A dollar or two more per employee sounds like an easy price for AI. Whether it is depends on what a company currently spends answering routine employee questions, a number the buyer has to know before the fee makes sense.
Tushneem Dharmagadda, CEO and founder of workforce experience platform HubEngage, says he has asked customers exactly that. When HubEngage asked whether they would pay a dollar or two more per employee for added AI, the answer was no, which he attributes to a lack of demonstrated return.
He sees the same pattern across the market. Vendors first sold AI as a separate add-on, and “a lot of customers did not see enough value to pay extra, so now it is being bundled into the core product,” he says. HubEngage is built for frontline and distributed workforces, the employees who often lack a desk or a corporate email address.
The price is moving
Bundling is not the end of the pricing story. A recent Harvard Business Review analysis by analyst Stacia Garr says software vendors have been absorbing the cost of GPUs, inference and tokens, labeling premium AI agent features unmetered, complimentary or included. As they move to usage-based charges, she writes, organizations are replacing fixed labor costs with variable AI consumption costs, and she advises modeling those costs now.
That sits awkwardly with a per-employee fee. A flat charge per employee is a fixed number, while the cost of serving AI requests rises and falls with use.
Cheaper tokens, costlier tasks
The usual counterargument is that AI keeps getting cheaper, and unit prices support it. Stanford’s AI Index 2025 reported that the cost of querying a model at a constant performance level fell from $20.00 per million tokens in November 2022 to $0.07 in October 2024, a drop of more than 280-fold.
A price per token says little about the price of a task. A preprint co-authored by Erik Brynjolfsson and Alex Pentland found that agentic coding tasks consume more than 1,000 times the tokens of code chat or code reasoning. The study tested coding, not HR questions, but it shows how hard a multi-step agent can be to budget for.
What the savings have to cover
Buyers weigh that cost against the time AI saves. HubEngage’s own ROI calculator for automating routine HR and finance questions uses a simple structure. Savings equal the cost of staff time spent on routine questions multiplied by the share the AI handles, and ROI is those savings minus the AI cost, divided by the AI cost.
The break-even point can be worked out on the back of an envelope. At the US Bureau of Labor Statistics median wage for human resources specialists, $36.51 an hour as of May 2025, a routine question that takes five minutes to answer costs about $3.04 in staff time. If the fee were charged monthly, a one-dollar fee is covered once the AI fully absorbs about one question for every three employees each month. A two-dollar fee needs about two questions for every three employees. If answers take ten minutes, both thresholds halve.
Company size does not change those thresholds. Whether a company clears them depends on how many routine questions each employee actually sends, a number a buyer has to measure before the fee can be judged.
The calculator labels its headline savings as time cost avoided and tracks the share of saved time converted to strategic work as a separate figure. Dharmagadda makes a related point: “time saved does not automatically equal payroll eliminated.”
The practical question for buyers is whether an AI add-on pays for itself at their own volumes, at a price that is no longer subsidized. Employers modeling that, as Garr advises, may find the answer is yes for some workloads and not yet for others.
