Esker Adds AI Token Costs to Employee Expense Planning

Esker is folding AI token costs into employee expense planning after its AI spending ran roughly four times over budget this year. CFO Scott McDermott attributed the overrun to increased use and model providers shifting toward usage-based pricing, and said the company repeatedly exceeded monthly token allowances in recent months. The account of Esker’s spending and planning approach shows how a finance team can connect AI usage to workforce forecasts, even when the business return is harder to pin down.

From provider usage to an employee cost estimate

Esker starts with actual usage data from its AI providers to estimate consumption across functions. At each month’s end, finance calculates a run-rate AI cost per employee, then uses assumptions about future usage and productivity to forecast the following year. The company includes AI spending alongside salary, bonuses and payroll taxes. McDermott said this gives Esker a more consistent view of employee costs than relying on when AI expenses appear on the profit and loss statement.

That distinction matters because a reported expense can reflect when a charge is booked, not necessarily which parts of the workforce are driving consumption. Esker’s approach uses usage patterns to make AI a visible part of workforce cost planning. The account describes estimates by function and a per-employee run rate, but does not specify a tracking tool or how individual activity is attributed.

Higher usage does not automatically mean higher returns

AI consumption varies by role at Esker. McDermott singled out finance and R&D as teams that run up substantial overages. That makes a single company-wide allowance a poor description of where the spending occurs, though the account provides no detailed cost breakdown by department. Esker is also trying to assess whether the added spending produces productivity gains or revenue growth. McDermott said productivity can be more straightforward to quantify; revenue impact is harder.

The wider figures in the report point to the same measurement challenge: a survey released by Esker found that 72% of finance leaders spent more than planned on AI initiatives over the past year, while 65% of CFOs struggled to connect AI use with specific business outcomes. Those figures describe survey responses, not a measured rate of budget overrun across all companies. The survey findings and McDermott’s comments underline why tracking consumption and demonstrating value are separate tasks.

Agentic workloads could make forecasts less stable

The report also cites a Futurum Research survey, conducted with AI infrastructure provider QumulusAI, which found agentic AI can multiply token consumption per task by up to 100 times. That is a reported survey finding about possible per-task consumption, not a guaranteed increase in every deployment’s bill. The cited research frames agentic usage as a potential cost-control challenge as use scales.

Esker’s method makes usage easier to include in a workforce forecast, but it cannot by itself establish whether the spending pays off. That judgment still depends on connecting the cost estimates to outcomes the company can measure, especially when revenue effects are less clear.