Six months ago, the mandate was clear: get AI tools into employee hands as fast as possible. Now, the bills are arriving, and they’re triggering uncomfortable conversations in finance departments across India and globally.
The pattern is consistent. A company rolls out GitHub Copilot to its engineering team, enables ChatGPT Enterprise for customer support, or grants marketing access to image generation APIs. Usage starts modest, then explodes. By quarter-end, AI spending has blown past projections by 40 to 200 percent — often without a clear trail showing what that money bought.
The Thousand Small Cuts Problem
The issue isn’t one rogue project burning through budget. It’s death by a thousand small requests. An engineer hitting Copilot suggestions hundreds of times daily. A sales team running every email through an AI rewriter. Customer support agents pasting entire conversation logs into Claude for summarisation, over and over.
Each individual request costs fractions of a rupee. But API pricing — the per-use fees companies pay every time software calls an AI service — scales linearly with usage. When thousands of employees each make dozens of daily requests, costs compound fast. Microsoft, OpenAI, Anthropic, and Google all charge based on tokens processed, meaning verbose prompts and lengthy outputs directly inflate bills.
One Bangalore-based fintech, speaking on condition of anonymity, reported that a single team of 15 analysts generated over $8,000 in unexpected API charges in one month by feeding lengthy compliance documents into GPT-4 for analysis. The task could have been handled with cheaper models or smarter document chunking. Nobody had set guardrails.
Finance Teams Are Building New Control Systems
CFOs and CIOs are responding with frameworks borrowed from cloud cost management — a discipline often called FinOps. The core idea: make teams financially accountable for the AI resources they consume.
Internal chargeback models are emerging as the preferred approach. Instead of AI costs sitting in a central IT budget, departments receive allocated quotas and get billed internally for overages. This forces team leads to justify usage and prioritise high-value applications. Infosys and TCS have both indicated they’re implementing departmental tracking for generative AI tool consumption across client projects.
Quota systems are the blunter instrument. Some organisations now cap the number of API calls or Copilot suggestions each employee can trigger per day. It’s crude but effective at preventing runaway spending while teams develop more sophisticated governance.
Third-Party Tools Are Rushing to Fill the Gap
The market opportunity hasn’t gone unnoticed. A wave of startups and established vendors now offer AI cost observability platforms — dashboards that track which teams, projects, and even individual prompts are driving spending.
Portkey, a Mumbai-based startup, provides an AI gateway that monitors and optimises API usage across multiple providers. Their pitch to enterprises: route requests to cheaper models when premium capabilities aren’t needed, and flag wasteful patterns before they hit the invoice. Similar solutions from Helicone, LangSmith, and established players like Datadog are gaining traction with cost-conscious enterprises.
The more sophisticated tools go beyond tracking to active optimisation — caching frequent queries so identical requests don’t trigger repeat API calls, or automatically downgrading model selection for simple tasks.
The ROI Question Nobody Wants to Answer
Cost control is only half the problem. The harder question: are these AI tools actually delivering value proportional to their expense?
Most enterprises cannot answer this today. They know how much they’re spending on Copilot licenses or API fees. They don’t know if that spending translated to faster code shipping, better customer satisfaction, or fewer manual errors. The metrics simply aren’t being captured.
This gap is pushing companies toward pilot frameworks with defined success criteria. Rather than blanket rollouts, forward-thinking organisations are running 90-day experiments with specific teams, measuring productivity changes against control groups, and only expanding access when numbers justify the cost.
What This Means for You
If you’re a CIO or CTO who enabled broad AI access in the past year, audit your actual spending against initial projections now — not at year-end. The gap is likely larger than you expect.
Implement departmental chargebacks before your next budget cycle. Teams that feel the cost of their AI usage behave differently than teams treating it as a free resource.
Evaluate AI observability tools, particularly if you’re using multiple providers or building on APIs directly. Portkey, Helicone, and similar platforms pay for themselves quickly when usage is high.
Finally, don’t fund AI experiments without defined success metrics. The companies that will extract real value from this technology wave are those treating AI adoption as a business initiative with measurable outcomes — not a technology checkbox with an open budget line.