How to Map One Business Workflow Before Adding AI

Updated September 6, 2026. This practical worksheet replaces the earlier version of “The AI-Native Map.”

Before adding AI to a task, write down how the work happens today. A useful workflow map names the trigger, information, decisions, people, and finished result. You can start with a sheet of paper.

This article focuses on mapping one process. For the broader business perspective, read The AI-Native Map: A Field Guide for Small Businesses.

Choose one task with a clear finish

“Improve customer service” is too broad for a first map. “Prepare a response to a new service inquiry” is easier to examine. Start where the request arrives and stop when a person approves the reply.

Choose a task that happens often enough to observe, has a recognizable result, and has someone who can explain the exceptions. Avoid starting with a decision that you cannot safely test or reverse.

Fill in these seven lines

1. Trigger: What starts the task? A submitted form, incoming email, or scheduled review?

2. Inputs: What information is needed, where is it kept, and is your business allowed to use it in the proposed tool?

3. Current steps: What does someone actually do, including copying, searching, checking, and waiting?

4. Decisions: Which steps follow a fixed rule, and which require interpretation or judgment?

5. Owner and reviewer: Who is responsible for the result, and who approves it?

6. Exceptions: What happens when information is missing, a system is unavailable, or a request falls outside the normal service?

7. Finished result: What must be true before you call the task complete?

An example: preparing a service inquiry reply

This is an illustrative example, not a customer result.

A customer submits a form asking about a service. The office coordinator reads it, checks the service area, looks up the approved service description, and prepares a reply. The owner checks any price or scheduling commitment before sending.

The map reveals several different jobs. Saving the form to the correct contact record may be ordinary automation. Summarizing a free-text request may be a useful AI task. Approving an unusual discount remains a person’s decision.

If the service location is missing, the workflow should flag that gap. It should not guess the address or promise availability.

Choose one change to test

Mark each step as keep manual, simplify, automate with a rule, or test with AI. You do not need AI at every step. Sometimes improving a form or keeping an approved service sheet up to date removes the problem.

For this example, the first pilot could prepare a summary and a draft reply for review. Keep sending under human control until the workflow has been tested and you have explicitly agreed to a different scope.

Record a baseline before the pilot

Time a representative set of tasks. Include searching, drafting, checking, and correcting. Record the number of requests with missing details and any replies that need to be redone.

After the pilot, compare the same measures. Count review and maintenance effort as well as drafting time. A faster first draft is useful only if the complete process improves.

Bring your map to a conversation

You do not need a polished diagram. Bring your seven-line map and one example of a good result. We can use them to discuss whether a small AI pilot makes sense.

Book a free 30-minute discovery call.

Reference

The NIST AI Risk Management Framework provides voluntary guidance for considering trustworthiness throughout an AI system’s lifecycle. This worksheet is TAG's practical planning tool, not a certification or compliance assessment.