Two “AI projects” can have the same headline and wildly different invoices.
The word AI tells you almost nothing about the amount of work underneath it.
What actually drives cost
- Number of systems involved
- Quality and structure of the data
- Permissions and security
- Number of real edge cases
- Whether a human approval queue is needed
- Reliability and volume
- Custom interface requirements
- How many people need to change how they work
A better way to scope it
OutcomeWhat should be different when this works?
Current workflowWhat happens today?
ExceptionsWhat breaks the normal path?
Minimum useful scopeWhat is the smallest version worth building?
AccessWhat systems and data can we actually use?
PriceOnce those are clear, quote the work.
Why some fixed prices are still useful
If the problem is repeatable enough, a starting price reduces uncertainty. That is what the FrictionList Fix Menu is for. Custom work stays custom.
What I would not accept from a vendor
- A price based only on “number of AI agents”
- A proposal that never maps the current workflow
- No explanation of ongoing software or model costs
- No plan for failures and exceptions
Talk to us about your workflow
You do not need to know the technology. Start with what is taking too long or falling through the cracks.
Start there