Treat AI Budgets Like Headcount, Not Blank Checks

The Uber Wake-Up Call

Uber executives disclosed earlier this year that the company exhausted its entire 2026 AI budget in just four months. The harder admission followed: no one could connect the billions spent to any measurable customer benefit. While Uber deserves credit for transparency, the problem is widespread—most companies are simply staying quiet about similar budget overruns.

The root cause is structural. Traditional software budgeting relied on predictable per-seat licensing. AI operates on consumption-based pricing, where a few autonomous agents running unchecked can incinerate a quarter's budget in days. Leaders face pressure from competitors, investors, and headline-chasing executives to keep spending, but open-ended funding isn't strategy.

Why it matters

Only 28% of global finance leaders report clear, measurable value from AI spending, according to Deloitte research cited in the source article. Outside companies selling AI infrastructure—chipmakers, cloud providers, and model developers—most application-layer technology firms haven't seen AI translate to revenue or earnings growth. This disconnect signals that companies are buying capability without converting it to business results. The winners will be organizations that impose the same spending discipline on AI that they apply to hiring.

Quarterly Reviews Over Annual Plans

Annual AI budgeting fails because model pricing, capabilities, and workflows shift too rapidly for long planning cycles. Companies should move to quarterly reviews that tie spending directly to current business outcomes rather than future projections.

The key is pairing two numbers at every level—company, department, and team: the business outcomes targeted and the AI budget allocated to achieve them. Every manager should carry both a token budget and an outcome goal, evaluated together.

Three implementation strategies stand out:

• Joint accountability: Pair finance and engineering leaders who share responsibility for both token spend and impact metrics.
• Monthly strategic reviews: Track token costs alongside the outcome metrics AI was meant to improve. Rising costs with flat results demand immediate investigation.
• Dynamic reallocation: Redirect AI budgets to high-performing teams quickly, just as companies reassign headcount to successful projects.

Flexible budgets matter because rigid caps punish winning teams at their peak performance while allowing underperforming teams to waste resources unnoticed.

Build Efficiency Into the System

Expecting employees to manually select the most cost-effective model for each task is unrealistic given the constant launch of new models at varying price points. Routing layers solve this by automatically dispatching requests to appropriate models based on task complexity.

Complex, multi-step planning deserves frontier-model intelligence. Meeting summaries don't. Most routine work runs effectively on cheaper or open-source models, and routing systems make these decisions automatically. Coding tools generate the majority of token spend at most companies, and modern solutions can deliver frontier-quality results at roughly 60% lower cost by intelligently selecting models based on task requirements.

Transparency Without Leaderboards

Internal AI usage scoreboards backfire predictably. One company shut down employee rankings after workers gamed the system, spinning up unnecessary AI tasks to climb the board. Another engineer admitted inflating usage after a performance review criticized insufficient AI adoption.

Public leaderboards reward token consumption rather than business outcomes. Employees should see their own usage and costs—essential for preventing budget surprises—but hitting limits should trigger manager conversations about work impact, not competitive pressure to outspend peers.

From Vision to Line Item

AI spending must transition from funding abstract visions to demonstrating clear outcomes. The companies that apply the same rigor to AI budgets that they apply to every other expense will spend less and achieve more.

These details were first reported by Vinay Kuruvila, CTO of Tinder, writing for Forbes.