Every CPA firm we talk to is experimenting with AI right now.
Some are building workflows inside Microsoft Copilot. Others are testing Claude or ChatGPT for financial summaries, benchmarking reports, or client communication. A few firms are already creating internal advisory tools using these platforms.
And honestly, a lot of it is impressive.
It has never been easier to generate commentary around financial statements, identify trends, or create a polished client-facing report in seconds.
For firms trying to expand advisory services, that’s a real opportunity.
But there’s another conversation happening underneath all of this that doesn’t get discussed as often.
The hard part usually isn’t the AI.
The hard part is the financial infrastructure underneath it.
Most firms today can build something useful with AI.
A smart internal team can create:
That’s no longer the barrier.
What many firms are discovering is that the complexity shows up later, once the project starts moving beyond a pilot or a single office.
Questions start coming up that sound operational, not technical.
How do we normalize financial data across multiple systems?
How do we standardize trial balances when every client structures their chart of accounts differently?
How do we maintain consistency across offices, industries, and advisors?
How do practice leaders trust that the benchmark data is actually comparable?
That’s where projects often become much larger than originally expected.
In theory, benchmarking sounds straightforward.
In practice, accounting data is incredibly inconsistent.
Even within the same industry, two companies can categorize expenses completely differently. One restaurant client may classify labor a certain way. Another may split it across multiple accounts. A third may bury costs in overhead.
Now multiply that across:
Then layer in system migrations, evolving workflows, and changing industry structures.
This is the part AI alone does not solve.
AI can analyze data remarkably well. But it still depends on the quality, consistency, and structure of the information it receives.
Without that financial intelligence layer underneath it, firms often end up with fast answers built on inconsistent foundations.
This is where the conversation tends to shift inside firms.
What started as:
“Let’s build some AI-powered advisory capabilities”
slowly becomes:
“We need a reliable financial data infrastructure.”
Because once leadership wants visibility across the entire firm, the requirements change.
Now the firm needs:
The challenge is no longer generating one impressive report.
The challenge is maintaining trust in the data over time.
That’s a very different problem.
One of the more interesting trends right now is that firms are becoming increasingly sophisticated about AI very quickly.
Many are already using:
The experimentation phase is happening fast.
But the firms getting the most traction are usually the ones thinking beyond the AI layer itself.
They’re asking operational questions:
That’s a much more mature conversation.
None of this means firms should slow down their AI efforts.
Quite the opposite.
AI is already helping firms:
The opportunity is very real.
But over time, we think firms will increasingly realize that AI becomes dramatically more valuable when paired with structured financial intelligence underneath it.
The firms that scale advisory successfully probably won’t treat AI as a standalone solution.
They’ll combine AI with:
That combination is what turns isolated AI experiments into repeatable advisory infrastructure.
Right now, a lot of firms are asking:
“Can we build this ourselves with AI?”
In many cases, the answer is yes, at least partially.
The more important question may be:
“Do we want to build and maintain the financial infrastructure underneath it?”
Because that’s the part that tends to grow more complex over time.
And it’s also the part that ultimately determines whether advisory capabilities scale consistently across the firm or remain isolated experiments inside a few teams.