AI steps inside an automation
Sort responses into categories, pull out details, score leads, or draft a reply — with a path for when the AI is not sure.
Free-text answers are where the useful information hides and where rules stop working. AI steps read the response and turn it into something the rest of your automation can act on.
The four steps
- •Categorize with AI — sorts the response into one of your categories, so a Branch after it can send each category somewhere different.
- •Pull out details with AI — finds specific things in the text, like a company name or a budget, and hands them to the steps that follow.
- •Score the lead with AI — scores 0–100 against a description of what a good lead looks like to you.
- •Draft a reply with AI — writes a reply you can send, or hand to a person to check first.
Setting one up
- 1Add the AI step and choose which answers it should read.
- 2Tell it what you want in plain language — your categories, the details to look for, what makes a lead good.
- 3Use its result in later steps the same way you use an answer: in a Branch, in an email, in a contact update.
It tells you why, and when it is unsure
Every AI step reports its reasoning and how confident it is, and both show up in the run history next to that step. Each one also has a second exit called "Not sure". Set the confidence you need, and anything below it goes down that path instead of guessing.
The best thing to put on the "Not sure" path is a person — an approval step, or an email to whoever would have read it anyway. You keep the speed on the clear cases and lose nothing on the hard ones.
AI steps use credits each time they run, like the other AI features. The step shows what it costs before you save it, and it only spends when a real run reaches it.
Was this article helpful?