We are using an AI check to identify when a comment is left by a person on a task and not by a rule. However, this consumes many of our credits and we would like to find a way to reduce the overall usage.
On one of our dashboards, there is a rule to leave auto-follow up comments when a due date approaches and the ‘follow up’ field is set to yes. To help identify if there is activity on the task, we have a separate rule that is triggered when a comment is left and then checks if the comment was left by the system automation (the follow up rule). If the comment was left by an actual person, then the Follow-up field is cleared, and the due date is set.
We have multiple checks listed to attempt to narrow down when the rule needs to be run and included the AI check last. We are also using GPT-5.4 nano which is listed as the lowest cost. This rule runs between 40-60 times a week and consumes an average of 235 credits per task.
Does this AI check consume an abnormal number of credits? Does anyone have any recommendations to reduce either the number of times the rule is being run or the overall credit usage?
Model, context size and complexity of the task at hand are the three levers you can pull. By the sound of it you already apply all of them. 235 credits per run sounds like a reasonably low avg to me.
This is our first rule using AI, so I wanted to try and gain a better understanding of credit usage, thank you for clarifying. The only other area I can think of narrowing checks are time. The auto follow up rule runs at midnight, are you aware of any setup to prevent a rule from running at a certain time?
Hi @Isabella_Aullet!
At the moment, there isn’t a way to prevent a rule from running at a specific time like midnight.
However, you can work around this by adding time-based conditions to the rule (for example, only running during certain hours or days) so it only evaluates when needed.
Regarding the current situation, there’s a rule with a trigger called “Comment is added.”
If we can select who made the comment using “Check if,” it seems like we could create branching logic without using AI.
As an update, I toggled off the setting for the rule to be triggered by other rules. This also removed the need to use AI to identify who left a comment!