An AI agent is useful when the task is repeatable, the volume is real, and a person can tell quickly whether the result is right. It is a poor fit when the task is rare, vague, or expensive to get wrong.
Pick one specific task, such as answering booking calls or sorting incoming support email, and answer the questions below about that task only. Do not score “AI in general.”
#The ten questions
- Does the task repeat with clear steps: the same kind of input, the same kind of output, in roughly the same order?
- Does it happen many times a week: enough that saving a few minutes each time adds up?
- Is the input text or speech: an email, a chat message, a form, a phone call?
- Can someone describe what a correct result looks like: so well that two people would usually agree?
- Can a person check results quickly: in seconds, not by redoing the whole task?
- Is a wrong result cheap to fix: a rework or a short apology, not a lost client or a legal problem?
- Is the data it needs in a system you can give it limited access to: read-only, or only the records it needs?
- Can the task be done without judgment across many systems at once: or does it stay inside one or two?
- Will someone on the team own it after launch: reviewing results, updating instructions, and handling exceptions?
- Can you run it alongside the current process for a few weeks: so you can compare results before you rely on it?
#How to score it
Count your yeses. This is a rule of thumb, not a statistically validated test, so treat the bands as prompts for discussion.
- 0 to 4: not yet. Look at which questions got a no. Missing volume or a vague definition of “correct” are the most common. Write down the steps, define a good result, or pick a different task. Often a simple automation or a cleaner process is the better first move.
- 5 to 7: promising, with preconditions. The task could work, but each no is a risk to settle first. If the problem is data access, decide what the agent may see. If it is checking, decide who reviews output and how. If no one owns it, assign someone before you start.
- 8 to 10: a good candidate. Start small, run it alongside the current process, and keep a person reviewing results until you trust the pattern.
#What to do with your score
If you scored 5 or higher, write a short description of the task: the input, the steps, what a correct result looks like, and who reviews it. That page is also the start of a good project brief.
Then check where your task sits against realistic wins. Our post on where AI actually pays off in a small business covers the kinds of tasks that tend to work and the ones that do not.
If you want to talk it through, you can book a 30-minute call with an engineer. Bring your score and the task you picked, and we will tell you plainly whether we think it is ready.
