash flow.
Back and forth with Claude, edge case after edge case, making explicit decisions about what should happen when inputs were missing or ambiguous, and encoding every one of those decisions into the skill.
The result replicates his framework exactly. But the investment to get there is nothing like what people expect when they first open Claude and start typing.
Lance calls this system thinking, and it’s a core concept in the AI-Powered Accountant programme.
Design for the beginning, the middle, and the end of every AI-assisted process.
Where does the data come from? What happens in the middle? And how do you know the output is correct?
The validation layer is not optional. That’s where professional judgement lives. That’s also where you build or lose client trust.
Lance also outlined the five-layer stack he sees emerging for serious AI users in finance:
- The prompt at the base,
- Then the skill,
- Then connectors and MCPs that pull live data,
- Then artefacts (dashboards, HTML reports, interactive outputs), and
- Finally digital workers: skills built from multiple skills, orchestrating end-to-end processes.
His own digital worker manages his daily operations briefing by pulling from Google and Microsoft 365, applying skills, and generating a live dashboard artefact.
It takes time to build to that level.
It’s also exactly where the profession is heading.
Build Your Own Skills. Don’t Just Use Anthropic’s
Both Lance and Carl were clear about the limitations of Anthropic’s open-source skills.
Carl tested the financial statement skill and found it functional but generic.
Actuals, budget, dollar variance, percentage variance.
Clean, but undifferentiated. And you can’t edit them.
Anthropic’s built-in skills are locked.
What Carl built instead was a financial modelling foundation skill encoding his own standards and decision logic, with specialist skills referencing it as a base.
His auditor skill ranks vulnerabilities by materiality and writes narrative commentary in Carl’s own voice. No open-source template does that.
Lance hit Anthropic’s platform limit at 200 skills and had to restructure: retiring old ones, converting others into plugins.
He’s written about the sprawl. The discipline of building purposeful, well-governed skills is exactly what creates competitive advantage that lasts.
If everyone just runs Anthropic’s standard templates, everyone delivers the same analysis on different numbers.
That’s the commodity trap.
Practical Takeaways and Actionable Strategies
Where to Start: Pick Something Low Risk and Annoying
For finance professionals just getting started with AI, Carl’s advice is simple: don’t start with the month-end close.
Don’t touch client work yet.
Start with something low-risk, frustrating, and repeatable.
Commentary is the obvious entry point.
Get it wrong and you rewrite it. You don’t lose a client.
Carl walked through his expense coding example.
Rather than telling Claude to “build me a process,” the approach is to first ask it to map out the full process: steps, order, decision points.
Then work through each one individually.
Test it, find where it fails, encode the fix into the skill. Run it again, find the next failure, fix that too.
This iterative approach is the opposite of “give it a prompt and see what happens.” It’s the only way to build something you’d actually trust in front of a client.
The Commodity Trap: What Actually Separates You
Carl put it plainly: with enough time and a modest amount of skill, anyone can write prompts and build skills to produce the technical outputs finance professionals currently deliver.
The accounting equations are in the textbooks.
The modelling frameworks are publicly available.
The risk isn’t that AI replaces finance professionals.
It’s that it commoditises the output, and the only thing left that matters is the layer you can’t automate.
Critical thinking.
Institutional knowledge.
Industry understanding.
The ability to look at an output and recognise when something’s off. Carl was also direct about something most finance people would rather not hear: sales, marketing, and communication.
Human-verified logic, proprietary frameworks, and the trust you build through years of showing up and delivering: those are what clients pay for.
No open-source skill set replicates any of it.
Important Quotes from the Session
Some of the key quotes from the session that are worth replaying and considering:
Carl Seidman
“I am now running my entire accounting and finance function for my business off of Claude.” (~7:51)
“What separates us is the critical thinking, is the institutional knowledge, is the ability to say, I can look at something that AI cannot.” (~16:00)
“If everyone in the next couple of years will have that ability, then we’re all commodities.” (~16:30)
“It’s almost like if you could tell a staff member on your team, this is exactly what I want you to do every step of the way, every single time. You got to tell Claude to do that, otherwise it’s going to run amok and it’s going to make stuff up.” (~51:00)
“You have to be very, very thoughtful and not just say, hey, I’m just going to go all in and be reckless with a wrecking ball.” (~57:00)
“Sales, marketing, and effective communication. Everything that we’re talking about here, accounting and finance, those are table stakes.” (~43:00)
Lance Rubin
“You can’t suddenly make a probabilistic engine deterministic. That’s not going to happen. That’s not the way this technology was built.” (~19:00)
“You almost have to go in with the mindset that it’s completely wrong and you need to validate.” (~46:00)
“This is someone who’s been using AI that’s never heard of the word hallucination. And I had to explain hallucination.” (~20:00)
Frequently Asked Questions
What is the difference between probabilistic and deterministic AI?
Deterministic tools like RPA bots execute exactly as programmed every single time. AI tools like Claude, ChatGPT, and Gemini are probabilistic: they make educated guesses based on context. You cannot prompt your way around this. A more sophisticated model simply makes incorrect outputs look more polished.
What is AI hallucination risk in finance, and how high is it?
Hallucination risk in finance AI ranges from 20% to 90% depending on the quality and specificity of context provided. Even with excellent prompting, it never reaches zero. Treat every AI output as unverified until you have checked it.
How should finance professionals start using AI safely?
