Rippling's AI Overspend Sparks Employee ROI Tool

HR software provider Rippling recently introduced its AI Spend Console, a new solution designed to help companies meticulously track and manage their AI expenditures. A notable feature of this product is its ability to map AI spending across individual employees, teams, and specific roles, allowing organizations to discern whether usage genuinely enhances productivity or merely contributes to excessive, unproductive AI output.

In its official blog post, Rippling highlights that the tool can pinpoint "which engineers have high AI spend whose peers frequently ask them to redo work in code reviews," thereby demonstrating the direct impact of AI usage on development efficiency.

The genesis of this tool stems from Rippling's own experience, having enthusiastically adopted extensive AI usage early in the year, only to discover an alarming rate of expenditure by employees. Chief Product Officer Matt MacInnis vividly recalls an executive team meeting in March where CFO Adam Swiecicki presented a staggering figure that left the leadership team in disbelief.

At that point, Rippling was projected to allocate 40% of its Research & Development headcount budget to AI tokens. This meant the company was spending an amount on AI tokens equivalent to 40% of the total compensation paid to employees within that unit, amounting to millions of dollars. (It's important to note that the R&D organization typically houses engineering functions in most technology firms.)

Furthermore, spending was escalating by 80% month-over-month. If this trend persisted, projections indicated that within the following year, AI token expenditure would nearly match 90% of the cost of its highly compensated R&D unit employees, an unsustainable trajectory.

"We were incredulous," MacInnis shared with TechCrunch, reflecting the team's astonishment.

Management promptly initiated an "urgent" project to comprehensively understand these outlays and evaluate the return on investment, MacInnis explained. The gravity of the situation is further underscored by the product's launch advertisement, which visually illustrates CFO Swiecicki observing employees as they discard large sums of cash into a paper shredder.

Rippling's subsequent analysis revealed critical insights, such as "roughly 10–15% of our employees were driving about 60% of total AI spend." The company's blog post further disclosed a striking example: one individual engineer was responsible for spending $50,000 a month.

Rather than ceasing AI usage altogether, Rippling aimed to significantly mitigate its uncontrolled growth. The initial step involved negotiating maximum spending caps with its primary AI tool providers: Cursor, OpenAI, and Anthropic. This process quickly unveiled a clear problem: employees were predominantly utilizing the newest and most expensive frontier models for all tasks, regardless of complexity or necessity.

MacInnis commented on the dynamics with AI inference providers, stating, "The truth is that 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, and that’s exactly what they do. They don’t provide you with great usage insight, and they don’t collaborate with one another."

This challenge was prevalent in early 2026. However, eight months into the year, enterprises have developed a more sophisticated understanding of AI deployment. Firstly, they recognize the necessity of leveraging a diverse portfolio of models from multiple AI labs, offering various price points. This includes considering frontier open-weight options, potentially from Chinese developers, for enhanced cost-efficiency.

Rippling's founder and CEO, Parker Conrad, observed last month that the company's internal benchmarking for its own applications revealed SpaceX's Grok as an overall leader. However, their findings also indicated that "GLM 5.2 is 85% cheaper but [had] nearly identical performance" to the more expensive frontier models. (SpaceX currently owns Cursor, which provides access to Grok and numerous other models.) Z.ai’s GLM 5.2 has notably gained traction among tech companies, including Databricks, as a preferred Chinese model for coding-related tasks.

Secondly, enterprises now possess the crucial understanding that an AI gateway is essential for intelligently routing prompts to the most suitable and cost-effective model for each specific task. Rippling arrived at this same conclusion and, consequently, developed its own AI gateway, which is an integrated component of the new AI Spend Console. MacInnis clarifies that while companies using other gateways can still leverage the AI Spend Console, full access to its spending governance features necessitates the use of Rippling’s proprietary gateway.

The AI Spend Console generates comprehensive dashboards (which were previously referred to as leaderboards during the peak of "tokenmaxxing") that provide scores for various attributes. These include prompts per day, combined with quantifiable work output such as lines of code or pull requests, and the corresponding expenditure.

With the implementation of this new tool, Rippling reported a significant reduction in its AI token spend, decreasing from 40% to approximately 15% of its headcount budget, critically without curtailing AI usage. MacInnis revealed that the company reached a peak of 605 billion tokens consumed in the month the CFO issued his warning. By July, internal usage again approached 600 billion tokens; however, "the cost of July’s token spend was 37% of the cost of April’s token spend," marking a remarkable efficiency gain.

"That’s just because now we’re routing to the more effective models," he explained, humorously adding that they are "not letting the sales team do grammar updates using Fable," emphasizing the strategic allocation of resources to appropriate models.

Rippling acknowledges that technological solutions alone are insufficient. The company identified employees who were effectively leveraging AI and designated them as "AI captains," tasking them with assisting and guiding the rest of the organization in optimal AI adoption.

However, MacInnis notes that efforts to expand AI usage beyond the engineering department remain an ongoing initiative, as software engineers have been the predominant adopters to date. Rippling is actively working to extend these capabilities to areas such as customer onboarding teams, aiming to automate tasks like mailing data management and data reconciliation. In such scenarios, the dashboard will quantify productivity by measuring the number of customers successfully onboarded.

"We have to be able to link token consumption in G&A functions and in customer-facing functions back to productivity," MacInnis emphasized. He added, "If we can’t do that, all bets are off on any of this stuff being available to the broader employee base," highlighting the critical need for measurable ROI to justify widespread AI access.

Rippling's experience illustrates a significant shift: the initial enthusiasm for "tokenmaxxing" has now given way to a more measured approach, suggesting that universal AI access for employees may not mirror the ubiquitous availability of tools like Slack or email. If an organization cannot effectively measure the productivity gains derived from AI, then access might not be extended to all employees.

Regarding the product's availability, the AI Spend Console is complimentary for Rippling's HR subscribers, though supplementary AI usage-based costs apply. MacInnis also noted that it can be acquired as a standalone offering and seamlessly integrated with an existing HR system of record.