Corporate America has gone all-in on artificial intelligence — but when it comes to actually measuring the payoff, most companies are flying blind.
That’s the warning from Bob Michaels, a partner at CrossCountry Consulting who advises corporate finance chiefs on exactly this problem: how do you justify millions in AI spending when you can’t clearly show it’s making money?
“The spend is just escalating almost out of control,” Michaels said in a July interview. Thanks to accounting rules that don’t let companies spread out software costs over time, much of that AI spending is landing directly on current earnings — no delay, no cushion.
Bought First, Asked Questions Later
The core problem: companies rushed to buy AI tools before figuring out how to measure their value. Now boards want answers, and finance teams are stuck trying to retrofit return on investment (ROI) calculations onto money that’s already been spent.
“Most companies are going to have a hard time answering that clearly,” Michaels said.
Some are pointing to softer wins — faster financial closes, quicker reconciliations, hours saved chewing through purchase agreements. But there’s a catch: saving time doesn’t automatically save money. If a freed-up employee just gets reassigned to other work, the bottom line never actually moves.
“The ability to measure that ROI, it has to be on time saved, closing process shortened, more efficient outputs,” Michaels said.
When AI Gets It Wrong, Someone Pays
If gains are hard to measure, losses aren’t. AI-detection firm GPTZero found fake footnotes and fabricated claims in four PwC reports — and warned that popular AI chatbots were already repeating the false information, spreading it further simply because it carried a Big Four name.
“Such errors betray a lack of care and human verification that many come to expect from work produced by the Big Four,” said GPTZero co-founder Alex Cui.
The damage isn’t just reputational. Deloitte Australia had to refund nearly AU$98,000 — about 20% of the total contract value — after AI-generated errors were discovered in a government report last year. GPTZero’s review found 19 suspected hallucinations buried in the 234-page document. Cui noted that similar investigations into EY and KPMG previously forced report retractions, burning staff time and straining client trust.
The Accounting Trap
With costs piling up, some companies may be tempted to strip AI expenses out of adjusted earnings to make the numbers look better. Michaels says that’s a mistake — this spending isn’t going anywhere, and investors won’t buy the “one-time cost” excuse forever.
“The cost is real, it’s hitting earnings, and the ROI is hard to prove out,” he said.
It’s not just tech giants feeling the squeeze. Even mid-sized companies — those earning a few hundred million to $1 billion in revenue — are routinely spending $5 million to $10 million a year on AI tools and infrastructure, Michaels said.
Adoption Is Outpacing Oversight
Perhaps the bigger issue: employees are using AI daily with little training and no consistent rules for how much to trust it.
“That gap means, in practice, there’s a number of different potential issues,” Michaels said, noting some workers accept AI output without question while others don’t — with no company-wide standard for what needs to be double-checked.
That’s why, he argues, AI won’t be emptying out finance departments anytime soon. Someone still has to fact-check the machine. If anything, the oversight gap could create a new kind of job: people whose entire role is challenging what the AI produces.
The Fix
Michaels’ advice is straightforward, if not easy to execute: treat AI spending like any other major investment. Know what you’re buying, define the expected return, and build in safeguards before the money is spent — not after the damage is done.
“Treat the spend like any other typical capital allocation,” he said. “Have the discipline and the controls upfront.”
Take your tax and accounting research to the next level with Checkpoint Edge and CoCounsel. Get instant access to AI-assisted research, expert-approved answers, and cutting-edge tools like Advisory Maps and State Charts. Try it today and transform the way you work! Subscribe now and discover a smarter way to find answers.
