Rippling's AI token costs hit 40% of R&D budget, sparking new controls

Runaway AI spending forces enterprise reckoning

HR software provider Rippling faced a sobering reality check in March when CFO Adam Swiecicki presented figures showing the company was on track to spend 40% of its R&D headcount budget on AI tokens—millions of dollars burning at an 80% month-over-month growth rate. If the trend continued, AI token costs would reach 90% of what Rippling paid its entire R&D workforce within a year.

The discovery prompted what Chief Product Officer Matt MacInnis described as an "urgent" investigation into where the money was going and what value it delivered. This week, Rippling launched AI Spend Console, a product born directly from that internal crisis, according to TechCrunch.

Concentration and waste in employee AI usage

Rippling's analysis revealed stark patterns. Roughly 10-15% of employees drove 60% of total AI spending. One engineer alone consumed $50,000 monthly in tokens. The culprit wasn't malicious overuse but structural: employees defaulted to the newest, most expensive frontier models for every task, regardless of complexity.

The company immediately negotiated spending caps with its AI providers—Cursor, OpenAI, and Anthropic—and began building infrastructure to control costs without curtailing AI adoption. MacInnis noted that inference providers "have absolutely no incentive to help you control your spend," offering limited usage insights and no cross-provider collaboration.

Model routing and employee accountability

Rippling's solution centers on an AI gateway that routes prompts to appropriate models based on task requirements. The company's internal benchmarks found that while SpaceX's Grok performed best overall, Z.ai's GLM 5.2 delivered nearly identical performance at 85% lower cost than frontier models. The Chinese-origin GLM 5.2 has gained traction among tech companies for coding tasks.

The AI Spend Console produces dashboards tracking individual and team metrics: prompts per day, work output (lines of code, pull requests), and spending. The tool can identify engineers with high AI costs whose work frequently requires revision in code reviews—a direct productivity measure.

With these controls, Rippling reduced token spending from 40% to 15% of headcount budget while maintaining usage levels. July's 600 billion tokens cost 37% of what April's 605 billion tokens did, purely through better model routing.

Why it matters

Rippling's experience illustrates a broader enterprise challenge as AI tools proliferate without traditional IT governance. The company's trajectory—from unchecked adoption to granular employee-level tracking—suggests AI access may become more restricted than communication tools like Slack or email. MacInnis indicated that if productivity can't be measured, "all bets are off" on broad employee access. This shift could fundamentally change how enterprises approach AI deployment, moving from universal access to earned privileges based on demonstrated ROI.

Beyond engineering use cases

While software engineers remain primary AI users, Rippling is expanding measurement to other functions. Customer onboarding teams are testing automation for mailing data and reconciliation tasks, with dashboards tracking productivity through customer onboarding volume. The company appointed high-performing AI users as "AI captains" to assist colleagues—a recognition that technology alone can't drive adoption.

AI Spend Console is included for Rippling HR subscribers with additional usage-based costs, and can be purchased standalone for integration with other HR systems. Details were first reported by TechCrunch.