Start with something low-risk, frustrating, and repeatable. Commentary writing is the safest entry point: if the output is wrong, you rewrite it. Avoid using AI for month-end close or client-facing financial models until you have built robust validation workflows and understand the tool’s failure modes.
What is system thinking in AI-assisted finance workflows?
System thinking means designing an AI workflow across three stages: where data comes from (beginning), how the AI processes it (middle), and how outputs are validated (end). The validation layer is not optional. It is where professional judgement lives and where client trust is built or lost.
What is the five-layer AI stack for finance professionals?
The five layers are: (1) Prompt, (2) Skill, (3) Connectors and MCPs that pull live data, (4) Artefacts such as dashboards and HTML reports, and (5) Digital workers, which are multi-skill orchestrators that automate end-to-end processes. Each layer builds on the one below it.
What is the commodity trap in AI for finance?
The commodity trap occurs when AI democratises access to technical finance outputs. When everyone uses the same skills and templates, the technical work becomes a baseline rather than a differentiator. What separates practitioners is critical thinking, institutional knowledge, and the ability to verify AI-generated outputs.
How long does it take to build a proper AI skill for finance?
Much longer than most people expect. Carl Seidman spent a full day building one skill for a single model type: the 13-week cash flow. The five-minute-prompt mindset is the wrong frame entirely. The investment is front-loaded, but the efficiency gain compounds once the skill is built correctly.
What is the HACK framework for AI in accounting?
HACK stands for Hygiene, Automation, Capability, and Knowledge. It is a structured approach for finance and accounting professionals to integrate AI into their practice, starting with internal process hygiene and progressing through automation toward deeper capability and knowledge development.
The Future of Finance Lab, a Journey of Learning and Adaptation
Lance is presenting at the FMI Global Leaders 24-Hour Conference with a model evaluation across the latest frontier AI tools applied to real financial modelling tasks. If you want to see what rigorous, human-verified AI testing looks like in practice, that’s the session to watch.
For a deeper dive into AI system thinking, skills architecture, and the HACK framework (Hygiene, Automation, Capability, Knowledge), the AI-Powered Accountant programme covers all of it across ten structured modules.
More sessions coming in the Future of Finance Lab. Plenty more to get into.
This session encapsulated the core message of the entire Future of Finance Lab series: while technology and AI are transforming the finance function, the human element—curiosity, adaptability, and best practice, remains irreplaceable. The series has equipped participants with frameworks, tools, and mindsets to navigate ongoing change, emphasizing that continuous learning and critical thinking are the keys to success.
Call to Action
Ready to future-proof your finance skills?
Explore the full Future of Finance Lab series for in-depth sessions, practical resources, and expert insights. Whether you’re a seasoned modeler or just starting out, there’s always more to learn—and the journey is just beginning.
Check out the recordings, download the workbooks, and join the conversation as the world of finance continues to evolvecussions, and resource sharing, this ongoing and free series remains a must-attend for anyone looking to future-proof their finance skills.
Looking Ahead
The Future of Finance Lab is more than a webinar—it’s an interactive community dedicated to learning, sharing, and growing together. We invite you to join our next session, connect with peers, and continue building the skills that will define the future of finance.
For resources, recordings, and more, visit our Knowledge Hub and stay tuned for further updates!
The session was also recorded and is available to be viewed on YouTube.
Want to discuss how AI can transform your finance function? Get in touch with Model Citizn to learn more about our approach to AI-powered financial modelling and strategic implementation.
Want to take action? You can by clicking here
We launched our Future of Finance Lab series of FREE monthly lightening sessions hosted on Maven to unpack all the hype and give solid foundations to finance professionals. Join our Friday Future of Finance Lab by registering for the lightening sessions on Maven.
Friday’s Future of Finance Lab
Future of Finance Lab
Join our community where finance professionals share insights, solve real challenges, and stay ahead of the AI revolution (max 50 attendees).
Key Benefits:
✅ Live Problem Solving – Bring work challenges
✅ Expert Insights – Practitioners not trainers
✅ AI Integration – Leverage AI in finance
✅ Networking – like-minded professionals
✅ Zero Cost – Completely free, always
What You’ll Get:
✅ – Monthly Focus Areas in the following areas
✅ – Financial Modeling Deep Dives
✅ – Data Analysis & Power BI Techniques
✅ – AI Applications in Finance
✅ – Open Forum & Case Studies Exclusive Resources
✅ – Meeting recordings and AI-generated summaries
✅ – Templates and tools shared during sessions
✅ – Resource library access
Who Should Attend:
✅ – Finance professionals wanting to upskill
✅ – Data analysts working with financial data
✅ – Anyone curious about AI in finance
✅ – Business leaders seeking data-driven insights
Buckle up and look forward to seeing you there at the Future of Finance Lab!
AI Accounting Finance #CFO #FinancialModeling #AIAdoption #FutureOfWork
Lance Rubin is CEO and Founder of Model Citizn and EXL Cloud. He is a Chartered Accountant, FMI Certified Trainer, co-author of the CA ANZ Financial Modelling Study Guide, and creator of the AI-Powered Accountant programme on Maven, currently live at Level 1 with Level 2 in design, and author of the companion book “AI-Powered Accounting with Excel and Power BI” (Packt). He has worked at PwC, KPMG, Investec, and NAB, and is based in Melbourne, Australia.