The short answer
Do not start with the flashiest idea.
A good first AI workflow is understandable before AI enters it. Someone owns it. The trigger and desired result are clear. The work happens often enough to learn from. A person can review the output before a serious consequence. If the system fails, the team can fall back to a known process.
This guide is for leaders, workflow owners, and teams comparing possible starting points. It is a selection tool, not a security, legal, privacy, employment, or industry-compliance review.
In the Rock N’ Roll Method, Key Numbers clarify the value worth observing, Operations makes the current workflow and ownership visible, and Leverage tests whether AI is a responsible amplifier rather than a substitute for leadership.
Step one
List the work before naming the technology.
Write down five recurring workflows that create friction. Use ordinary verbs: receive a request, find the right information, prepare a document, route an item, follow up, reconcile a status, or produce a report. Avoid entries such as "use AI for sales" because they do not describe an operating path.
For each candidate, name the person responsible, the event that starts it, the information it needs, the output it should produce, and the person affected by the result. If those basics are disputed, clarify the process before considering automation.
Step two
Use the seven-part first-workflow filter.
- Repetition: Does the work happen often enough to justify learning a new way to do it?
- Boundaries: Can the team describe where the workflow begins, ends, and hands off?
- Information: Are the required inputs available, usable, and appropriate for the proposed tool?
- Consistency: Is there a stable path for normal cases, even if some exceptions need a person?
- Consequences: Can errors be caught and corrected before they cause significant harm?
- Ownership: Is one accountable person available to answer questions, approve changes, and stop the test?
- Observation: Can the team describe what better work would look like and review evidence after use?
A candidate does not need to be perfect on every dimension. It does need enough clarity to test responsibly. A weak answer on consequences, permission to use the data, or accountable ownership is a reason to pause.
Step three
Compare candidates in plain language.
Mark each of the seven dimensions as strong, uncertain, or weak. Add one sentence explaining the judgment. The explanation matters more than a numerical score because it exposes assumptions the team can check.
Prefer a candidate with strong boundaries, low or reversible consequences, available information, and an engaged owner. A smaller opportunity with these characteristics is usually a better first learning environment than a larger opportunity filled with hidden exceptions.
Step four
Write a one-paragraph pilot statement.
Use this structure: "When [trigger] happens, the workflow will use [approved information] to assist with [bounded task]. [Named role] will review [output or decision] before [external action]. We will observe [evidence] and return to [manual fallback] when [stop condition] occurs."
If the statement is full of vague nouns or nobody can fill in the reviewer and stop condition, the workflow is not ready. Return to the process map or choose another candidate.
Poor first candidates
Leave high-consequence ambiguity for later.
Avoid beginning with a workflow that makes employment, eligibility, safety, legal, financial, medical, or similarly consequential decisions; sends external communications without review; depends on data the business cannot confidently use; or spans several poorly understood systems and teams.
That does not mean AI can never assist near important work. It means the first project should create disciplined learning without quietly transferring authority or unacceptable risk to a tool.
Next step
Turn the strongest candidate into an opportunity map.
Bring the current steps, sample cases prepared to protect sensitive information, known exceptions, and the people who understand the work. An AI opportunity assessment can help compare the candidate's value, feasibility, risk, and readiness before implementation.