The plain question
If a company has ten possible AI ideas, which one should go first?
The best first workflow is usually not the most exciting idea. It is the workflow where the work is repeated, painful, understandable, reviewable, and owned by someone.
That sounds less glamorous than “AI transformation.” It is also much more likely to work.
Start with work, not tools
A tool-first question sounds like this:
“Should we buy an AI assistant?”
A workflow-first question sounds like this:
“Which repeated piece of work is costing the team time, slowing customers down, or creating avoidable errors?”
The second question is better because it gives you something concrete to evaluate.
AI is useful only when it changes a real activity. A real activity has a trigger, input, decision, output, owner, and place where the result is used.
Six signs of a good first workflow
1. It happens often
If the work happens many times a week or many times a day, improvement can compound.
Example: repeated support questions, recurring invoice checks, lead intake, document summaries, internal policy lookup.
2. It creates visible pain
The workflow should cost time, slow response, create quality issues, or block higher-value work.
If nobody feels pain, the pilot will be hard to justify.
3. It has patterns
AI works best when cases vary but still follow recognizable patterns.
Example: customer emails may be written differently, but many of them ask about the same five policies.
4. It has source material
AI needs something trustworthy to work from: policies, SOPs, examples, prior decisions, templates, product information, or structured records.
If the source material is weak, the first project may need to be source cleanup.
5. It can be reviewed
The first AI workflow should have a clear human review point.
That does not mean every output must be checked forever. It means the pilot has a safe way to learn before expanding.
6. It has an owner
Someone must care about the workflow after the pilot launches.
This person decides what is acceptable, which exceptions matter, and whether the workflow should expand.
A simple scoring exercise
Before choosing the first pilot, score each candidate workflow from 1 to 5:
- repetition
- pain
- pattern clarity
- source quality
- reviewability
- ownership
- measurability
The best first workflow is not always the highest total score. Sometimes a slightly smaller workflow is better because the team can test it quickly and safely.
Example: three candidate workflows
Imagine a company considering three AI ideas.
Idea A: company-wide knowledge assistant
This sounds useful, but the sources are scattered and ownership is unclear.
Good idea, but probably too broad as the first pilot.
Idea B: support replies for delivery policy questions
The work is repeated, source material exists, a human can review replies, and first-response time can be measured.
Strong first candidate.
Idea C: executive strategy analysis
High importance, but low repetition, unclear source boundaries, and difficult measurement.
Not a good first pilot.
What leaders should avoid
Avoid starting with the idea that has the strongest internal sponsor if the workflow itself is weak.
Avoid starting with the most technically impressive demo if the team will not use it daily.
Avoid starting with the riskiest workflow just because the business impact sounds large.
The first pilot should teach the company how to use AI responsibly. It should not create a political, technical, or compliance crisis.
The Glasrocks view
The first AI workflow should be small enough to test, important enough to matter, and structured enough to manage.
That is why Glasrocks starts with workflow fit before implementation.
If you are choosing between several AI ideas, use the AI Workflow Fit Assessment first, then read the Glasrocks Method to understand how the pilot should be designed.