AI automation providers have very different levels of experience, delivery process, and domain knowledge. Getting proposals is easy. Evaluating whether one fits your actual workflow is harder.
What to ask before you sign
Work they've actually shipped. Ask for relevant production examples, not only mockups or demos. Where confidentiality allows, ask to speak with a client who uses something they built. Ask what failed, how they detected and fixed it, and how the system is monitored now.
How they scope projects. A fixed price can be credible for a narrow, standardized deliverable. Complex custom work usually needs discovery, documented assumptions, acceptance criteria, and a change process before the estimate means much.
What happens when something breaks. Because things break. What's the response time? Is there a maintenance arrangement? Is there documentation that would let someone else step in if needed?
Who actually does the work. Ask who will design, implement, test, and support the system, including whether subcontractors are involved and who remains accountable for delivery.
Red flags
Proposals heavy on AI terminology and light on specifics about how your actual business process will work.
Specific outcome promises, "we'll cut your processing time by 60%", made before any discovery about your current process.
No mention of failure modes, edge cases, or what happens when the automation encounters input it wasn't designed for.
A timeline that is not tied to concrete deliverables, testing, dependencies, and client review.
The right framing
An AI automation project is a collaboration. You know your business. The builder knows the implementation options. Discovery should turn those two perspectives into an agreed scope, measurable acceptance criteria, and a support plan.
A builder who asks good questions, acknowledges what they don't know, and is honest about tradeoffs is more likely to deliver something that works than one who has all the answers before they've heard the questions.
