What Business Process Should You Automate First?

Start with friction, not a model. Look for work that is slow, repetitive, inconsistent, or difficult to see. A model is one possible ingredient, not the project definition.

Choose a visible result. You should be able to tell whether the change saved a step, improved access to information, shortened a handoff, or created a useful new capability.

Use context you can trust. The project needs an approved source of information, a defined input, or a process people can explain. Ai cannot rescue missing ownership.

Limit the first authority. Drafting, classifying, and recommending are safer starting points than sending, deleting, purchasing, or changing customer records.

Make learning cheap. A narrow project should reveal whether the problem, data, workflow, and user behavior support a larger investment.

• Can one person clearly own the outcome?
• Does the problem happen often enough to matter?
• Can we describe the current process from beginning to end?
• Do we have approved information or data to work with?
• Can the first version stop safely when it is uncertain?
• Can we judge the result without inventing a vanity metric?
• Is there a smaller version that answers the biggest unknown first?

Suppose a team spends time answering the same internal process questions. The first project does not need to become a company-wide Ai platform. It could begin with one approved set of procedures, source-linked answers, and a clear path for unresolved questions. That would test source quality, usefulness, and ownership before expanding. This is a hypothetical example.

• Choosing a flashy demo that is disconnected from a real operating problem.
• Measuring activity instead of whether the work became more useful.
• Starting with sensitive data or irreversible actions before the system earns trust.
• Expanding scope before the first decision or workflow is understood.

Bring me the recurring problem, the people involved, and what happens today. I would choose the smallest project that can create a useful result and teach us whether the larger idea is worth pursuing.

For any business I work with, step one is its brain: it reads the tools you already use, read-only, so whatever gets built starts from your own records.

How the brain works See Custom Builds