Training tax staff when AI does the prep

The skills tax professionals need to build a career haven’t changed. The path we used to build them is disappearing. 

When I was in practice, I used to say it took two to four years to bring a new staff member along to the point where I trusted them with a client return on their own. Those years were drafting, data entry, and compliance reps. AI now absorbs much of that work, and tax staff training hasn’t caught up. The firms that figure out how to build judgment faster will have a real talent advantage. The rest will end up with fast teams that can’t tell when the output is wrong. 

Firms see the gap. According to our recent AI Tax Research Solution Outlook report, 45% of regular users of AI for tax research strongly agreed that finding more efficient ways to train new team members was a reason their firm adopted it. No other motivation ranked higher. Yet only 28% of the firms that adopted AI-powered tax research reported using it to train new team members — the lowest of the 11 workflows the report tracked. Although training is the reason firms give for adopting it, most haven’t put it into practice yet. 

The repetitions that once built tax judgment are shrinking 

The old model was simple. Technical skills came from years of hands-on drafting and compliance reps, and client-facing work came after that foundation was in place. Learning advanced gradually on a fairly standard timeline. 

AI is replacing that learning curve, and firms are gaining efficiency faster than their training programs can adjust. Doing more with AI isn’t enough. The harder job is training people differently. 

Ryan Corcoran, a partner in Washington National Tax at RSM, does acknowledge that there are tradeoffs when AI takes the first pass: "You do lose some of those reps that you would get generally in just copying same as last year for the first two years of your career and then figuring out, oh, this is wrong and here's why." 

Seniors get faster reviews, and juniors lose the drafting reps that used to build their judgment. Judgement still has to come from somewhere, so we have to build it on purpose.  

Pattern recognition is not expertise 

Ryan described a type every tax leader has managed: the junior who has seen a particular book-to-tax adjustment so often and gotten the same review comment so many times that they simply know what to do. They just may not know why. 

"In an older model, that person gets rewarded, not because they know it, but because they know how to get to the outcome and can do it quickly," Ryan said. 

AI can get to the outcome in seconds, so speed is no longer what sets a good junior apart. What matters now is understanding why the outcome is right, so the question to ask junior staff changes. In Ryan's words, the question is, "Are you curious? Are you asking not just the reps, not just the pattern recognition, but why is it treated that way?" 

That makes curiosity worth looking for when you hire, and some firms are starting to screen for it alongside credentials. The three skills a trusted advisor needs 

AI is giving firms time back. In the same report, 84% of respondents agreed it saves time, and 46% said they’re putting that time toward higher-quality advice for clients. Advice takes more than technical knowledge, though, and that’s where most training stops. A trusted advisor needs three key skills: 

1. Technical proficiency to know whether what comes back from an AI tool actually makes sense 
2. Cognitive skill to apply that technical knowledge to a specific client's situation and circumstances 
3. Social skill to communicate advice in clear, plain words that clients can act on, whether the audience is an individual, an owner, or a board 

The old model assumed staff would pick up the second and third skills by accumulating years of practice. With fewer reps, firms have to train for judgment and client communication directly. Name the three skills in onboarding and build development around them. Leaving them incidental is how they get missed. 

Turn AI output into the training exercise 

Senior people are already doing more of the research themselves. According to Ryan, RSM's usage data shows that senior managers, senior directors, partners, and principals average more questions per user in its AI research tool than the levels below them. Research that once went to a junior now gets done at the top. 

If seniors keep all of that work, juniors never learn it. RSM's fix is to treat the AI-generated draft as a review and learning exercise, not output to sign off on. A senior can have the tool prepare a draft with sources cited, skim it and hand it to a junior to review. The junior's job is to find what's missing and ask why. 

Clients are creating a second training opportunity. More of them are sending in their own AI research, and reviewing it takes real time. Ryan treats it as a conversation and a chance to educate rather than correct. "I ask clients all the time, what did you prompt?" he said. Handing a junior that research and asking them to find the errors builds the same muscle. 

Many firms haven’t set up either exercise. In our recent webinar’s live poll, 50% of tax professionals said their firm hasn’t established a consistent approach to using AI-generated work product with junior staff. 

Find the people who will build the workflows 

Training gets easier when it is shared. Most staff will use AI, but a smaller group will build with it, and they can turn what they know into something everyone else can learn from. 

Ryan compares it to Excel. Everyone uses it, but only a subset become Power Query and Power Automate users. His observation about AI is the same: “Everybody’s going to use it, but there’s going to be that subset that is going to be building those workflows for their own teams that you got to go find them, you got to discover them.” 

At RSM, the people who stand out are thinking about systems and processes, not just their own efficiency. The opportunity is to turn what that minority builds into repeatable workflows for everyone else. A well-designed, easy-to-use workflow also builds in its own governance, because consistent use is itself a control. 

RSM learned a few things about finding these builders. Its rollout leaned on champions, use-case training, and peer demos, and Ryan said adoption of an AI research tool reached 91% within roughly two to three weeks. Asking staff to send in their prompts drew few submissions, though. People told RSM they weren’t prompt engineers and weren’t sure their approach would suit anyone else. 

So RSM shifted from requesting finished prompts to asking staff about their day-to-day work: What pain points do you have in your workflows? What issues are you running into? Where has AI already helped you? RSM then used those answers to tailor the use cases and build on them to create a library of useful prompts, tips and tricks, plus agents. 

Four steps to start 

You don’t need RSM’s scale to begin. These four steps work at any size:

1. Identify where AI is already changing junior staff workflows in your firm. 
2. Build one structured way to turn AI output into a learning exercise. A senior-drafts, junior-reviews routine is an easy first one. 
3. Name an internal AI champion, even informally.  
4. Revisit your onboarding and recruiting messaging. A move toward advisory work is a pitch to forward-thinking talent, and staff who feel developed are more likely to stay. 

Want to learn how to build advisory skills in an AI-enabled practice? Find out how an AI tax research solution can support staff as they build judgment at CPA.com/TaxAI. 

 
About the author
Brandon Allfrey, CPA, CGMA is the Senior Director of Tax
Transformation at CPA.com where he helps CPA firms modernize
tax compliance and expand advisory services by leading the
strategy and development of innovative tax technology solutions.