What a well-planned AI project answers
The strongest AI projects are operating changes supported by software. They begin with a workflow, a measurable outcome and a clear boundary for human responsibility. Technology choices follow those decisions.
- 01
Start with a workflow, not a model
Choose a recurring piece of work with a clear owner, visible friction and an outcome the business already cares about.
- 02
Define the decision or action
Describe what should become faster, more consistent or newly possible. Avoid vague goals such as “use AI” or “build a chatbot”.
- 03
Map the evidence
Identify the documents, records, system events and human judgment required to perform the workflow well today.
- 04
Set acceptance criteria
Agree on quality, speed, cost, safety and adoption thresholds before building. Include examples of unacceptable output.
- 05
Design human control
Decide where people review, approve, correct or override the system, including what happens when confidence is low.
- 06
Prototype the risky assumption
Build the smallest working path that tests whether the data and model can meet the acceptance criteria with real users.
- 07
Plan production ownership
Name who monitors quality, updates knowledge, handles incidents and decides when the system must stop or change.
Your one-page AI project brief
- The workflow and its owner
- The business outcome and baseline
- Users and affected teams
- Required data and permissions
- Quality and safety thresholds
- Human review and escalation
- Prototype scope and timeframe
- Production owner and operating cost