How to Scope an AI Project So You Don't Get Ripped Off by Dev Shops

Protect Your Budget

Development agencies love vague requirements. It means more billable hours. "AI" is a license to overcharge. Here is how to protect yourself.

The Three Rules

1. Define Success Metrics

"The AI must answer accurately 90% of the time" is a metric. "Make it smart" is not. Before any engagement, agree on measurable outcomes:

• Accuracy rate on a test dataset
• Response time under load
• Cost per query
• User satisfaction score

1. Demand Fixed-Price POC

For AI projects, demand a Proof of Concept at a fixed price first. This should take 1-2 weeks and prove the core concept works. If an agency cannot deliver a POC in two weeks, they cannot deliver the full project.

Never sign a large contract without a working POC.

1. Own Your IP

Ensure you own the prompts, the fine-tuned weights (if any), the data pipelines, and the code. Some agencies retain ownership of "proprietary AI systems" they built with your data. This is unacceptable.

Red Flags in AI Proposals

• Vague timelines ("6-12 months")
• No mention of evaluation or testing
• Emphasis on "research" over "delivery"
• No fixed-price option
• Refusal to show a POC first
• Proprietary lock-in language in the contract

What Good Looks Like

A good AI partner will:

• Show you a working demo within two weeks
• Give you clear metrics for success
• Work in fixed-price sprints
• Hand over all code and IP
• Be honest about what AI cannot do for your use case

We work in fixed-price sprints. You know exactly what you get and when.