The meter is running

Earlier this year, I ended a presentation to a room full of HR leaders from many CAHRS partner companies by asking them how much they thought it cost me to produce the slide deck I had just presented to them. At first, it sounded like a rhetorical question and I was met with an uneasy silence (albeit expected). I assured them that I didn’t quite know the answer either but that I had a ballpark figure in mind. The cost to create my deck was a combination of my time (paid for by my salary), some AI that I pay a monthly subscription cost for, and some small fraction of the cost of my laptop. Then I asked a hypothetical back to them: What does it cost when an employee at their company gets an AI system to produce a presentation? This is a task increasingly being delegated to AI, and the question, of course, was met by the in-unison chorus of “it depends.”

Depending on which model does the work, how many drafts it takes, and how many token-heavy elements (like images) get generated along the way, the same deck can cost four cents or four dollars. That is a 100× spread on a single, ordinary knowledge-work task. Leaders across organizations are generally convinced that AI is already reshaping the future of work, but don’t know how to price this new way of doing work.

For the sake of having a way to talk about pricing the cost of AI in workforce planning and job design, let’s call it workforce tokenomics. This term “tokenomics” originally came from the supply mechanics around crypto tokens, but is increasingly being adapted to describe how organizations weigh AI consumption against the cost of a human doing the same work and connect it to business outcomes.

In the coming years, there are going to be a lot of people with different perspectives and pricing models for this. But from the HR context, we need to recognize that something structural has changed. Traditional workforce budgets commit most knowledge-work costs up front: salary, benefits, equipment, and software. Additional tasks still consume scarce employee time and coordination, but they rarely appear as a newly metered line item. But now, enterprise AI that is capable of performing real knowledge work is breaking that model by commoditizing units of cognition and billing them by the token, a shift visible in recent reporting on enterprise AI spending at scale.

For CHROs, and even more for the HR business partners (HRBPs) and unit HR leaders who will inherit the redesigned jobs, the decisions this forces aren’t ones for IT. Questions like, “Which tasks can we reliably delegate to AI, which workflows do we rebuild around it, and what does the job become afterward?” are job analysis, work design, and workforce planning. These are the most consequential org design questions of the next five years, and should be the responsibility of strategic HR whether HR is ready or not. Especially now that enterprises are moving past the AI honeymoon phase, CFOs are critically evaluating token spend and asking who is approving it.

There are many thought pieces around this concept, but for the rest of this piece let’s try to use a specific example around pricing the AI augmentation of an HRBP. Leaders need a way to price the augmentation before redesigning jobs. The costing method I call the Total Cost of Augmentation takes a look at costs, benefits, and token sourcing to quantitatively inform job and workforce design. This method comes from working sessions with CAHRS partner companies over the past year, conversations with HR leadership teams of large global firms, and an embedded field study with one CAHRS partner where we are tying employee AI adoption, usage data, and token pricing to actual productivity. The honest answer, so far, is that there are many more variables at play in pricing AI that make human labor still the core focus of workforce design. AI, when deployed effectively, can create positive yield and labor surplus, but return on investment (ROI) only exists if that surplus is captured by the organization. It’s really not about what the workforce tokenomics analysis shows for your organization, but that you are confident in how you do the analysis.

Knowledge work now has variable costs

Consider what a knowledge worker has cost until now. Take a worker paid about $80,600 plus benefits (about 43% on top of wagesA) and that’s roughly $115,000 a year in loaded compensation. Add the laptop with the enterprise software seats, and across a large enterprise that IT layer averages out to roughly 2%C of the loaded cost of the employee. Everything after that is committed up front rather than metered. The five-hundredth task an analyst performs this year never shows up on an invoice, but I wouldn’t call that work marginally “free.” It consumes that analyst’s hours and usually some of a colleague’s, and the people who reliably absorb more work tend to cost more over time in raises, bonuses, and promotions. What’s new is not that doing tasks has a cost, but that enterprise AI attaches a visible, attributable price to the task itself.

Now put that same worker in front of enterprise AI. Published usage data from one major lab puts its average enterprise developer at about $13 a day, with nine in 10 under $30. For a typical enterprise knowledge worker I use a deliberately generous planning figure of $30 a dayC in consumption, which in this example is about 6% of loaded cost and nearly quadruples this worker’s technology line. While this may seem a manageable expense, we have to understand that the distribution of AI spend per employee is not a bell curve, but is very skewed. The power users, the people who genuinely “get it” and are already automating whole workflows for themselves, can run $100 to $200 a dayC in a heavy-use scenario as work shifts from simple prompting to multistep automation; enterprise token-cost analysis shows how workflow shape drives the bill. One large company I spoke with set a central annual token budget and burned through it in four months. Nothing had gone wrong, per se, but nobody was watching the meter.

I’ve started calling this the predictable surprise of AI pricing. The phrase belongs to Max Bazerman and Michael Watkins, whose 2004 book Predictable Surprises is about disasters leaders could have seen coming. One HR executive used this phrase with me to describe what had just happened to their team. During a single, week-long sprint to build two HR agents to automate transactional HR tasks, they exhausted their token allocation before the agents were even finished. Most companies don’t have mature pricing models per employee/task yet because adoption is still shallow. As usage scales, we have to ask ourselves which function pays, which business unit deploys, and how we measure if the spend was worth it.

These pricing questions are about labor, and I argue that even digital labor belongs to HR.

Start with the value of a human minute

Your organization already invests heavily in every hour of its people’s expertise, and it’s more important than ever to first know what that hour is worth. In order to illustrate the pricing of AI and have conversations around TCA, we need to know the fully loaded value of a minute of human expertise for a given role. Of course, this is a number that has no precise value in practice, because the real world is complex. As the familiar maxim “all models are wrong, but some are useful” reminds us, we need to suspend our disbelief to come up with some plausible numbers such that we have points to talk about, debate, and maybe get better at measuring within our companies.

For an HR business partner in a major metro area, blending federal wage data for HR specialists and HR managers and applying the standard benefits package, you land somewhere around a dollar a minute, give or take geography and seniority. That number gives shape to both the cost and the benefit side. Every minute spent checking AI output is a minute of that expertise being spent, and every minute the AI hands back to your team is worth that same rate.

For a planning estimate, start with annual cash compensation, employer-paid benefits and payroll costs, and recurring role-specific technology or overhead. Divide that total by productive minutes, not simply the 2,080 hours from the payroll calendar. State the utilization assumption explicitly, and use a range when geography, seniority, or role mix varies. This is a simple way to begin pricing work, not an actual person’s worth.

It’s pretty quick and uncomplicated to run through some of the math. Take an illustrative cash compensation for an HRBP of $94,300 plus the 43% that federal data puts on top for benefits and payroll costs and you get about $134,800 a year.B Divide by 2,080 paid hours and the minute costs about $1.08. Divide instead by the roughly 1,750 hoursC a person actually has left after leave, holidays, training, and the meetings that produce nothing, and it is closer to $1.28. Adding roughly $2,400 for tech we equip them with adds about two cents a minute.

We need this number so that the rest of the analysis stops being hand-waving. The AI bill will be an exact number, so we need something to go on that’s not a vaguely described “productivity gain,” which is how organizations end up with AI business cases that could easily be challenged.

What does a fully loaded minute cost in your organization? How would it make sense to measure that?