Key Takeaways

  • Accordion CFO Jon Apter notes that sponsor-backed companies are overspending on compute and software tokens, turning AI deployment into an immediate margin drag rather than an instant cost reduction.
  • Internal metrics inverted inside six months: firms that once celebrated employees with the highest token usage now audit those same users to halt unproductive API consumption.
  • Future M&A Quality of Earnings (QoE) analyses will require formal adjustments for experimental AI operational expenses to establish true baseline run-rate EBITDA at exit.
  • Back-office automation delivers measurable velocity when data governance is fixed first; Devin Mathews details a carveout that compressed its monthly close from six weeks to 36 hours.

The Token Audit: From Free Experimentation to Cost Control

Six months ago, private equity sponsors encouraged portfolio teams to test every generative tool on the market. High API usage was treated as a proxy for operational initiative. That phase ended as software bills arrived.

“It's definitely not a cost cutting exercise right now,” Apter explains. “Most people are spending more on AI than they should be, and so much of the job of the technology team at Accordion and my team is how do you actually do these things in a smart way.”

Unmonitored prompt queries, overlapping vendor subscriptions, and undisciplined cloud usage inflated operating expenses without creating revenue lift. Finance teams quickly realized that unguided adoption generates raw compute expense faster than it generates productivity.

“Six months ago we were celebrating people who had the highest token usage,” Apter says. “Now if you have the highest token usage in the organization you're getting a phone call saying, 'What were you doing?'”

Normalizing AI Operational Expense on Exit

When sponsor-backed companies prepare for an exit, sell-side Quality of Earnings reports routinely adjust for one-off professional fees, severance, and standalone migration costs. Apter expects AI operational spend to become the next standard line item subject to QoE normalization.

Because many companies are currently paying for redundant proof-of-concept tests, overlapping seats, and excessive query usage, their current profit margins understate normalized operating profitability. Conversely, targets claiming massive future labor savings without accounting for recurring infrastructure bills overstate their run-rate earnings.

“I imagine when all these companies go to exit, there will be some normalization around AI spend in these quality of earnings reports,” Apter notes. “People are trying to figure out what is a true stable AI usage that is appropriate.”

Sponsors seeking clean exits must separate speculative developer exploration from the permanent compute cost required to serve active customers.

Compressing Finance Cycles from Six Weeks to 36 Hours

The real operational gains in private equity finance functions come from structured data pipelines rather than standalone chatbot wrappers. When finance infrastructure is properly architected, cycle times collapse without expanding team headcount.

Mathews points to a direct operational proof point from a standalone carveout that inherited minimal legacy software infrastructure. The operational path required progressive elimination of manual reconciliations:

“We went from literally closing the books in six weeks because it was a pure carveout without a lot of systems that came with it to then four weeks, three weeks, two weeks,” Mathews notes. “He closed August in 36 hours.”

Achieving that velocity required clean data pipelines and targeted automation rather than unguided software spend. As Apter emphasizes, “Getting the right governance around your data and the right AI usage put in place, that's what the consulting model is going to look like in the future.”

Why It Matters

Private equity buyers are moving past narrative-driven tech premiums and demanding audited proof of margin expansion. As sponsors face elongated hold periods, uncontrolled software experiments directly depress bridge EBITDA. The firms capturing real enterprise value are treating AI as a disciplined balance-sheet allocation with strict usage controls, clear QoE normalization logic, and measurable back-office cycle compression.