This month, HR platform Rippling launched a system for tracking AI tool spending. It shows, in real time, which AI subscriptions each employee and team is paying for and how much. Rippling has been candid about why it built the system: after spending millions of dollars on AI tools over the course of several months, the company found itself struggling to figure out where the money had actually gone.
What's notable is that Rippling didn't keep this system as an internal tool. It launched it as a product for sale to other companies — presumably because it saw this not as a problem unique to Rippling, but as one facing the market at large. Rippling is hardly the only team that has lost track of where its AI spending is going.
Building a tracking tool is the right instinct. But this sequence — spend first, track later — has already left a gap that can't be filled in retroactively. If you don't record how many hours a week a task took before you adopted a tool, there's no way to know afterward whether that time went up or down. Rippling's new tracking tool can see spending from this point forward. What it cannot see is how actual working hours changed during the months when those millions were spent — because that data was never collected in the first place.
The Gap Between Login Counts and Productivity
The data AI tool vendors hand you is usage data: login counts, prompts sent, monthly active users. These numbers show up clearly on any subscription-management dashboard. But none of them directly answer the question that actually matters: has this tool cut the time a given task takes?
That's not an oversight. Usage is the metric that serves AI vendors' interests. The more people use a tool, the more likely they are to keep the subscription — and some will upgrade their plan. Whether a customer actually saved working hours has no direct bearing on the vendor's revenue, so there's no incentive to put that data on a dashboard. If you want it, you have to generate it yourself.
During the months Rippling was spending those millions, the data it had on hand was almost certainly dollar amounts and login logs. What it lacked was a baseline: a record of how long specific tasks took each week before the tools were adopted.
Without a baseline, there's no way to verify results. Without verification, there's no basis for either keeping a subscription or cutting it — so the default becomes simply continuing. The same phrases repeat: "the team seems to be using it," "it's expensive, but it's probably useful." That's how millions of dollars quietly pile up over a few months.
For a team of five or fewer, or a solo founder, the unit wouldn't be dollars but Korean won, and the totals wouldn't be in the millions but in the hundreds of thousands — yet the same mechanism applies. If a three-person team subscribes individually to ChatGPT Plus, Claude Pro, Notion AI, and Perplexity, that adds up to somewhere between 150,000 and 250,000 won (roughly $110–190) a month. Over a year, that's 1.8 million to 3 million won (about $1,300–2,300). Whether that amount is excessive varies by team. But if no one can answer the question "How many more hours a month would these tasks have taken without these tools?" there's no way to judge what that spending actually represents.
Ask a team in this position whether its AI subscriptions are worth it, and most will say, "they're useful enough." That's not a lie — it reflects a genuine sense of the tools helping. But that sense doesn't prove hours were actually saved. The gap between feeling and data is exactly what a baseline is supposed to fill.
Is "Allowing" a Subscription the Same as Designing One?
Many teams manage AI subscriptions by simply "allowing" them — a team member asks, and it gets approved. Sometimes there isn't even a request; people pay with their own cards and expense it later. This approach has real advantages: individual team members can adopt tools quickly, and experimentation is easy.
The problem is that no one defines, at the moment of subscribing, exactly which task the tool is meant to replace. When a tool gets used a little bit for many different tasks, there's no unit to measure it by — and it becomes hard to tell even when the tool isn't working.
"Designing" a subscription starts from a different question, asked before you sign up: Which task's time will this tool reduce? How many hours a week does that task currently take? And three months from now, what would that number need to be for us to call the subscription a success?
For a small team, this kind of design doesn't need to be elaborate. A single spreadsheet row is enough.
| Tool | Target Task | Current Time (weekly) | Target Time (weekly) | Review Date | |------|-----------|---------------------|---------------|------------| | Claude | Weekly report draft | 3 hours | 1 hour | Nov 2026 | | Perplexity | Competitor research | 4 hours | 2 hours | Nov 2026 |
With those two lines in place, you can actually check in November whether the time went down. If it didn't, you can ask why: was the tool being used wrong, were expectations too high, or was it simply the wrong fit for the task from the start? Without those two lines, every renewal cycle turns "should we keep this?" into a question answered by gut feeling.
If you're already running several AI subscriptions and never recorded a baseline, you can still start in reverse. Next to each tool, write down an estimate: "how many more hours a week would this task take without it?" If you can't answer — if all you get is "I don't know" — that subscription's value remains unverified to this day. In that case, the fastest way to measure it is to cancel it for a month at the next renewal. If the work genuinely gets harder without it, it was earning its keep. If nothing changes, it wasn't necessary.
Tracking Time Comes Before Tracking Spend
Rippling's new tool now lets a finance lead say, "here's what we spent on AI subscriptions this month." But it has no answer for the next question: what did that money actually change, and by how much? A tool that answers the first question is now on the market. The second question can still only be answered by teams generating their own data.
A small Korean team doesn't need a Rippling-grade spend-tracking system. What you can do right now is simpler: pull up the list of AI subscriptions you already have running, and next to each one, write down the task that would take longer per week without it. Any subscription left with a blank next to it is one that keeps renewing without ever being evaluated.
There's a question every subscription renewal date should force: "without this subscription, how many more hours a month would this task take?" If you can answer that with a number, you can make a real call — keep it, expand it, or cut it. If you can't, what you have isn't a designed subscription. It's just one that was allowed.
