Rippling's AI Token Bill Was Out of Control

Rippling, the HR software company, built a new product called AI Spend Console after discovering it was burning millions on AI tokens. In March, CFO Adam Swiecicki revealed the company was on track to spend 40% of its R&D headcount budget on AI tokens. That's the equivalent of paying 40% of its engineering salaries to OpenAI and Anthropic. Spending was growing 80% month-over-month. If unchecked, it would hit 90% of R&D salaries within a year.

Chief Product Officer Matt MacInnis described the executive team's reaction: "We were incredulous."

The root cause? Employees defaulted to the most expensive frontier models for everything. One engineer alone spent $50,000 per month. About 10–15% of employees drove 60% of total AI spend.

MacInnis didn't mince words about the AI labs: "The inference providers, like Anthropic and OpenAI, have absolutely no incentives to help you control your spend. They have every incentive for it to be a runaway expense."

The Solution: AI Gateway and Spend Console

Rippling built two things: an AI gateway that routes prompts to cost-effective models, and AI Spend Console, a dashboard that tracks spending and productivity by employee, team, and role.

The gateway routes each prompt to the best model for the task. Rippling benchmarked models for internal use and found that SpaceX's Grok was the overall leader, but Z.ai's GLM 5.2 was 85% cheaper with nearly identical performance. The gateway likely uses a rules engine or LLM to pick models based on task complexity and cost.

For example, a simple grammar fix might route to GLM 5.2, while complex code generation might go to Grok. MacInnis joked, "We're not letting the sales team do grammar updates using Fable" (a high-end model).

AI Spend Console provides dashboards (formerly called "leaderboards" in tokenmaxxing days) that score prompts per day, work output (like lines of code or pull requests), and spend. It can show "which engineers have high AI spend whose peers frequently ask them to redo work in code reviews," according to Rippling's blog post.

Results: 40% to 15% of Headcount Budget

With the gateway and console in place, Rippling dropped token spend from 40% to about 15% of its headcount budget. Crucially, AI usage didn't drop. In April, the company hit a peak of 605 billion tokens. In July, usage hit 600 billion tokens again, but the cost was 37% of April's cost. That's a 63% cost reduction per token, achieved by routing to cheaper models.

Product Details and Availability

AI Spend Console is included for Rippling's HR subscribers, with additional AI usage-based costs. It can also be purchased standalone and integrated with another HR system. Companies already using another AI gateway can still use the console, but to enforce spending controls they'd need Rippling's gateway.

Beyond Engineering: Measuring Productivity

Rippling is extending AI to customer onboarding teams, automating mailing data and data-reconciliation tasks. The dashboard will measure productivity in terms of onboarding more customers. MacInnis stressed the importance: "We have to be able to link token consumption in G&A functions and in customer-facing functions back to productivity. If we can't do that, all bets are off on any of this stuff being available to the broader employee base."

The company also designated "AI captains" — employees who use AI effectively — to assist others.

What This Means for Developers

If you're building AI-powered applications, this is a cautionary tale. Token costs can spiral. You need observability into spend per user and per feature. Consider building a gateway that routes to cheaper models when possible. Benchmark models for your specific use cases. Don't blindly use the latest frontier model for everything.

Rippling's experience shows that cost control doesn't have to mean reducing AI usage. With smart routing and monitoring, you can cut costs dramatically while maintaining performance.

Practical Steps You Can Take

  1. Benchmark models for your workload. Test GLM 5.2 or other open-weight models against frontier models. You might find a 85% cost saving with negligible quality loss.
  2. Implement an AI gateway. Use tools like LiteLLM, Portkey, or build your own to route prompts based on cost and complexity.
  3. Track spend per user. Add logging to your AI calls to see which users or features are driving costs. Set alerts for anomalies.
  4. Negotiate caps with providers. Rippling negotiated max spending caps with Cursor, OpenAI, and Anthropic. You can do the same.

Conclusion

Rippling's AI Spend Console is a direct response to a real problem: runaway AI token costs. It's a product born from necessity, and it offers a blueprint for enterprises to control AI spending without stifling innovation. If you're managing AI costs, study Rippling's approach. The tools and strategies are replicable, and the stakes are high.