AI in Accounting. Do We Use It? Carefully.

AI is already part of your business.

You may not have signed up for ChatGPT. You may not be asking it questions every day. You may not even particularly like it.

But if you use modern email, accounting software, search engines, client portals, or other business technology, there’s a good chance some form of AI is already built into the tools you use.

So the question probably isn’t whether businesses should use AI anymore.

The better question is:

How do you use it without letting it do your thinking for you?

That came up in Episode 7 of the Planting Seeds Podcast when we answered a question we’ve started getting more often:

How does Harvest Advisory Group use AI?

Nick’s answer was simple.

Carefully.

And I think that’s probably the right answer for most businesses.

Ignoring It Doesn’t Make It Go Away

From the beginning, we’ve taken the position that our team needs to understand these tools.

Not because AI should replace people.

Not because every new piece of technology automatically makes a business better.

But because you can’t recognize the limitations of something you don’t understand.

Our team has spent time testing different uses, talking about what works, figuring out what doesn’t, and establishing some boundaries around where AI belongs.

Sometimes the uses are pretty ordinary.

I’ve used it to help organize a route when I had several places to visit in one day.

I use it regularly to organize thoughts I’ve dictated out loud.

Nick uses it to help gather research and point him toward sources he can review.

We’ve used it to look at a plan and ask:

What are we missing?

That may be one of its most useful applications.

Not replacing the person making the decision.

Helping that person look at the decision from another angle.

AI Is Very Good at Getting You to the Information Faster

Research is a good example.

You could spend a couple of hours searching for information, opening different pages, sorting through results, and trying to find the original source.

Or you can ask an AI tool to help gather that information into one place.

That can save a tremendous amount of time.

But saving time and knowing the answer are not the same thing.

That distinction matters.

If AI gives us information about a tax issue, for example, the next step isn’t:

“Well, ChatGPT said it, so we're done.”

The next step is checking where that information came from.

What does the IRS actually say?

What does the source actually say?

Does the source support the answer we were given?

AI can shorten the path.

It does not remove the responsibility to verify where that path leads.

Check Your Work

Nick used an example on the podcast that I liked.

Remember being in school and having a math teacher tell you to work the problem backward to check your answer?

Everyone hated doing it.

You already solved the problem.

Now you have to solve it again?

But there was a reason for it.

You were verifying the result instead of simply trusting the first answer you got.

AI should be treated the same way.

Give it the problem.

Look at the answer.

Then check the work.

That becomes especially important when the information affects somebody’s business, taxes, finances, or another important decision.

Because AI can be confidently wrong.

And the internet it learns from contains plenty of bad information too.

Give It Guardrails

One thing we’ve learned quickly is that broad questions tend to produce broad answers.

Better instructions produce better results.

If Nick is researching something tax-related, he may tell the tool specifically where he wants the information to come from.

Look at IRS guidance.

Look at applicable case law.

Bring back sources.

That is very different from simply asking the internet, “What should I do?”

The same idea applies outside of tax research.

If you’re asking AI to help think through a business decision, give it context.

What kind of business are you running?

What problem are you actually trying to solve?

What limitations exist?

What assumptions are you already making?

And one of my favorites:

Tell me what’s wrong with this idea.

AI is very agreeable if you let it be.

Sometimes you need to specifically ask it to disagree with you.

Sometimes It Helps Us Communicate Better

One of the more interesting uses we talked about had nothing to do with numbers.

Nick had received questions from a client and realized there was some kind of communication gap.

He understood what he was trying to say.

The client understood what they were trying to ask.

But those two things weren’t quite meeting in the middle.

So he used AI to help look at the conversation and identify where the misunderstanding might be.

That helped him approach the explanation differently.

I use it similarly.

Sometimes I'll explain something the way that makes perfect sense in my own head and then ask:

What am I not explaining clearly?

Or:

What does it sound like this person is actually asking me?

That's useful.

We all have blind spots in the way we communicate.

Sometimes another perspective helps you see them.

But There Have to Be Boundaries

This matters particularly for businesses that handle sensitive information.

Not every AI tool belongs in every situation.

Not every piece of information belongs inside an AI prompt.

And “this would be faster” is not a good enough reason to ignore privacy or security.

We have internal rules around how our team can use AI, what platforms we use, and the kinds of information that can and cannot be shared.

We also don't rely on AI to do the accounting for us.

It can assist with research.

It can organize information.

It can help us communicate.

It can help us pressure-test an idea.

But there still needs to be a person on the other side who understands the work well enough to know when something isn't right.

That part matters.

The Skill Isn’t Asking AI Questions

Anyone can type something into a prompt box.

The useful skill is knowing what to do with the answer.

Can you recognize a bad source?

Can you spot something that doesn’t make sense?

Do you know enough about the subject to ask a better follow-up question?

Can you separate a useful suggestion from a confident-sounding bad idea?

That’s why I don’t think the answer for businesses is avoiding AI altogether.

It’s becoming familiar enough with it to use it responsibly.

Technology should make good people more effective.

It shouldn't remove judgment from the process.

Use the Tool. Keep the Responsibility.

There will probably be things AI can do a year from now that seem ridiculous today.

That doesn’t mean every business needs to chase every new tool that comes along.

It also doesn’t mean pretending the technology isn't here.

Find the places where it genuinely saves time.

Put boundaries around it.

Protect sensitive information.

Verify important answers.

And keep a human involved where judgment matters.

That’s basically how we approach AI at Harvest Advisory Group.

Not as something that replaces the work.

As something that can help us do the work better.

As long as we keep checking it.

Watch the Conversation

In Episode 7 of the Planting Seeds Podcast, Nick and I talk about how Harvest Advisory Group uses AI, owner draws and distributions, the numbers business owners should know, tax myths, paying yourself as an owner, fear around sales opportunities, and quite a few other things along the way.

At Harvest Advisory Group, technology is useful when it creates better information, clearer communication, and more time for the conversations that actually matter.

The tool may change.

The responsibility to make a good decision doesn’t.