The price is not the cost

The number on the pricing page is the easiest cost to compare, which is why everyone compares it, and why it is the least useful basis for a decision.

None of what follows is an argument against paying for tools. It is an argument for knowing what you are agreeing to before the workflow depends on it.

The seat you did not want

Per-seat pricing is fine when everyone uses the thing daily. It stops being fine at the boundary, and AI tools sit on that boundary more than most software does, because usage is usually spiky and concentrated.

Two people use it constantly. Six use it occasionally. Four were added because the plan had a minimum. You are paying ten seats for what is genuinely two and a half, and the tool’s own dashboard will happily show you this if you look.

Check the minimum before you check the per-seat price. A cheaper tool with a five-seat floor costs more than an expensive one that bills for three.

Usage pricing is a forecast, not a price

Metered pricing looks honest, and it mostly is. What it is not is predictable.

The failure mode is not the big month you planned for. It is the automation somebody wires up in week six that quietly runs on every record instead of every new record, and the invoice that arrives four weeks later. Metered pricing turns a purchasing decision into an ongoing operational risk, and that risk needs an owner and an alarm.

Before you adopt anything metered, ask two questions. What is the worst month this could produce if something loops. And can I set a hard cap, not a warning email, at a number I choose. If the answer to the second is no, price that in.

The evaluation you have to run anyway

This is the cost nobody puts in the spreadsheet, and it is often the largest one.

You cannot know whether an AI tool is good enough for your work without trying it on your work. That means somebody assembling real examples, running them through, and judging the output against what a competent person would have produced. For anything touching customers or money, it means doing that carefully rather than casually.

That is days of somebody’s attention, and it recurs. Models change underneath you. A tool that was right in March can drift by September without anyone telling you, because the change happened at a provider two layers down.

The practical move is to keep the examples. Twenty or thirty real inputs with the outputs you consider correct, in a file, so that re-checking is an hour rather than another week. Almost nobody does this and almost everybody wishes they had.

The integration tax

A tool that does not connect to where your work already lives costs more than its price, because the difference gets paid in copy and paste, forever, by whoever is unlucky enough to own the process.

Look at what the tool exports, not what it imports. Import is easy and everyone supports it. Export is where you find out whether you can leave, and whether the output is usable by anything other than the tool that made it. A CSV of your data is a different proposition from a proprietary format behind an API you need a paid plan to call.

The exit

Ask early what happens when you stop paying. Not because you plan to, but because the answer tells you what you are actually buying.

Some tools hand back everything you put in. Some hand back everything you put in and nothing the tool generated, which for an AI product can be most of the value. Some keep working read-only. Some go dark on the day the card fails.

The most expensive tool is the one you cannot leave, and you find that out at the worst possible moment unless you check at the start.

What actually to do

Four things, in order:

Work out the real seat count, then check the plan minimum before the per-seat price.
Cap anything metered at a number you chose, and give the cap an owner.
Keep your evaluation examples in a file. Re-run them when something feels different. This is the cheapest insurance available.
Export once, early, while you still have the option, and look at what comes out.

None of this is exotic and none of it takes long. It is just the part that happens after the demo, which is the part that decides whether the tool was worth it.

Why the price is on our cards anyway

Given all of the above, printing a price on a card might look like the least useful thing a directory could do.

It is still worth doing, because it is the fastest filter available. Knowing whether something is free, has a free tier, or is paid answers “is this even in range” in about a second, and that is what a shelf is for. The card gets you to a shortlist. The questions above are what you do to the shortlist.

What the card should never do is make you click through to a pricing page to learn something that fits in four characters.