AI Operations
Rippling's AI Bill Exploded, So It Built a Console to Measure Token ROI
The enterprise learned that generous AI access without routing, attribution, and outcome measures can turn productivity enthusiasm into a runaway expense.
AI adoption without cost attribution is not empowerment; it is an unmetered utility. Rippling's experience suggests teams need task-level model routing, spend limits, outcome measures, and coaching so access grows with demonstrated value rather than enthusiasm alone.
Rippling says its AI token spending was on track to equal 40% of its R&D headcount budget and was growing 80% month over month. Roughly 10% to 15% of employees generated about 60% of total spend, while one engineer reportedly consumed $50,000 a month. The company had encouraged heavy usage, then discovered usage and value were not the same dashboard.
Employees often selected the newest, most expensive frontier model for routine tasks. That is understandable when the interface makes every model look one click away and somebody else receives the invoice. It is also the AI equivalent of taking a helicopter to buy tea from the corner shop: impressive machinery, questionable unit economics.
Rippling built AI Spend Console to connect spending with employees, teams, roles, and work output. It also created an AI gateway that can route requests toward a more cost-effective model. The company says token spend fell from the equivalent of 40% of headcount budget to about 15% without meaningfully reducing usage.
Internal usage reached roughly 600 billion tokens again in July, close to its earlier peak, yet Rippling says July cost only 37% as much as April. These are company-reported results, but the operating principle travels well: use expensive reasoning where the task earns it, cheaper models for routine work, and a fallback path when quality is insufficient.