An AI budget request is often a proxy for a more fundamental question: "What value are we getting for this spend?" Framing the discussion around outcomes and ROI, rather than just the cost of tools, helps align finance and technology leaders on the strategic purpose of the investment.
Around 80% of AI pilots fail not because the technology is ineffective, but because there is no clear owner. Initiatives without a designated leader lack a defined ROI, upfront goals, and ongoing spend tracking, leading them to wither without demonstrating value.
FP&A teams must develop a new muscle for managing AI costs, which behave like consumption-based cloud spend, not predictable recurring software fees. Uncapped token usage can lead to massive budget blowouts, requiring more dynamic tracking and forecasting than traditional expenses.
For mid-market companies, the primary ROI for AI isn't cost reduction through layoffs. Instead, the key metric is increasing or flattening revenue per head. AI should empower smaller, leaner teams to become more productive and drive growth without proportional increases in headcount.
Encourage experimentation with AI, but mitigate risk by giving every pilot a clear budget limit ('a leash') and a deadline ('a clock'). This framework allows for innovation while ensuring underperforming projects are killed quickly, successful ones are funded, and great ones are scaled.
AI is only as good as the data it analyzes. Companies must treat shoring up their data infrastructure as a distinct, foundational investment. This includes cleaning not just quantitative data (financials) but also contextual data (policies, strategy documents) to avoid the "garbage in, garbage out" problem.
Even massive, engineering-forward organizations like Uber and Meta are struggling to accurately forecast and manage their AI spend. This indicates that the challenge is universal and not a sign of failure for mid-market companies, who are facing the same immature, evolving landscape as tech giants.
Beyond traditional ROI, a key metric for AI success is the 'acceptance rate'—how often users accept an AI-generated output without edits. This KPI measures the tool's practical utility and trustworthiness. Defining a 'good' acceptance rate helps quantify when an AI tool is genuinely driving value versus creating rework.
While most AI bets should be tied to ROI, CFOs should also establish an 'internal enablement fund' or 'slush fund.' This dedicated budget allows for low-ROI, high-learning experiments. It formally separates pure exploration from value-capture initiatives, fostering innovation without compromising financial discipline on major projects.
Annual budgets are insufficient for managing volatile AI costs. Because token spend can balloon unexpectedly and value capture can change quickly, finance teams must implement monthly or quarterly re-forecasts for AI-specific line items. This agile approach allows for faster decisions to kill, fund, or scale projects.