Rippling Almost Blew 90% of Its R&D Budget on AI. Most Companies Have No Idea What Theirs Is Doing.

In March 2026, Rippling's finance team walked into an executive meeting with a number that stopped the room. The company was on pace to burn 40 percent of its entire R&D headcount budget on AI tokens. Not 40 percent of a small experimental line item: 40 percent of what it pays its engineers, in tokens. And the curve was steep. Monthly AI spend was growing at 80 percent month over month. Left unchecked, the following year AI tokens alone would cost almost as much as the entire R&D payroll: 90 percent of it.

That is not a hypothetical. It is what happened inside a real company, reported by TechCrunch this week, and it is worth sitting with, because most SMEs and mid sized organisations are running the same experiment right now, just with fewer zeros and no finance team watching closely enough to notice.

What is actually happening

Rippling's Chief Product Officer Matt MacInnis told TechCrunch that when the company dug into the numbers, the pattern was stark: 10 to 15 percent of employees were driving about 60 percent of total AI spend, and one engineer alone was spending 50,000 dollars a month. Employees defaulted to the newest, most expensive frontier models for every task, from serious engineering work to what MacInnis called grammar updates.

Rippling did not respond by cutting access. It built AI Spend Console, a product that tracks spend by employee, team and role, and pairs it with output: pull requests, code review rework, actual delivered work. It also built an internal AI gateway that routes each task to the most cost effective model capable of doing it well, instead of defaulting to the most expensive one. The results, as MacInnis described them: token spend dropped from 40 percent of the R&D budget to about 15 percent, while actual usage held steady. In July, the company used almost as many tokens as it did at its April peak, but at 37 percent of the cost, because the tokens were being routed intelligently instead of spent by default.

The strategic reframe

The obvious conclusion, the one most leadership teams will reach first, is: we need to control AI costs. That is true, but it is the smaller insight. The real one is this: Rippling did not have an AI spending problem. It had a visibility problem. Nobody could see, at the point of use, what a token was buying: real output, or a habit. The fix was not austerity. It was measurement, paired with routing intelligence and a handful of internal AI captains tasked with helping colleagues use the tools well.

This is the same lesson MAKIA Labs keeps repeating in a different key: the technology was never the bottleneck. Adoption without visibility just becomes spend without accountability, at whatever scale the organisation operates.

What this means for an SME or leadership team

First, if nobody in your organisation can currently answer what did we spend on AI tools last month and what did it produce, you do not have an AI strategy. You have an AI habit. That gap closes with a simple monthly review, not a new platform.

Second, resist the instinct to standardise on a single frontier model for everything. Rippling's own benchmarking found that a much cheaper model performed nearly as well as the frontier option for most coding tasks. The expensive model is not automatically the right tool for a routine task, and most SMEs are paying frontier prices for tasks that do not need frontier capability.

Third, name your AI captains before you need them. Rippling formalised something most organisations already have informally: the two or three people who quietly figured out how to use AI well and get asked for help constantly. Giving them time and recognition to support others costs nothing and compounds fast.

Fourth, measure output, not usage. Prompts per day and tokens spent are activity metrics. They tell you nothing about whether the work got better. Tie AI use back to something the business already tracks: cycle time, rework rate, customer response time. Otherwise you are optimising for the wrong number.

Rippling is a well funded tech company, and building a custom spend console is not realistic for most SMEs. But the underlying discipline is: know what you are buying, route work to the right tool for the job, and measure the work, not the activity. That does not require a product launch. It requires someone in the room willing to ask an uncomfortable question before finance has to.

Try this week

Pull whatever AI usage or billing data you have access to, ChatGPT, Claude, Copilot, or whatever your team uses, and identify your one or two heaviest individual users. Ask them directly what they are using it for and whether it is saving real time or just producing more output to review. Fifteen minutes, one honest conversation, and you will already know more than most companies do about their own AI spend.

Sources

• After Rippling blew millions on AI in months, it built an employee ROI tool, TechCrunch, Julie Bort, August 7, 2026