What an AI employee costs, compared with hiring one

Most people price AI against software. That is the wrong shelf. The honest comparison is the person you were about to put on payroll, and once you make that comparison the interesting number stops being the monthly fee.

When a small service business asks what an AI worker costs, they are usually holding a comparison in their head: this thing versus the CRM they already pay for. It is a natural comparison and it produces a useless answer, because a CRM does not do work. It stores what work happened.

The comparison that produces a decision is the hire. Not because AI replaces the person in some clean one-for-one way, it usually does not. Because the hire is the alternative you were actually going to choose.

What the hire actually costs

Salary is the number everyone quotes and the smallest part of the total. On top of it sits payroll tax, benefits, equipment, and a desk. Then the parts nobody budgets:

• The ramp. Two to three months before someone is genuinely productive in a small business, because nothing is documented and they are learning it from whoever is least busy that day.
• The management overhead. Every direct report costs the owner hours a week that were previously spent on the business.
• Turnover. Industry estimates put the cost of replacing someone at anywhere from half to twice their annual salary once you count recruiting, ramp and the gap in between.
• The knowledge that leaves with them. This is the one that hurts and the one nobody prices. In a business running on spreadsheets and memory, the process lives in the person. When they go, it goes.

What the AI side actually costs

Here the surprising part is how little of it is the model. On a system we run in production, the measured cost of an AI-drafted reply is $0.00061. At a thousand conversations a month that is well under a dollar in inference. It is a rounding error against any subscription, let alone a salary.

So if inference is nearly free, what are you paying for? Three things, and they are worth naming honestly:

1. Standing it up. Discovery, configuration, connecting it to the systems you already run, and getting it to speak in your business's actual vocabulary rather than a generic one.
2. Keeping it right. An AI worker is not a thing you install. Prices change, services change, the way your team answers a question changes. Something has to keep the system current or it slowly starts being wrong.
3. Support time. This is the real variable cost, and it is human. Not tokens.

Why we price in conversations, not seats or tokens

Seat pricing punishes you for adding staff, which is backwards for a tool meant to let you grow without hiring. Token pricing asks a business owner to care about a unit they have no way to estimate. Neither maps to anything you can plan around.

A conversation is a thing an owner can count. You know roughly how many enquiries you get in a month. That is the number the price should attach to.

The question underneath the question

When someone asks what it costs, they are usually asking something else: is this going to be another subscription that quietly stops being used?

That is a fair fear and it has nothing to do with price. It is answered by adoption, and adoption is answered by whether the tool is better than the workaround. One deployment we ran went from zero to 1,536 captured conversations in its first month with no mandate and no training programme, because the inbox became the easier place to work. A tool people choose is worth its cost. A tool people are told to use is expensive at any price.

Related: What happens when it gets something wrong · How you know it is actually working