What is the line teams most often forget?

Review time. Somebody has to read the output before it is used, and that effort scales with volume rather than with the number of licences. It is invisible at pilot scale, where one person checks a handful of results, and becomes the largest single line once the tool is working across a department.

Over what period should we calculate it?

Long enough to include a renewal and at least one significant change of direction, which for most teams means two to three years. A one-year view flatters tools with heavy setup because the setup lands once, and it hides the cost of a migration that a shorter horizon simply never reaches.

How do we compare a per-seat tool with a usage-priced one?

Model both against the volume you actually expect rather than the volume you are testing at, and look at where each becomes uncomfortable. Per-seat is predictable and wasteful when only some people use it; usage-priced rewards light use and can move sharply in a busy month unless a cap is in place.

Does this apply to a free tool?

It applies more, because the licence line being zero removes the only number anybody was watching. Free tools still take integration effort, still need somebody to own them, still consume review time, and still cost something to leave. A zero on one line is not a zero on the total.

Glossary

Total cost of ownership (TCO)

Total cost of ownership is what a tool costs across its whole life rather than what it costs to buy, adding setup, wiring it into your systems, training and the staff time to keep it running.

In plain terms

The quoted price is the part everybody sees and rarely the largest part. Around it sit the days somebody spends connecting the tool to your systems, the time the team loses learning it, the person who ends up owning it, and the work of checking output that nobody costed. Add those and two products with similar quotes can differ by a wide margin in what they actually take out of the year.

Why it matters

Because AI purchases are unusually good at hiding their real cost behind a small headline number. A per-person subscription looks modest and sits alongside integration work, prompt and template maintenance, review time on every output, and usage charges that move with how much people actually use the thing. Comparing two tools on the quote alone tends to select for the one with the friendliest pricing page rather than the one that costs less to live with, which is a different and more expensive mistake than overpaying.

How it works

Start with the licence, then keep going. The subscription or usage charge is the anchor, and it is the only line most comparisons contain. Everything below it is real spend that lands on somebody's budget, usually not the budget that approved the purchase, which is part of why it goes uncounted for so long.

Count the setup properly, including the work you will do twice. Connecting a tool to your systems, importing content, setting permissions and configuring it for how your team actually works is a project, and the first version is rarely the one you keep. Teams routinely budget the first pass and are surprised by the second.

Count the human time on both sides of every output. Somebody writes the instructions, somebody keeps them current as the work changes, and somebody checks what comes back. Review time in particular scales with volume rather than with headcount, so it grows exactly when the tool is working well and nobody is watching for it.

Include the cost of the thing not working. Support, the person who becomes the internal expert, the fallback when the vendor has an outage, and the rework when output was wrong and shipped anyway. These are hard to estimate and consistently non-zero, and leaving them at zero is the same as asserting they are.

Finish with the exit. What it would take to move off the tool later belongs in the total, because a low running cost paired with an expensive exit is a different proposition from a low running cost you can walk away from. That figure is easiest to estimate before you sign and hardest to face afterwards.

What a year actually contains

The proportions here are illustrative rather than measured, and the point they carry is the shape rather than the sizes: the line everybody compares is one of five, and it is routinely not the biggest. Which of the others dominates varies by tool and by team. A product that connects to nothing has little integration cost and may still consume enormous review time. A product that slots in cleanly may cost most in the person who ends up owning it. What does not vary is that four of these five lines are absent from the quote, so a comparison built from quotes is a comparison of one bar. The useful exercise is not to get the numbers exactly right, which nobody can do in advance, but to force each line to be non-zero and argued, because the ones left at zero are the ones that surprise a budget later.

Seen in the wild

Price a per-person subscription for a whole department, then add the days spent connecting it to the systems the work already lives in.
Notion AI
Put a model step inside an automation that runs on every incoming record, where the charge follows volume rather than headcount.
n8n
Run something on your own hardware to remove the per-answer charge, and notice the operating time that replaces it.
Ollama

Common misconceptions

People assume

The subscription is the cost.

In fact

It is the visible cost and frequently the smaller half. Integration, training, maintenance of instructions and review time together often exceed it, and none of them appear on the quote. A comparison that stops at the licence line is really a comparison of pricing pages.

People assume

Self-hosting is cheaper because there is no licence.

In fact

It moves the cost rather than removing it. Hardware, operating time, upgrades and the person who keeps it running replace the subscription, and for small volumes that trade is usually worse. It becomes genuinely cheaper at scale or where the data cannot leave, which are good reasons on their own terms.

Questions

What is the line teams most often forget?: Review time. Somebody has to read the output before it is used, and that effort scales with volume rather than with the number of licences. It is invisible at pilot scale, where one person checks a handful of results, and becomes the largest single line once the tool is working across a department.
Over what period should we calculate it?: Long enough to include a renewal and at least one significant change of direction, which for most teams means two to three years. A one-year view flatters tools with heavy setup because the setup lands once, and it hides the cost of a migration that a shorter horizon simply never reaches.
How do we compare a per-seat tool with a usage-priced one?: Model both against the volume you actually expect rather than the volume you are testing at, and look at where each becomes uncomfortable. Per-seat is predictable and wasteful when only some people use it; usage-priced rewards light use and can move sharply in a busy month unless a cap is in place.
Does this apply to a free tool?: It applies more, because the licence line being zero removes the only number anybody was watching. Free tools still take integration effort, still need somebody to own them, still consume review time, and still cost something to leave. A zero on one line is not a zero on the total.

Key takeaways

The quoted price is usually the smaller half of what a tool costs to live with.
Setup is a project, and the first version is rarely the one you keep.
Review time scales with volume, so it grows just as the tool starts working.
Self-hosting moves the cost rather than removing it.
The cost of leaving belongs in the total, and is easiest to estimate before you sign.

Tools that use this

Notion AI
Per-person pricing across a department, plus the integration work around it.
n8n
Where charges follow volume rather than headcount, inside an automation.
Ollama
Running it yourself, where operating time replaces the per-answer charge.

Last checked July 2026

All glossary terms