Why this matters more than picking a tool
Most automation projects fail for the same reason: the tool gets chosen before the process is understood. Someone buys the software, the team tries to bend their work to fit it, and six weeks later everyone is back to doing it by hand — plus a subscription.
The order that works is the opposite. Understand the process, pick the one worth changing, and only then decide what runs it.
This guide is the «pick the one worth changing» part.
One rule before anything else
Never automate a process you could not explain to a new hire in five minutes. If you cannot explain it, it is not a process yet — it is a habit, and habits have exceptions nobody wrote down.
First, the difference nobody explains
A chatbot answers. An agent finishes the task.
| The customer asks | A chatbot | An agent |
|---|---|---|
| What are your hours? | Tells them the hours | Tells them, and offers the next open slot |
| Can I book Thursday? | Says someone will confirm | Books it, confirms it, sets the reminder |
| Where is my order? | Asks for the order number | Looks it up and answers |
If what you need is answers, a chatbot is enough and it is cheaper. If what you need is the work done, that is an agent — and it is worth reading on.
The difference is not intelligence. It is whether it can act.
The four questions
Take one process. Score each question from 1 to 5.
How often does it happen?
Daily is a 5, monthly a 1. Frequency beats complexity: a simple task forty times a week is worth more automated than a complex one twice a month.
How much does the answer vary?
Same shape of answer is a 5. «Varies» is not the same as «complicated» — quoting a job can have twenty inputs and still be a 5, as long as the rules are the same.
What does it cost when it is late?
A lost customer is a 5. This is the question that turns hours into money: a reply eleven hours late is usually a customer who already called someone else.
Could someone else do it?
Anyone could, given the information, is a 5. Only you could is a 1.
Add them up
| Score | What it means |
|---|---|
| 16 to 20 | Start here. This is the one. |
| 11 to 15 | Worth doing, but after the first one works |
| 6 to 10 | Leave it. Revisit in six months |
| 4 to 5 | This is a human process. Protect it |
The trap in question 4
Owners almost always want to automate the task only they can do — because it is the one that hurts. It is the wrong place to start.
A task that only you can do usually depends on knowledge that was never written down. Automating it means first getting it out of your head, which is a project in itself. Starting there means months before anything works.
Start with the 5: something anyone could do given the information. It moves in days, it proves the idea to your team, and it buys you the time to do the hard one properly.
What should stay human
- The conversation with somebody who is angry
- The decision that changes what you charge
- The exception you make because you know the customer
- Anything where being wrong is expensive and being slow is not
The point of the system is not to remove people. It is to give them back the hours so they are available for exactly these.
And the rule that makes it safe
Nothing goes out without a human seeing it first.
Every task the agent produces gets reviewed: approved, adjusted, or sent back. That is not a limitation on the technology — it is the design. It is also what makes the difference between a system your team trusts and one they quietly stop using.
Want us to score yours?
We do a free Process Diagnosis: we sit with how your team actually works, run these questions across your real processes, and tell you the three worth automating first — with your numbers, not generic ones. No cost, no commitment, nothing to install.