Jordan Zamir is the CEO of Turnstile, a San Francisco–based software company focused on sales-led business-to-business startups. Views are the author’s own.
For decades, software budgeting followed a predictable formula: Finance estimated headcount, mapped employees to software licenses, and forecast spending accordingly.
More salespeople meant more customer relationship management software licenses. Larger engineering teams required more development tools. Software costs generally grew alongside the workforce.
However, AI breaks that paradigm.
Software spending no longer scales primarily with the number of employees. It scales with the amount of work software performs. That's why many finance teams struggle to forecast AI spending — not because variable pricing is difficult to understand, but because the assumptions that governed software budgeting for decades no longer explain how AI software is priced.
AI introduces a primary driver of software spending that grows much faster than staffing.
Customer interactions, automated workflows, AI-powered product features, and internal assistants all consume compute resources.
None of these activities appear in a staffing forecast, yet each one directly affects infrastructure costs. A finance department can create an accurate staffing forecast and still overshoot its AI budget because the organization has generated significantly more output than it has added staff.
This shift explains why AI spending often feels unpredictable. Traditional budgeting models measure people. AI invoices measure activity.
Most organizations initially approached AI in the same manner they handled other new technology investments. They assigned budgets to departments, established corporate innovation funds and approved pilot projects with the expectation that spending would eventually stabilize as experimentation declined.
However, this rarely occurs.
As employees discover new use cases, customers adopt AI-powered products, and product teams expand AI capabilities, compute demand continues to increase. The organization’s AI budget grows not because more people were hired, but because more work is being performed by software.
That requires a different budgeting approach.
Performance marketing provides a useful example. Proven campaigns receive predictable operating budgets as the outcome (pipeline created, bookings) is known with high certainty as a function of the input (campaign spend on known channels). Experimental campaigns receive limited funding until they demonstrate measurable returns. Organizations do not treat every campaign as a permanent expense from the day it launches.
AI deserves the same discipline.
Experimental workflows should receive separate budgets, defined evaluation periods and measurable financial objectives. Those objectives may include reducing operating costs, increasing revenue, improving customer retention, or accelerating internal processes. Success should be measured by the economics of the workflow.
Once an AI workflow consistently creates value, it should move into the operating budget. That transition should occur because finance understands the underlying economics, not because enthusiasm for the workflow has grown within the organization.
Finance should evaluate whether the workflow creates enough value to justify ongoing compute costs, whether the underlying model remains appropriate as utilization increases and whether safeguards exist to prevent unexpected increases in spending.
Separating experimental workflows from operational workflows gives finance a structured way to encourage innovation without treating every pilot as a permanent operating expense.
That doesn't mean finance should focus solely on reducing AI costs.
Most employees should not have to think about model pricing or token consumption to use AI effectively. Those decisions belong in the technology stack, where requests can be routed automatically to the most appropriate model based on cost, performance and complexity.
Finance gains far more by understanding the drivers of AI spending rather than by imposing arbitrary spending limits. Transparency allows organizations to connect infrastructure expenditures to products, workflows, and business functions. Once those relationships become clear, decision-makers can distinguish between investments that do and do not create economic value.
For years, finance budgeted for software used by employees. AI requires finance to budget for software that performs work on behalf of companies. Those are fundamentally different economic models, and they require different budgeting disciplines.
Organizations that build lasting competitive advantages with AI will not have the lowest infrastructure expenditures. They will understand which AI investments deserve additional capital, which ones require further optimization, and which ones should never move beyond experimentation.
