Start here
How to buy AI
without being
sold to.
Six steps, no jargon, and nothing you need a technical background to follow. If you read one thing on this site before spending money, read this.
Work out what you're buying
Almost every AI product is one of three things: a wrapper around someone else's model, a workflow built on top of one, or the model itself. The wrapper is not worthless. The workflow is often the whole value, but knowing which you're looking at tells you what you're really paying for and how easily it can be replaced.
Start with the job, not the tool
"We should use AI" is not a requirement. "Our support team spends nine hours a week writing the same four replies" is. Every good outcome we have seen started from a specific, boring, measurable task that somebody hated doing.
Check what you already pay for
This is the single highest-return step and almost nobody does it. Your existing tools have shipped AI features in the last year. Audit those before adding a line to the budget. A large share of AI spend is duplication.
Understand the two numbers
For anything using a language model, context window and price per million tokens decide most of the outcome. Context is how much the model can consider at once; price is what it costs per unit of text in and out. Everything else is detail.
Run it against real work
Not a demo, not the vendor's sample data. Take last week's actual work and run it through. The gap between a demo and your data is where most disappointment lives.
Ask where your data goes
Does it train the model? Where is it stored? What happens when you cancel? If the vendor can't answer plainly, that's the answer. This matters most in healthcare, legal and finance, and it matters everywhere.