Someone on your team is using AI and nobody knows if it paid off
Maybe your service manager runs quotes through it. Maybe your front desk writes recall emails with it. The subscription looked cheap, so nobody asked what came back out.
A well funded software company just found out how expensive that question gets when you skip it. And the lesson has almost nothing to do with AI.
What happened at Rippling
HR software provider Rippling unveiled a product called AI Spend Console this week, which TechCrunch describes as an anti-tokenmaxxing product that helps a company track and contain its AI spending. Tokens are the units of text an AI model reads and writes, and you pay per token, so heavy use means a heavy bill.
The tool exists because Rippling went all in on tokenmaxxing at the start of the year and then discovered employees were wildly burning cash. Chief Product Officer Matt MacInnis recalls an executive meeting in March where CFO Adam Swiecicki presented a number that shocked them. Rippling was on track to burn 40% of its R&D headcount budget on AI tokens, meaning it was spending as much on tokens as 40% of all compensation paid to that unit. Spending was growing 80% month over month.
When they dug in, the picture got sharper. Roughly 10 to 15% of employees were driving about 60% of total AI spend, and one engineer was spending $50,000 a month.
The fix was mostly bookkeeping
Rippling did not ban AI. It negotiated a max spending cap with each tool it used: Cursor, OpenAI, and Anthropic. It found that employees defaulted to the most recent and most expensive models for every task, including tasks that did not need them.
So it built an AI gateway, a piece of software that routes each request to the cheapest model that can do that particular job. Token spend fell from 40% of the headcount budget to about 15%, and usage stayed high. Rippling used a peak of 605 billion tokens the month the CFO raised the alarm, hit 600 billion again in July, and paid 37% of April's cost for it.
MacInnis had a blunt line about why nobody warned them: "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."
Sean's take: token maxing was never the goal
Token maxing. What a thing.
One of my philosophies for life is that things are often in and out of balance. Optimizing for maximum token usage makes no sense to me. It only drives up costs. Trying to get your people to actually use AI is reasonable, and I would encourage it. But it comes down to incentives, and incentives have to be thought through carefully because they are so important.
Rippling measured maximum token usage. It got maximum token usage. It also got a bill the size of nearly half its engineering payroll. The technology did not fail. The scoreboard did.
I have a background in accounting, and there is a saying that gets passed around managerial accounting circles: you get what you measure. It sounds obvious until you watch a company live it out with millions of dollars. Rippling's own dashboards were called leaderboards during the tokenmaxxing period. If you put a leaderboard in front of smart, competitive people, they will climb it. They will climb whichever one you hand them.
The same mistake is sitting in most ad accounts
Paid ads is our area of expertise, and the incentives here are clear once you say them out loud.
Great agencies optimize for cost per new customer. Good agencies optimize for cost per lead. Bad agencies optimize for cost per click and cost per impression. Every one of those is a real number you can put in a report. Only one of them pays your staff.
Watch what happens at each level. An account optimized for clicks gets you cheap clicks, and cheap clicks come from people who are curious rather than ready to buy. An account optimized for leads gets you form fills, and some of those form fills are tire kickers, wrong service areas, or people who will never answer the phone. An account optimized for new customers gets fewer, better conversations, and your booked revenue goes up while the click count sometimes goes down.
We measure the actual new customers or clients our clients acquire. That is what they get. Over 10 years, more than 3.2 million dollars in managed ad spend, and more than 38,000 leads generated, the pattern has not changed once.
Cheap numbers only matter when they are attached to booked work
One of our clients runs new patient leads at 1.23 dollars each on Meta. That number is only worth talking about because we know those leads turn into booked appointments. Detached from bookings, a 1.23 dollar lead is a rumor.
This is the balance point Sean keeps coming back to. A single number pushed to its ceiling almost always breaks something else. Maximum tokens broke a budget. Maximum clicks breaks a sales team's morale and your cost per job.
MacInnis said something about Rippling's next phase that applies directly to your business: "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." Same test for your marketing. If you cannot link spend to booked work, you are guessing.
What to do this week
Write down the one number your marketing is judged on. Cost per booked job. Cost per new patient. Cost per signed client. Pick the thing that has money attached and say it out loud so your team knows it too.
Then open your last ad report and see whether that number appears anywhere. If the top of the report is impressions, clicks and click through rate, the account is being managed to the wrong scoreboard, whoever is running it.
After that, get real outcomes flowing back into Google and Meta. When you send the platforms your booked jobs and closed sales instead of only form fills, they start finding more people like your actual customers. That is offline conversion tracking, and it is the single change that moves an account from lead volume to revenue.
Do the same audit on your AI subscriptions. List the tools, list the monthly cost, and ask each person what output changed because of them. Rippling found the useful users and made them AI captains to help everyone else. You can do a smaller version of that with three people and a Friday afternoon.
Last, check what you are quietly rewarding. If you praise your front desk for call volume, you will get call volume. If you praise them for booked appointments, you will get booked appointments.
Measure what you actually want
Pick one real number → Feed platforms real sales → Cut what never books → Reward booked jobs
Final Thoughts
Rippling did not solve this by turning AI off. It kept usage flat and paid a third of the price, because it changed what it counted and how the work got routed. Your ad account works the same way. The spend is not the problem, and the tools are not the problem. The number at the top of the report decides everything underneath it.
If you want a second set of eyes on which number your ads are currently being managed to, book a call with us. We will look at your account and tell you what we see, whether or not you end up working with us.
