A spreadsheet is the correct way to start tracking AI spend. It is free, everyone can use it, and the discipline of maintaining it matters more than any tooling. Most organizations should begin exactly there. The mistake is not starting with a spreadsheet. It is staying with one past the point where it quietly stops telling the truth.
Where the spreadsheet works
In the early stage, a spreadsheet does everything you need. It holds the vendor inventory, maps API keys to teams, and supports a monthly reconciliation. When spend is modest, comes from a few sources, and is reconciled by one person, a maintained spreadsheet is genuinely sufficient and far better than nothing.
The first crack: manual reconciliation
The first sign a spreadsheet is straining is the time the monthly reconciliation takes. When pulling every vendor export and mapping it by hand starts eating a full day, the process has outgrown the tool. Manual reconciliation also drifts: a key gets missed, a new tool goes unlogged, and the spreadsheet slowly diverges from reality without anyone noticing.
The second crack: no early warning
A spreadsheet is a record, not a monitor. It shows you what happened last month, not what is happening today. Once consumption spend is large enough that a spike matters, a monthly spreadsheet cannot warn you in time. The gap between when a problem starts and when the spreadsheet shows it becomes a real cost.
The third crack: task and agent granularity
A spreadsheet tracks totals and line items well. It cannot track cost per completed task, retry amplification, or agent workflows that make many calls per task, because that data does not arrive in a form a person can transcribe. Once agents and complex workflows enter the picture, the spreadsheet is tracking the wrong unit entirely.
The fourth crack: too many owners
A spreadsheet works when one person maintains it. When ten teams need to see and update their own spend, a single shared file becomes a source of conflict and stale data. The coordination cost rises until the spreadsheet is more argument than record.
Knowing when to graduate
The line is not a dollar figure. It is a set of signals: reconciliation takes too long, spikes arrive unannounced, agents make the unit wrong, and too many people need the same file. When two or more of those are true, the spreadsheet has done its job and the organization needs infrastructure built to track spend continuously, at the task level, with attribution that maintains itself. Starting with a spreadsheet is smart. Noticing when it stops working is what keeps tracking honest.