Finance software firm Esker is factoring artificial intelligence token consumption into its employee expense planning as the company seeks to better manage the technology’s increasingly unpredictable costs, CFO Scott McDermott said.
The company ramped up its use of AI this year, with costs running roughly four times over budget, McDermott said in an interview.
“As the models have gotten smarter, and the technology has become more addictive, companies’ reliance on it has grown and that’s leading to more and more costs,” he said.
The software provider has repeatedly exceeded its monthly AI token allowances in recent months as consumption has increased across the company and model providers have shifted toward usage-based pricing, according to McDermott, who became Esker’s CFO in January after more than two decades in corporate finance.
Tokens are the basic units of data processed by AI, with greater token consumption generally resulting in higher costs.
Esker’s experience with rising AI costs reflects a broader challenge.
A survey released by the company last week found that 72% of finance leaders spent more than planned on AI initiatives over the past year, while 65% of CFOs had difficulty connecting AI usage with specific business outcomes.
The cost challenge could become more pronounced as companies move from basic AI assistants to agentic systems. A recent survey from Futurum Research, completed in partnership with AI infrastructure provider QumulusAI, found that agentic AI can multiply token consumption per task by up to 100 times. The report said that can turn per-token pricing into a cost-control problem as usage scales.
As part of its effort to make AI spending more predictable, Esker is now treating AI spending as another component of an employee’s cost, alongside salary, bonuses and payroll taxes, McDermott said. The approach allows the company to track AI consumption across functions and use those patterns to improve budgeting for the technology.
Esker starts with actual usage data from its AI providers to estimate how much employees in each function are consuming. Finance can then calculate a run-rate AI cost per employee at the end of each month and apply assumptions for how usage and productivity will change the following year, McDermott said.
The approach gives Esker a more consistent view of its total employee costs, rather than relying on when AI expenses hit the profit and loss statement.
AI costs can vary significantly by role, with some employees consuming far more than others, according to McDermott.
“The finance and R&D teams — they run up overages like crazy,” he said.
McDermott said the company is also trying to determine whether higher AI spending is translating into returns such as productivity gains and revenue growth. Quantifying increased productivity can be relatively straightfoward, but revenue impacts remain more difficult to measure, he said.
“We’re certainly trying to strike a balance between investment and growth,” he said. “It’s probably one of the biggest challenges that I’ve faced in my career, but I’m pretty confident that we’re on the right track for 2027.”
