This summer, KPMG brought nearly 1,000 audit interns to its Lakehouse training campus outside Orlando and spent a week teaching them how to think critically. The program swapped some of the routine bookkeeping drills for exercises built on judgment calls, teamwork under pressure and a fraud investigation simulation the interns solved as a team.
A Big Four firm needs to devote intern training to judgment because AI is absorbing the work that used to build it. KPMG plans to begin a pilot this year, expand use in 2027 and shift most routine checking to AI by 2029. The tie-outs and confirmations that used to fill a first-year’s calendar also taught them how a client’s numbers fit together.
Two other stories caught my attention the same week. Alex Romero, director of CPA Pipeline at the AICPA, shared on an episode of The INSIDE Public Accounting Podcast that students still want to join the profession. But once inside, they struggle to see the path forward. He pointed to communication, judgment and relationships as essential skills right now. And placement data covering more than 400 public accounting hires showed pay growth stalling at Senior and jumping at Manager.
Together, these stories point to one conclusion: The market pays a premium when a person stops preparing work and starts making the calls. Meanwhile, we’re automating the work that prepared people for that moment. And now we’re paying to teach it back.
We measured what was easy to measure
I spend a lot of time helping firms adopt AI. We evaluate most automation use cases based on cycle time, error rates and hours saved. Those metrics make sense. A bot that ties out a bank reconciliation in seconds beats a first-year who takes two hours, misses a transposition error and plugs the difference to “Ask My Accountant.”
But those two hours also (eventually) produce a team member who understands how cash moves through a business and recognizes when a reconciling item looks off. Simulations don’t capture that value because it stems from inefficiency.
So before we rip out the tasks AI can replace, we need to ask what else those tasks delivered and how we plan to replace it.
A loop is only as good as the human in it
Every AI governance conversation I join involves “human in the loop.” It’s the right principle, but it assumes the human knows what a good answer looks like. When the reviewer never built the pattern recognition that comes from doing the work, the loop turns into a rubber stamp.
Chess offers a useful example, believe it or not. After IBM’s Deep Blue beat Garry Kasparov in 1997, Kasparov began exploring what happens when humans and computers play as a team. In a 2005 freestyle tournament, human+machine teams dominated even the strongest computers. The winners were a pair of amateur players running three computers. They beat grandmasters by coaching their machines to look deeply into positions.
Angus Fletcher pushes the idea further in Primal Intelligence. He argues that in volatile, uncertain environments, intuition and imagination outperform logic and data, and humans still win at adapting and spotting exceptions.
That’s where client work is. Being able to spot a going-concern risk or an aggressive revenue position comes from someone who’s seen enough to sense when the numbers feel wrong.
AI multiplies the value of that person. But your firm still has to develop them.
Rebuilding the reps
Before you approve the next use case, consider where your first-years get their reps right now. If the answer is a new line in the Learning & Development budget, you’ve funded a training program. That helps. A redesigned first two years builds a bench, and five years from now, your bench determines what kind of work your firm can take on.
Here’s where to start:
- Add a development column to every automation business case. For each task you automate, identify the skill it used to build and how team members will build it going forward. If you don’t have an answer, pause the rollout until you do.
- Rebuild early work around judgment. Have first-years review AI output against source documents, run down the exceptions the tools flag, and explain what they found to a senior. Let them draft the client email about the issue. Each task forces a decision.
- Pull decision-making earlier. Give associates and seniors real calls to make with a reviewer beside them, such as leading a controls walkthrough with the client or owning a small client relationship.
- Make the path visible. Map the skills each role requires and show your team how today’s assignments build toward the next level. People stay when they can see where they’re going.
Technology is the easy part. The hard part is deciding what your people are for. Answer that question first, and choosing the right AI use case gets much easier.
ABOUT THE AUTHOR:
Marc Staut is a shareholder and chief innovation and technology officer at Boomer Consulting Inc., where he helps CPA firms align technology strategy with firm-wide innovation and growth. With a degree in psychology and a career that spans nearly every role in IT, Marc brings a people-first perspective to the tech world, bridging the gap between systems and strategy with empathy, clarity and contagious enthusiasm.
Sign in to get access to this free resource, and all of our whitepapers and reports.
Download this content today!
Register Now Already registered? Click here to Log In
![Marc_Staut_(grey_BG)[1]](https://www.cpapracticeadvisor.com/wp-content/uploads/2022/04/Marc_Staut__grey_BG__1_.5edff0550f4dd.png)