If you’ve ever worked in or around an accounting firm during tax season, you know that calling software support isn’t optional. It’s basically a scheduled activity. The software is complex, the tax code changes every year, edge cases appear constantly. Something always requires a call.
Amanda, a CPA I’ve been consulting with, made it through her most recent tax season without a single one.
“I didn’t have to call tax software support. Not one time. After I created that GPT.”
And then she added something that stuck with me: “That’s unheard of. No tax practitioner would tell you they didn’t have to call the software company at all.”
She’s right. This isn’t a minor improvement. It’s the kind of thing that makes other CPAs in the industry do a double-take.
What She Actually Built
Before the season, we worked together to build a custom GPT trained specifically on her practice.
Not generic tax knowledge — she’s not trying to replace her own expertise with Wikipedia. The GPT was built around her specific methodology: how she structures S-corp entities for her client base, the deductions that come up most often, the error patterns she’s seen repeat across hundreds of returns, and the review criteria she applies when checking her team’s work.
The questions Amanda would normally escalate to software support — nuanced edge cases, situations where the software behavior wasn’t matching what she expected — the GPT already understood those. Because the GPT was built from Amanda’s own knowledge of how those situations play out in practice.
It’s a subtle but important distinction. Generic AI tools know a lot about tax law in general. Amanda’s GPT knows her firm’s way of handling tax situations specifically.
The Team Effect
The impact showed up in another way that surprised even her.
One of Amanda’s team members spent two hours trying to work through a complex tax scenario. They couldn’t get to a clear answer. Amanda ran the exact same question through the custom GPT.
Fifteen minutes.
This is the part of custom GPT builds that most people don’t anticipate when they start. You think you’re building a tool for yourself. What you’re actually building is a way to distribute your expertise across your team, available at any hour, for any question, at no marginal cost per use.
Amanda’s senior judgment used to be a bottleneck. Complex questions had to wait for her bandwidth. Now that same judgment is accessible to everyone on her team, whenever they need it.
What Most People Get Wrong About Custom GPTs
There’s a version of this that most people build — and it’s not this.
The typical approach: take a general-purpose AI tool, give it a generic prompt about your industry, and use it as a faster search engine. Ask it things. Get answers.
That’s useful. But it’s not what Amanda has.
When you build a GPT trained on your actual methodology — your documented approach, your historical cases, your review criteria — you’re doing something different. You’re codifying expertise that normally lives only in your head and making it deployable.
The software support team answers questions from tens of thousands of accounting firms. They know the software well. They don’t know Amanda’s firm, her clients, her preferred structures, her typical edge cases.
Amanda’s custom GPT does.
That gap is where the result comes from. It’s not that the GPT is smarter than the software support team. It’s that it knows things about Amanda’s practice that no generic support line ever could.
The Broader Principle
There’s a way to think about AI in professional services that goes beyond “using AI to speed things up.”
It’s this: your expertise is an asset. And like any asset, it can be structured so it works harder.
Right now, your expertise works when you’re working. It’s available when you’re available. It answers questions when you’re in the room. When you’re not, things wait — or get answered incorrectly by someone with less experience.
A well-built custom GPT changes that. Your judgment becomes available around the clock, to any team member who needs it, without requiring your direct involvement.
That’s what eliminated Amanda’s support calls. Not because the software got easier — it didn’t. But because her team now has access to a system that carries her knowledge of how to navigate it.
How to Start
If you want to build something like this for your own practice:
Start by identifying where the support calls or escalations actually come from. What are the questions your team asks you repeatedly? What are the scenarios that require you to stop and think through carefully before answering?
Those are your training materials.
Record yourself explaining your reasoning on complex cases. Document the patterns you look for in review. Write out your criteria for the decisions that come up most often.
Then build the GPT from those materials.
You don’t need to be technical to do this. What you need is your own expertise, structured and written down. That’s the hard part — and it’s the part that makes the GPT useful rather than generic.
Amanda went through that process with me over several months. By the time tax season arrived, the system was substantial enough that the software support line never rang.
That’s the goal: not faster answers to generic questions, but the right answers to your specific ones.
I help professional service firms build custom AI systems trained on their specific expertise. My AI consulting and workshop programs are where we do this work. If you want to see what this looks like for your practice, that’s the place to start.
