An agent built on a messy process just produces mess faster. Many small businesses pay for an agent when a simple automation was enough, or when they first needed to sort out the process.
Who it is for
Business owners who are being offered an "AI agent" or want to build one, and need to decide in 10 minutes whether it is the right time.
Answer the three questions with one concrete process in mind, not "customer service" in general. If even one sign shows up, the agent can wait. This is not being against agents: it is putting them where they actually help.
The protocol, step by step
-
Pick one process
Write in one line which process you want to automate, what goes in and what should come out. Example: "A quote request arrives by email and a sent quote comes out".
-
Sign 1: the process changes every week and nobody can explain it
Ask two people who do it to describe it separately. If the versions do not match, or if it changed in the last month, there is nothing stable for an agent to follow.
Copy and pasteDescribe step by step how you do [PROCESS] from the moment it arrives until it is done. Include what you check, what you decide and who you notify.
-
Sign 2: the task has fixed rules
If you can write the process as "if A happens, do B" without important exceptions, you do not need reasoning: an automation is enough and it is cheaper, faster and more predictable.
-
Sign 3: there is no metric to know whether it worked
Write which number will change if the agent works: response time, cases closed, errors, hours freed. If you cannot write it, you will not be able to tell whether it was worth it.
-
If any sign showed up: stabilize, document and measure
For two weeks do the process the same way every time, write it in a one-page document and log the metric by hand. You can use AI to organize what you already do.
Copy and pasteHere are the descriptions of how we do [PROCESS]: [PASTE]. Merge them into a procedure of at most 10 steps, mark where they disagree and tell me what simple metric could measure whether it works well.
-
Then add reasoning only where there is real ambiguity
With the process stable, mark the steps where a person has to read free text or decide with judgment (classify an email, understand a complaint). Only there does AI come in. The rest stays rule-based automation.
Common mistakes
- Buying the agent to discover the process. The agent needs a process; it does not invent one.
- Using an agent where a rule is enough. You pay more and get less predictable results.
- Judging by a demo instead of your real cases. Always test with examples from your own business.
Who this is NOT for
If your process is already stable, documented and has a metric, this filter does not stop you: go straight to the PILOT protocol to test the agent without risk.
Want it applied to your business?
If you would rather not build it alone, book 30 minutes with me and we will review it with your real case. If you still do not know where to start, the map tells you what to automate first in 5 questions.