One of the biggest reasons accountants tell me they’re nervous about AI is pretty reasonable: You can’t trust everything it tells you. Correct!
However, that doesn’t mean we don’t use AI. I think we just have to get better at questioning it.
I was reviewing a tax return that an AI created the other day when I noticed something didn’t look right. There was a small discrepancy I wouldn’t have caught if I didn’t have decades of experience reviewing returns.
Here’s the funny part. I used the AI agent to help me find its own mistake.
I didn’t necessarily know what was wrong at first. I knew enough to see that something didn’t make sense. Then I started asking the agent questions.
What looks off?
Why doesn’t this tie?
Where is this number coming from?
What should have happened here?
Eventually, we got to the problem.
Trust Doesn’t Have to Be Binary
It made me think about the way accountants talk about trust and AI. We tend to make it very binary. Either we trust AI and let it do the work, or we don’t trust it and shouldn’t use it.
There’s a lot of space in between.
Think about a junior accountant. You wouldn’t give a brand-new staff accountant a return and say, “Great, you did it! Send it to the client.” You’d review their work. You’d ask questions. You’d teach them what they missed and why. Over time, as they demonstrated that they knew what they were doing, the way you reviewed their work would change.
I think we need to approach AI with some of that same skepticism.
The agent is kind of like a newbie. It can do an incredible amount of work, sometimes ridiculously fast, but that doesn’t mean you blindly trust the output. You still need:
A review process
Controls
Someone who knows what good looks like
And, most importantly, judgment
But Judgment Is Changing
We’ve always assumed that the person reviewing accounting work needs to know how to do the work themselves. And there’s obviously truth to that. You need accounting knowledge to know when something doesn’t make sense.
But preparing something and reviewing something aren’t exactly the same skill.
I can write something. That doesn’t automatically make me a great editor of someone else’s writing. The same thing happens in accounting. Knowing how to prepare a return and knowing how to look at a completed return, identify what seems wrong and ask the questions that uncover the problem are related skills, but they aren’t identical.
AI makes that distinction much more important.
The skill might not always be knowing exactly what’s wrong the second you look at something. It might be recognizing that something is wrong and knowing how to interrogate the AI until you understand why.
That can be taught.
AI Doesn’t Automatically Understand Your Job
We need to remember that most AI tools accountants are using weren’t built by accountants who have spent 15 years closing books, preparing returns or sitting across from clients. They were built by engineers.
That doesn’t make them bad tools, but software doesn’t automatically understand your job just because it can perform pieces of it.
So now we have to figure out the layer between the technology and the final work product:
Who teaches the AI what good looks like?
Who reviews what it produces?
What does the preparer need to know?
What does the manager need to know?
What questions should each person be asking before something goes out the door?
Those are new skills, and we need to start training people on them.
Humans Get Things Wrong, Too
Because I worry that when accountants hear, “AI gets things wrong,” we use that as proof that AI can’t be trusted and go right back to doing everything ourselves.
Humans get things wrong too. That’s why we have review procedures in the first place.
We can’t expect AI to be perfect, but we need to make sure that our people and our processes can catch its mistakes.
Put Your Skepticism to Work
So question your AI. Push back on it. Ask it to show its work. Tell it something doesn’t make sense even when you can’t immediately articulate why. Make it explain itself. Ask the same question a different way.
When I tell you to implement AI in your firms, I’m not saying to trust it blindly. In fact, one of accountants’ greatest advantages could be our natural skepticism of new technology.
If we can learn to channel that skepticism into asking better questions, challenging the output and knowing what needs a second look, it could actually put us ahead.
