How ConnectCPA is using AI to run a better accounting practice

Implementing AI in an accounting firm sounds straightforward in theory. In practice, it's messier, and more interesting, than most firms expect. Tom Sun, VP of Operations at ConnectCPA, a cloud accounting firm established in 2014, shared an honest account of what's actually working, where the friction is, and what the profession needs to reckon with.

AI is running in two parallel tracks

At ConnectCPA, AI isn't a single initiative. It operates in two distinct areas that happen to feed into each other: service delivery to clients, and internal operations.

On the client-facing side, AI shapes the output that reaches customers: how information is structured, how advice is framed, how quickly practitioners can respond. On the operational side, it's doing the work that used to require hours of manual consolidation, such as time tracking analysis, profitability by department, capacity planning. Both matter, and the quality of one depends heavily on the other.

The client dynamic has shifted

Tax season now comes with a new wrinkle: clients are arriving better prepared than ever before. They've already consulted AI tools on their own. They know what positions they want to take. They have views on their planning.

This raises the bar for practitioners. The conversation is no longer about conveying basic knowledge. It's about applying context. A client who comes in saying "AI is telling me to take this tax position. What do you think?" isn't looking for education. They're looking for professional judgment that accounts for the specifics of their situation. That's a fundamentally different interaction, and it requires AI-assisted practitioners who can move quickly and engage at a strategic level.

Where AI has made the biggest internal difference

The clearest win Tom described was in capacity planning. Before automation, the process was painful: pull data from the CRM, extract time tracking, reconcile staff hours with client MRRs, consolidate everything into a report. Hours of work for analysis that needed to happen regularly.

With automation tools like Zapier, that consolidation dropped to minutes. But automation had limits. It couldn't do the higher-level analysis: figuring out which departments were overstretched, modelling scenarios for new client intake, stress-testing what happens if four or five clients are added in a quarter. That's where AI came in.

ConnectCPA now uses AI to take the consolidated data and run scenarios: what does capacity look like in two or three months? What are the financial implications of onboarding this client? The AI doesn't make the decision, but it does the analysis that makes the decision possible, quickly and with enough rigour to be a useful sanity check.

The same logic applies to sales. ConnectCPA uses AI to summarise discovery calls, pulling out key details about client revenue, volume, complexity, number of employees, and whether they have existing processes for things like deferred income or depreciation schedules. That summary feeds into a pricing model that weighs up capacity data from the CRM. What used to require significant back-and-forth can now surface a well-informed quote much faster.

The honest picture on reluctance

Tom was candid about why AI in accounting firms is still more talked about than adopted, why many firms are still sitting on the sidelines. It’s typically because of the scale of what proper implementation actually requires.

The accounting industry in general tends to move slowly with technology adoption. But beyond that, AI implementation isn't a feature you bolt on. For firms that have been running the same way for ten or twenty years, it means rebuilding workflows from the ground up. And if you're going to partially implement, you're probably not getting the full benefit anyway. That tension, between wanting the gains and reckoning with the disruption, is where most hesitation lives.

There's also a learning cost that people underestimate. AI is only useful when you know exactly where and how to deploy it. Getting that right requires upfront investment, ongoing refinement, and a team that's genuinely bought into the change. Build the infrastructure without team buy-in, and it doesn't matter how well you've designed it.

The distinction between automation and AI

The difference between old-school automation and what AI adds now is meaningful, and Tom drew it clearly. Bookkeeping automation is good at transactional and administrative work: if this condition is true, do that action. Given the right setup, it runs reliably and fast.

What it couldn't do was anything that required reasoning over context. Looking at consolidated data and saying: given current capacity, anticipated intake, and the nuances of how this particular department operates, here's what the outlook looks like. AI adds that layer. It's not replacing the automation; it's extending what becomes possible on top of it.

AI as companion, not replacement

On whether AI will eventually handle everything without human involvement: Tom's view was clear. Accounting is a relationship business. Clients put their businesses in the hands of people they trust. They don’t just want people who can process information correctly, but people who understand the context of their specific situation. That's something AI, for all its capability, doesn't have in the way a trusted adviser does.

This reframes what AI for accountants actually does. The role that's emerging isn't data entry or transactional processing. It's reviewer, interpreter, and adviser. This is the foundation of modern AI advisory services: the practitioner's value shifts from doing the work to evaluating what the work means, and helping clients make decisions based on it.

Practical advice for firms thinking about the next step

When asked what he'd tell other practitioners right now, Tom's recommendation was to start while things are fresh. Tax season, for all its pressure, is actually the best diagnostic available: it's when your real constraints show up. The workflows that held under normal volume, the manual steps that nobody questioned, the places where everything slowed down. Those are visible now in a way they won't be in July.

The advice: write it down. Identify the areas where time was spent that didn't translate to client value. Ask whether each one could be automated, or whether AI could help. Then start building during the slower months because it will take longer than expected, and there will be trial and error. Get the team trained and involved early. And focus on the areas where the ROI is clearest first.

The firms seeing the most transformation aren't the ones who implemented the most AI. They're the ones who were most deliberate about where they put it.