What AI consulting costs, and how the price should be built

Three inputs to the price

Start with revenue. What does the company make in a month, and how much of it passes through the workflow in question? An intake process that feeds every new client touches all of it. A reconciliation step at month end touches the part that gets refunded, disputed or written off.

Then wages. Who touches the workflow, how often, and what do those hours cost fully loaded? This is the number companies already know and rarely add up.

Then the change. If the workflow worked, what would be different next quarter: fewer people on it, faster cash, less spoilage, fewer findings? That is the value the system is priced against.

Why the client confirms every number

A consultant who arrives with a benchmark is guessing. A consultant who asks you for the numbers and reads them back is building a price you can defend to your own board. The discipline cuts both ways: if the confirmed value is small, the price should be small, or the work should not happen.

Fixed fee, retainer or outcome share

Once the value is on the table, there are three ways to pay for it. A fixed fee for a scoped build. A monthly retainer for ongoing operation. Or a share of the measured saving.

Most companies, offered a share of the saving or a fixed number, take the fixed number. It is easier to budget and easier to explain. The share is worth offering anyway, because it shows the firm is willing to be paid on the result.

Where outcome-only pricing goes wrong

Pricing purely on outcomes sounds fair and often is not. The builder carries all the risk of the client's adoption, data quality and timing. When the math favors the outcome and not the time, the builder cuts corners to reach the outcome, and the client inherits a fragile system.

The honest version is a fixed fee that reflects the confirmed value, with a smaller outcome component on top. The firm should know its own delivery well enough to price the risk of running long. If it does not, it should not be quoting.

Red flags

• A price before anyone has traced a workflow
• A proposal that treats AI as a one-time project with an end date and no operating plan
• A prompt library or a set of chatbot instructions sold as a transformation
• Pricing tied to hours with no statement of what changes for the business
• Benchmarks from other companies standing in for your numbers

Questions to ask before you sign

• Which of our numbers is this price built on, and can we see the math?
• What will be different in ninety days, and how will we measure it?
• Who on our side has to do what, and for how many hours?
• What happens after it goes live, and what does that cost?
• What do you refuse to build?