What is the best way to handle events that normally happen at the end of each month?

I have a time series where starting from the 28th of each month, the situation might change dramatically. There is a sharp increase on the 28th and then a sharp decrease in the next few days (29-31) as opposed to the values prior to 28th.

I have tried using monthly seasonality with a period of 30.5 and different values for Fourier order, but that didn't help. Adding custom regressors such as `is_28th` or `is_end_of_month` didn't work either. It seems that the model just doesn't take this information into consideration even though there is daily data for almost three years.

What else is possible in such cases?