There's probably a lot you could do here to avoid so much nested looping.
Given the plans are related to groups, you can start your loop of each group by doing a routine to get all the group members using the Office 365 Groups or Entra ID connector - this will avoid having to do an API call for each assignee within the loop later.
Also, you may benefit from using a Select action to transform your arrays instead of appending objects to an array variable within a loop. Variables are quite resource heavy in the backend vs using data operations such as Compose and Select.
My Approach to this would be:
For each group:
Get all users in the group
Get all plans for the group
For each plan
List tasks
List buckets
Instead of looping round each bucket and appending to an array variable, use a Select action to transform the array
For each Task, instead of getting the user profile (I assume this is the created by/owner of the task?) you can filter your array of group members from earlier
Use the same technique to join bucket info to the tasks.
N.B You can have a Compose inside an Apply to each loop (let's say it's called Inner Compose), then below the loop you can add another Compose and put the expression outputs('Inner_Compose') there, and the result will be an array with the contents of each compose from inside the loop for its items. This is waaay quicker than appending to a variable - however, it doesn't work in nested loops.
Basically your goal is to have as few loops as possible - use Select and join using xpath. Keep any loops you must have to a single layer to produce a bunch of arrays by referencing a Compose inside the loop using another compose and thereby avoiding array variables. Then join the resultant arrays together at the end using Select and xpath.
When you stop using variables and nested loops, you can set your loop concurrency right up to 50.
This should speed your flow up massively.
Even if you absolutely can't avoid those loops, it's much better to have an array of users ready you can filter rather than making an API call for each instance of the loop, especially when many of the tasks are going to have the same assignees, so you're looking up the same user's details from the API many times over during each run.
