Want unlimited access to top ideas and insights?
Every week there's a new headline, a new tool or a new platform that claims artificial intelligence will "transform" our profession. True. AI can already read our source documents, draft our returns, write our client emails, summarize our engagement letters, and generate advisory memos while we sleep. If you're an owner or managing partner at a CPA firm, the pace of change doesn't feel like progress. It feels like a fire hose aimed at your face.
And the natural response is paralysis. When everything seems to be changing all at once, it's hard to know where to start. When every vendor promises the future, how do you separate signal from noise? When your partners are split between "we need to move now" and "let's wait and see," how do you build a strategy that doesn't become obsolete before you finish implementing it?
Here's the good news: Those questions were answered over 25 years ago by someone who had nothing to do with accounting. Jeff Bezos.
Start with what won't change
When Bezos built Amazon, he wasn't thinking about what would change in society. He was thinking about what wouldn't change. He knew no customer would ever say, "I wish your prices were higher," or "I wish you'd deliver more slowly." So, he built his entire business around speed, selection and price. He let everything else evolve around those three constants.
CPA firms need to ask the same question: What will our clients always want, regardless of what technology does?
The answer isn't complicated. No client will ever say, "I wish my CPA understood my situation less." No client will ever say, "I want slower turnaround, less accurate work, or an advisor who doesn't pick up the phone when the IRS sends me a letter." Clients will always want trustworthy judgment, a human being who makes the process feel easy, and the confidence that someone competent is coordinating all the moving pieces.
Those are your constants. The question is whether your firm is actually built around those constants, or around something else entirely.
The problem with the organizational pyramid
For decades, CPA firms have been built on the same organizational hierarchy model. It's shaped like a pyramid. Partners sit at the narrow top, billing at the highest rates and capturing the margin. Below the partners sits a widening base of managers, seniors and staff that do the production work: preparing returns, building workpapers, ticking and tying, populating schedules. The wider the base, the more revenue the firm generates. The business model is, at its core, labor arbitrage: hire people at one rate, bill them at a higher one and extract the spread.
This model worked for a long time because the work of accounting firms required bodies. Someone had to key the data manually from a stack of K-1s. Someone had to populate every line of the return. Someone had to reconcile the trial balance. There was no way around it. The process was very labor-intensive.
But notice what the traditional pyramid structure is built around: process. It's not built around value. The wide base at the bottom of the pyramid doesn't exist because clients want it. It exists because the work demanded it. Clients don't care how many staff associates touch their return. They care that the return is done right, that it is done on time, and that someone they trust reviews it and can explain to them what it means.
The pyramid, in other words, is built around something that will change. And it's changing right now.
What AI actually does (and doesn't do)
Let's be realistic about what's happening with AI in tax and accounting. I bring this up because the conversation tends to oscillate between two unhelpful poles: breathless hype and dismissive skepticism. The reality is more nuanced and, frankly, more interesting.
I have found AI to be exceptionally good at the work that lives in the middle of a typical engagement: ingesting a document, reading a W-2, a 1099 or a K-1 and extracting the data. Data population, compliance checks and draft preparation are rapidly getting there. Take the initial review that a second-year associate used to spend three hours on. A model can flag the same discrepancies in seconds. These are pattern-recognition tasks performed on structured data, and they're precisely what AI was built for.
What AI is not good at, and won't be good at for a meaningful period, are the things that sit at either end of the engagement. On one side you have the nuanced judgment calls that partners make:
- Should the client make an 83(b) election?
- If so, how should they structure the entity to optimize for both tax efficiency and liability protection?
- What's the right approach when the tax code creates a genuine tension between short-term savings and long-term positioning?
These decisions require contextual understanding, risk tolerance assessment, and the kind of professional judgment that comes from years of seeing how decisions play out across dozens of similar situations.
AI is also not equipped to handle the other side of the equation: human relationships. It doesn't know how to handle a client who calls in a panic because they got a scary letter from the IRS. It doesn't know how to help a business owner who needs someone to walk them through their options after a bad quarter. Same goes for the family going through a divorce who needs a steady hand coordinating between attorneys, financial planners and the firm. No model handles this. And clients have never wanted it to.
Introducing the hourglass
If AI compresses the middle — the process-heavy, labor-intensive production layer — but the top and bottom of the organization remain irreplaceable, then the pyramid is the wrong shape for the modern CPA firm. The right shape is an hourglass.
The top of the hourglass is the Advice Layer. This is where your partners, senior advisors and experienced managers live. They review the output, make the calls and deliver the advice. They're the ones who look at the complete picture and tell the client: "Here's what you should do and here's why." This layer is defined by expertise, pattern recognition and professional judgment — things that require years of accumulated experience and are extraordinarily difficult to automate. This is the trust layer.
The bottom of the hourglass is the Relationship Layer. This is your client-facing team. The client service associates, the people who answer the phone, collect the documents, walk clients through what's needed and explain what happened after the work is done. These are the people who make the process feel human. They know the client's name, remember they're always late with their K-1s, and follow up without being asked. This layer is defined by empathy, communication and service, skills that are fundamentally human and that clients will always value.
The narrow middle of the hourglass is the Processing Layer. This is where AI lives. Document ingestion, data extraction, return population, compliance checks, workpaper generation, initial review, etc. It's all the work that used to require a wide base of staff and seniors grinding through files. The hourglass pinches here in the middle because you simply don't need the same number of people doing processing work anymore. The work still gets done. It just gets done by models, not by a bullpen of second-year associates.
Build around what won't change
The technology in the middle will keep evolving. Don't try to stop it. When you build your firm around the things that don't change, you stop worrying about what the next tool does and start asking the only question that matters: "How do we use it to deliver more of what our clients have always wanted?" That strategy worked out pretty well for Jeff Bezos and Amazon. Why not you?
In Part 2, we'll explore what the hourglass means in practice: how it changes the way you hire, how you develop talent and where to invest first.
What is your firm doing to adapt to the new accounting firm structure? I'd love to hear from you.
