Software bill prices work

I have started looking at AI pricing less like a software bill and more like an operations report.

More AI plans are not only charging for access. They are charging for pieces of work: a resolved support conversation, an agent credit, a workflow task, an analysis response, or a scheduled run.

That can feel annoying. But it also turns a vague question into a better one: what work are we actually paying this system to do?

The bill is naming the workflow

Intercom prices Fin around outcomes. Notion agents use credits. Airtable AI analysis uses credits. Zapier prices automation through tasks, including AI steps and code.

The details vary by product, but the shift is clear. Software used to mostly ask how many people needed access, which plan tier you wanted, and how much volume you used. Now it is also asking:

• How many issues should this resolve?
• How many actions should this run?
• How many credits will this workflow consume?

That changes the buying conversation. A support resolution is not valuable just because the bot closed the conversation. It is valuable if the customer got the right answer, did not come back frustrated, and did not create cleanup work for the team.

The unit of work is not the same thing as the value. It is just the place to start looking.

AI ROI is not really an AI question. It is a workflow question with a new cost line.

The hidden cost is around the output

This is where I think small teams get tripped up. The visible price is rarely the whole cost.

The real cost includes the stuff around the output: setup, messy source data, unclear rules, review, cleanup, and the time it takes to manage another system.

AI can make an output appear so quickly that it feels like the work happened. The draft appears, the summary appears, the resolution appears.

Sometimes the actual work is still waiting for a person to check whether the output is useful.

A support agent may be cheap per answer, but expensive if the help docs are stale and half the conversations get reopened. A weekly reporting agent only helps if it pulls from the right numbers and answers a decision someone actually needs to make.

That does not mean the AI feature is bad. It means the workflow was not ready to be priced yet.

A better buying question

If I were deciding whether an AI feature was worth paying for, I would start smaller than the product page. I would pick one repeated workflow and define the work unit first.

For example:

• "One resolved support issue" means the customer got a correct answer and did not need human escalation.
• "One qualified lead" means the company, need, fit, and next action are clear enough for sales to use.
• "One useful weekly report" means the team can make a decision from it in under five minutes.

That definition is not bureaucracy. It is how you keep the bill connected to reality.

Without it, the team ends up measuring whatever is easiest to see: seats, prompts, credits burned, generated drafts, closed tickets.

Those are activity metrics. They are not proof that the work got better.

The more useful test is boring:

1. Name the repeated workflow.
2. Define what a good work unit looks like.
3. Decide who reviews the output.
4. Compare the cost, review time, and result before and after.

Some benefits will still be hard to measure. Better customer experience and less mental load do not fit neatly in a spreadsheet.

But small teams do not have unlimited attention. They cannot afford AI projects that add another system to manage without making the work better.

So before upgrading the next plan or turning on the next agent, start with the work.

Name it. Price it. Review it.

Then decide whether it is worth paying for.