The “Wait and See” Era Is Over
I’ve been watching the “AI is coming” headlines cycle for three years now. Every quarter, a new report says this is the year AI stops being a pilot and starts showing up in the numbers. And every quarter, the goalposts move.
So when the Wall Street Journal’s CFO Council Summit wrapped up last week in Palo Alto, I paid attention. Because this time, it wasn’t vague promises — it was public company CFOs putting specific numbers on the record.
What the CFOs Actually Said
ServiceNow’s CFO Gina Mastantuono said AI investments have produced $355 million in savings — with roughly $125 million flowing directly to the bottom line and the rest reinvested. The company earned $1.75 billion in 2025, up 23% year over year.
Shopify CFO Jeff Hoffmeister pointed to a memo the company sent a year or two ago: no new hires unless managers can prove AI can’t do the job. The result? Headcount has stayed flat for three years while annual revenue growth has held at around 30%.
Levi Strauss offered a different kind of story. A junior employee — not someone from IT — built an AI agent to handle wholesale order input. Something that used to take days now takes minutes, with higher accuracy. The workers who used to do that work? They’ve been redeployed to collecting receivables. That’s upskilling in practice.
Yes, these are public companies with an incentive to talk AI up to investors. But they also have a legal duty not to mislead. When a CFO puts a dollar figure in a public forum, there has to be something behind it.
The Research Context: The Productivity Paradox Is Real
At the same time these CFOs were speaking, a new working paper dropped from Duke University and the Federal Reserve Banks of Atlanta and Richmond. They surveyed nearly 750 CFOs and found companies self-reported an average 1.8% AI-driven productivity gain in 2025. That sounds solid.
But when researchers compared that to actual revenue data, the implied gains were smaller — suggesting a 12-month lag between AI investment and measurable financial results.
This isn’t surprising. It mirrors what economists call the productivity paradox — a pattern first described by Robert Solow in 1987, when computers were everywhere but invisible in productivity statistics. It took about a decade for computing investments to show up in GDP data. The gains were real. The timing just lagged the headlines.
AI appears to be following the same S-curve. Finance and high-skill services are leading all sectors in realized AI productivity gains right now — about 0.8% implied labor productivity growth versus 0.4% in manufacturing and lower-skill services. That’s consistent with finance being data-rich and process-heavy — exactly the kind of environment where AI gets traction first.
What This Actually Means for Accountants
There are two accounting issues here that deserve real attention as AI investments scale.
1. Capitalizing vs. Expensing AI Costs (ASC 350-40)
The accounting treatment of AI spend is becoming material. Under US GAAP, internal-use software rules under ASC 350-40 govern how AI spend gets classified — but many AI deployments sit in a gray zone.
Ongoing SaaS subscriptions and employee training are generally expensed. Custom model builds in the application development stage can be capitalized. The gray zone is everything in between: off-the-shelf LLMs with a custom configuration layer, prompt engineering labor costs, AI agent setup and configuration.
The key question to ask about any AI spend: is this a subscription, a custom build, or an implementation cost? Each answer leads to a different accounting treatment — and a potentially material difference in reported earnings.
2. Measuring and Disclosing AI ROI
ServiceNow stating “$355 million in AI savings” is a public claim. It has to be supportable, consistently defined, and not misleading. The SEC has been increasing its focus on AI-related disclosures in MD&A sections. Companies overstating AI benefits — intentionally or not — face real scrutiny risk.
On the last round of major earnings calls, executives said the word “AI” more than they said “earnings.” That’s a lot of AI talk — and a lot of exposure if the numbers don’t hold up.
The gap between what CFOs perceive and what revenue data shows isn’t fraud — it’s timing. But accountants and auditors need to make sure that public statements about AI productivity tie back to a defensible, consistently applied methodology. What’s the definition of “savings”? Is it headcount avoided, process time saved, or revenue attribution? How is it calculated period over period? Does it align with what the financial statements actually show?
Key Takeaways
- ServiceNow reported $355M in AI savings ($125M to the bottom line); Shopify held headcount flat while growing revenue ~30% annually for three years — real, on-record numbers from public company CFOs
- A Duke/Fed Reserve NBER study found CFOs self-report 1.8% AI productivity gains in 2025, but revenue data suggests a 12-month lag — consistent with how computing and other transformative technologies historically played out
- Finance is leading all sectors in realized AI productivity gains, making this directly relevant for accountants and finance professionals
- AI spend classification under ASC 350-40 (CapEx vs. OpEx) is an emerging material accounting issue as investments scale
- AI ROI disclosure accuracy is a growing SEC focus area — public claims about AI savings need to be supportable and consistently defined
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply