get past success rate at each timepoint using pandas

I have a dataframe like as below

data = {
    'project_id': [11, 22, 31, 45, 52, 61],
    'user_id': [10001, 10002, 10001, 10003, 10002, 10004],
    'customer_id': [20001, 20002, 20001, 20001, 20003, 20002],
    'project_start_date': ['2023-01-01', '2023-02-01', '2023-03-01', '2023-04-01', '2023-05-01', '2023-06-01'],
    'project_end_date': ['2023-01-31', '2023-02-28', '2023-03-31', '2023-04-30', '2023-05-31', '2023-05-10'],
    'product_name': ['Product A', 'Product B', 'Product C', 'Product A', 'Product B', 'Product C'],
    'success_flag': [0, 1, 1, 1, 0, 1]
}

df = pd.DataFrame(data)

I would like to do the below

a) for each user, compute two metrics - past_project_success_count, past_project_success_ratio

The logic to compute these two metrics is given below

past_project_success_count : count the number of past projects where the success_flag = 1
past_project_success_ratio : find the ratio using formula -> all past success projects (for each user) divided by all past projects (for each user)

b) Similarly, I would like to compute the same metrics for each customer

c) Similarly, I would like to compute the same metrics for each user and customer combo

d) we should only look at the past and compute these metrics. Should not consider the current row/current project. (only past projects)

I tried the below but not going anywhere with this

t1 = t1.sort_values('project_end_date')
t1['user_past_project_success_count'] = t1.groupby('user_id')['success_flag'].cumsum() / t1.groupby('user_id').cumcount()
t1['user_past_project_success_count'] = t1.groupby('customer_id')['success_flag'].cumsum() / t1.groupby('customer_id').cumcount()
t1['user_past_project_success_count'] = t1.groupby(['user_id','customer_id'])['success_flag'].cumsum() / t1.groupby(['user_id','customer_id']).cumcount()

one issue that I observed was sometime the cumcount returns the same value I don't know why. For past project count, it should just be in increasing order of 1 for each of the project.

I expect my output to be like as below. I have shown only for the user column. But we need to do the same for customer column and user. customer combo. Remember that first record for each group will always be zero as there is no past to look at.

This loop loops through the groups you want to group on and creates new columns with the statistics:

for name, cols in zip(
        ["user", "customer", "user_and_customer"],
        ["user_id", "customer_id", ["user_id", "customer_id"]]):
    # create new column for count
    df[f"past_successes_count_{name}"] = df.sort_values(
        "project_end_date").groupby(cols)["success_flag"].transform(
            # cumulative sum subtract current so only previous
            lambda x: x.cumsum() - x)
    # and another for ratio by dividing column above by cumulative count
    df[f"past_successes_ratio_{name}"] = df[f"past_successes_count_{name}"].div(
        (df.sort_values(
            "project_end_date").groupby(cols).cumcount()))
# fill any nan with 0
df.fillna(0, inplace=True)

Whilst .cumsum() includes the current row in calculations, thus the need for x.cumsum() - x, .cumcount() does not - so only uses 'past' data.