Today everything is sold as an "agent", but most small business processes are solved with something simpler. Choosing too much costs money, maintenance time and less predictable results.
Who it is for
Owners and operations leads who must decide how to solve a process with AI or compare vendor proposals.
There are four levels: rule-based automation, automation with one AI step, agent, and agent with mandatory human approval. The golden rule: start with the least complex option that closes the full result, and move up only when you have evidence the previous one is not enough.
The protocol, step by step
-
Write the result, not the tool
Describe the process by the result it must close. "Every lead gets a reply and lands in the CRM" is a result. "A sales agent" is not.
-
Question 1: are the rules stable and repeatable?
If every case is solved with "if this happens, do that", use rule-based automation (n8n, Activepieces, your CRM's automations). It is the cheapest and most predictable level. If you answer yes, stop here.
-
Question 2: do you need to read text or documents to decide?
If the flow is fixed but one step needs to understand an email, a PDF or a message (classify, extract data, summarize), add an AI step inside the automation. The flow is still yours; AI only solves that step.
-
Question 3: does the goal vary and require several tools?
Only if each case needs to choose which steps to take and in what order, using different tools (search, calculate, write, schedule), consider an agent. Before that, run the FILTER check in the Vault.
-
Question 4: does a mistake cost money, reputation or is it irreversible?
If the answer is yes, at any level add mandatory human approval before that action: payments, sends to important customers, price changes, deleting data.
-
Use the tree with AI for a concrete case
Paste your process with this prompt for a second opinion. Use it also to review a vendor's proposal.
Copy and pasteProcess: [DESCRIBE INPUT, STEPS AND RESULT]. Evaluate it with this tree: 1) stable, repeatable rules? automation. 2) need to read text or documents? automation + AI step. 3) variable goal with several tools? agent. 4) high risk? mandatory human approval. Tell me the minimum level that closes the result, why the previous level is not enough and which actions must have human approval.
-
Move up a level only with evidence
Start at the level you got and measure for two weeks. If many cases go unresolved and they all require judgment, then move up one level. Never skip two.
Common mistakes
- Starting with the agent because it sounds more advanced. The advanced move is closing the result with the minimum.
- Forgetting question 4. Risk is assessed at every level, including a simple automation.
- Comparing vendors by the tool instead of the result they commit to closing.
Who this is NOT for
It does not help you choose which AI model to use or compare platform prices: it is for deciding the complexity level of a solution.
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.