I would do it this way but feel free to modify my idea:

1. I will create a parallel hierarchy for dates which will have suffix _AVG90 (Example: 2024-08-26_AVG90)
2. Go to Integrator and add two variables type groovy (today ${Today}: return new Date().format("yyyy-MM-dd") and 90 days prior ${Today-90} return (new Date()-90).format("yyyy-MM-dd"))
3. Create an extract from the desired cube and make filtering on dates accept.inAlphaRange
4. Make sure you include empty cells so avg can be more accurate. These empty cells will be presented as null which you can replace by zeros in Groovy Function by just writing "return _input1" and check Null Default.
5. Create a Table aggregation where input will be your day as a constant value ${Today}-AVG90 (meaning you would like to write a value to this cell)
6. In Measure choose #Value and make sure aggregation is AVG.
7. Write this back to the cell and schedule the job to run every day after midnight.

Tadaaaa