Rippling builds ROI tool for employees after investing millions in AI in a few months

HR software provider Rippling this week announced AI Spend Console, an anti-tokenmaxing product that helps companies track and contain their AI spending. One of the most interesting features is mapping how much individual employees, teams, and roles are spending and whether they are truly being productive or producing more AI output in general.

The tool will show “which engineers are spending the most on AI and whose colleagues are frequently asking for rework in code reviews,” the company said in a blog post.

The tool comes after Rippling, like many companies, went all-in on token maxing earlier this year, only to discover that its employees were hoarding cash. Chief Product Officer Matt McInnis still remembers when CFO Adam Swieczycki announced some shocking numbers at a March leadership meeting.

Rippling planned to spend 40% of its R&D headcount budget on AI tokens. This means that they were spending an amount equal to 40% of the compensation they paid to employees in that department in tokens. Millions of dollars. (The research and development organization is home to the engineering department of most technology companies.)

Spending is up 80% month-on-month, and if this trend continues, we will spend almost as much on AI tokens next year (90%) as we do on highly paid R&D employees.

“We couldn’t believe it,” McInnis told TechCrunch.

He said management immediately embarked on an “urgent” project to understand what it was spending and what it was getting for its money. In fact, the new product’s launch ad depicts Swieczycki sitting on a stool while an employee picks up wads of money and throws them into a shredder.

Rippling’s analysis revealed that “approximately 10-15% of our employees accounted for approximately 60% of our total AI spending, with one engineer spending $50,000 per month,” the company shared in a blog post.

Rippling didn’t want to stop the use of AI, they just wanted to suppress it. They started by negotiating spending limits for each of the tools the company uses (Cursor, OpenAI, Anthropic). An obvious problem was immediately discovered. By default, employees used the latest and most expensive Frontier models for all tasks.

“The truth is, inference providers like Anthropic and OpenAI have no incentive to help manage spending. They have every incentive to let spending run wild, and that’s exactly what they’re doing. They don’t provide great insight into usage, and they don’t collaborate with each other,” MacInnis said.

That was a common problem in early 2026. Eight months into this year, companies have figured out a few things. First, they know they need multiple models at different price points from multiple AI Labs, including perhaps a Chinese-made Frontier open weight option.

Rippling founder and CEO Parker Conrad said last month that when the company ran its own benchmarks for internal use, SpaceX’s Grok was found to be the overall leader, but that “the GLM 5.2 is 85% cheaper, but (had) almost the same performance as the Frontier model.” (SpaceX currently owns Cursor, which provides access to Grok and dozens of other models.) Z.ai’s GLM 5.2 has recently become a particularly popular Chinese model among tech companies for coding tasks. Databricks supports it too.

Second, businesses know they need an AI gateway that routes prompts to the most cost-effective model for the task. Rippling has also come to that conclusion. So we built our own AI gateway, which is also part of this product. MacInnis said companies that are already using another gateway can continue to use the AI ​​Spend Console product, but if they want the ability to manage their spending, they should use Rippling’s gateway.

The AI ​​Spend Console creates dashboards (known as leaderboards in the days of tokenmaxxing) that score attributes such as prompts per day, which combines work output (lines of code/pull requests) and spending.

With the tool in place, Rippling said it has reduced token spending from 40% of its staffing budget to about 15%. However, the use of AI was not curtailed. McInnis said the company spent a peak of 605 billion tokens in the month the CFO issued the warning. In July, internal usage once again reached 600 billion tokens, but “the cost of spending tokens in July was 37% of the cost of spending tokens in April,” he said.

“That’s because we’re now routing to a more effective model,” he says, joking, “We’re not going to force our sales team to use Fable to do grammar updates.”

But Rippling points out that technological solutions alone are not enough. The company has appointed an “AI captain” tasked with finding people who are using AI effectively and supporting the rest of the company.

Still, such efforts to leverage AI beyond engineering are underway, as software engineers have been the primary users so far, MacInnis said. But Ripling is working to help customer onboarding teams automate some email data and data reconciliation tasks, for example. The dashboard measures productivity in terms of onboarding more customers.

“We have to be able to link token consumption in general administration and customer-facing departments to productivity. If we can’t do that, all bets are off that any of these capabilities will be available to a broader employee base,” McInnis says.

So, using Rippling as an example, tokenmaxxing could have taken a big turn in a different direction where employee AI access could become less like Slack or email. If a company cannot measure productivity, it may not be accessible to all employees.

This product includes the AI ​​Spend Console for Rippling HR subscribers, but there are additional costs based on AI usage. It can also be purchased as a standalone product and integrated with another HR records system, MacInnis says.