Find the first process worth automating

A new tool is not the project. The project is a piece of work moving from a recognizable beginning to a useful end.

Owners often start with a product name: “We need AI,” “We need a CRM,” or “We should automate the office.” Those ideas may be reasonable, but they are too broad to build from. First describe what a person actually does today.

For example, “customer management” is not yet a process. “A new inquiry arrives, we collect the details, decide who should reply, and know the reply was sent” is.

The best clue is rarely one dramatic failure. It is the small interruption that keeps asking you to remember, search, retype, or explain.

Do not fall in love with the first idea. Three candidates are enough to create a real comparison without turning this into a consulting project.

Choose work you can observe now. Avoid a future process nobody has performed yet. Write each candidate in one sentence:

1. Candidate one: something that repeats and annoys you.
2. Candidate two: something that creates avoidable waiting or uncertainty.
3. Candidate three: something small that would make the day calmer even if it did not increase sales.

If you cannot describe where a process starts and ends, keep observing it. That does not mean it is unimportant. It means it is not ready to be the first build.

Score each candidate 0, 1, or 2. The number does not make the decision for you. It makes your reasons visible enough to discuss.

Some work matters too much, changes too often, or carries too much consequence to be your practice field.

Imagine a local service business where inquiries arrive by email, phone, and handwritten notes. The owner wants to respond reliably without replacing the systems that already work.

Three ideas surface. These example scores are illustrative, not universal. A real owner might score the same work differently because the risks, volume, and exceptions differ.

The winner is not the most ambitious idea. It is the one that creates a trustworthy record of incoming work and shows what needs attention. Later versions can help draft replies, prepare a quote, or connect to scheduling, once the owner trusts the basic flow.

Small does not mean a toy. It means the shortest complete path from a real input to a result someone can use.

Now name what is deliberately outside version one:

• no automatic email sending;
• no quote calculation;
• no calendar or accounting integration;
• no import of every old customer record;
• no customer login.

A good first app removes clerical friction while keeping judgment, responsibility, and unusual conversations with people.

If AI is later added, begin in “draft for review” mode. Keep sensitive or proprietary information out of tools that are not approved for it, and make the reviewer’s responsibility explicit.

You do not need a technical specification. You need a short, testable description of the work and its boundaries.

PROCESS
When:
We:
Done when:

PEOPLE
Who uses it:
Who approves unusual or sensitive actions:

INFORMATION
What comes in:
What must be remembered:
What must stay private:

SMALLEST USEFUL VERSION
The user can:
The app shows:
The app does not do yet:

HUMAN CHECKS
A person reviews:
A person must approve:
If something is wrong, we:

PILOT
We will test with:
We will know it is useful when:
We will review the decision after:

Fill this in with ordinary business language. If a developer or coding agent asks a technical question you do not understand, ask for a recommendation and the tradeoff in plain words. Your job is to protect the business meaning, not to recite code syntax.

Build the smallest path, try it with safe examples, and compare it with the work you actually do. Learning is the first deliverable.

1. Walk through the brief with the person who does the work. Correct missing steps and exceptions before building.
2. Use copied or test data first. Do not put the first experiment in charge of live money, commitments, or irreplaceable records.
3. Run it beside the current process. Let the app show what it would have done while a person keeps control.
4. Write down friction, not wishes. Fix the points that stop a real item from reaching “done.”
5. Review the boundary. Keep, expand, narrow, or stop based on what the pilot taught you.