Your AI Budget Is Already A Headcount Decision

Prashant Jalan is the founder of Guickly and previously led applied AI teams in the mapping technology industry.

gettyWhen I was a tech lead manager at Google, adding even one person to my team took almost a quarter. I had to write the justification, defend the scope in a planning meeting and then wait. Sometimes the answer came back as “yes”; at times, it didn’t. Every person on that team was a number somebody had signed their name to.

The same money moves very differently now. AI tools can be purchased through a corporate card or software subscription, often without the review or accountability that comes with adding a person to the team.​

​I hear a version of this from almost every company I talk to. A role opens up and never gets backfilled, because the team found a way to cover it. A manager who cannot get a requirement approved grows the AI tooling budget instead. Nobody questions it, because tooling is not headcount.

Those are labor decisions. They are just never called that.

It is also not little money anymore. Bain & Co. has modeled a scenario in which AI tokens eventually replace roughly a quarter of headcount operating expenses. Nothing reaches a quarter of operating expenses without displacing some other line item. In most companies, that line item is payroll.​​

The uncomfortable part is that nobody can actually compare the two options, because the two options do not live in the same place.​

The cost of an employee is relatively straightforward to calculate. Finance tracks a fully loaded cost tied to a specific person, with a manager, a start date and a place in the budget.​​

AI spend is the opposite. It arrives through annual contracts, corporate cards, employee expense reports and features quietly switched on inside software the company already bought. It changes every month. Most of it never gets attributed to a team, let alone a person. Plenty of companies cannot produce a full list of the AI tools their own employees use, which makes it difficult to determine how much they are spending on AI, what work that spending supports or how it compares with the cost of hiring.​

​The standard advice is to measure cost per unit of work and compare. Everyone tries this now, but it has limitations.

In sales, support and marketing, it is traceable. A resolved ticket. A processed claim. Those success metrics already exist and somebody already tracks them, so it is easier to map the cost against it and assess the impact.

Engineering, on the other hand, is harder to measure. It is also where much of the AI spend tends to go. Unlike a resolved support ticket or a processed claim, engineering output has no single, universally accepted unit of measurement. Lines of code, story points, commits and tickets closed can all offer some insight, but none captures the full value of the work. When these metrics become targets, they can also encourage activity that looks productive without necessarily improving the software.​

AI complicates this further by making it easier to generate code and complete discrete tasks without necessarily improving reliability, maintainability or business outcomes, leaving companies with a difficult measurement problem: engineering is a major area of AI investment, yet its impact is hard to isolate and quantify. Anyone claiming there is a simple, settled way to measure AI’s return in software development is either overlooking how much depends on the work being done, or they’re trying to sell you something.

When a team needs more capacity, it has two paths. One needs a headcount request, a finance review, a planning cycle and often a quarter of waiting. The other needs a card and 10 minutes.

Capacity flows toward the path of least resistance. That is not an anecdote about what AI can do, but more about how procurement works. A good share of the work shifting from people to AI is shifting because AI is easier to buy, not because anyone proved it was better at the job.

Sometimes it really is cheaper and better. When that holds, the shift is the right call. But a company substituting on friction rather than evidence has no way of knowing which case it is in. It finds out later, when quality slips or a bill arrives that nobody was able to accurately forecast.

I see this tension at my own company. When a team is small, even one hire represents a meaningful investment, and the choice between adding a person and expanding AI tooling becomes tangible. The challenge is that these options are evaluated differently, and only one of them comes with a number I trust.​

None of this gets better by arguing about whether AI replaces people. It gets better when the legitimate comparison becomes possible. ​

A useful starting point is an inventory of the AI tools running across the company, including those purchased outside formal procurement channels. Where possible, that inventory should connect usage to teams, users and business purposes, giving leaders a clearer picture of where the money goes and what work it supports. From there, they can begin evaluating what those tools actually deliver.​

Many organizations still lack that baseline, making it difficult to assess the value of their AI investments, let alone compare them with the cost of additional staff.​

Until AI spend becomes part of the same planning conversations as headcount, companies risk making labor decisions through procurement processes that were never designed to evaluate them. These decisions are already shaping how enterprise teams work and grow. The question is whether leaders have the visibility they need to understand the trade-offs.​​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?