Script 807: SBA Campaign Pacing

Purpose

SBA Campaign Budget Pacing - Minimize Lost IS (Budget)

To Elaborate

The Python script solves the problem of allocating budgets to campaigns in order to minimize lost impression share (budget). It takes into account various factors such as remaining budget, remaining weekdays in the month, historical spend and spend potential, campaigns with spend in the lookback period, and a minimum daily budget. The goal is to allocate the budget in a way that maximizes impression share while staying within the budget constraints.

Walking Through the Code

The script starts by defining a configurable parameter for the minimum daily budget.
It then checks if the code is running on a server or locally.
If running on a server, it skips the initialization step. If running locally, it loads the dataSourceDict from a pickled file.
The script sets up the necessary imports and defines column constants.
It checks the data types of the input DataFrame and converts the allocation columns to percent strings if they are floats.
The script coerces the “Program End Date” column into a date type.
It sets the input DataFrame as the index using the “SBA Strategy” column and groups it by the “SBA Strategy” column.
The script calculates the full-potential spend by adjusting the historical spend based on the lost impression share due to budget.
It removes date segmentation and calculates the MTD spend for each campaign.
The script aggregates the data by grouping it based on certain columns and applying aggregation functions.
It excludes campaigns that are not active, have no spend in the lookback period, or have a program end date in the past.
The script calculates the budget allocation ratio by capping the full-potential spend and comparing it to the total spend within the same Salesforce Item ID budget group.
It calculates the remaining budget for each SBA Strategy budget group by subtracting the MTD spend from the SBA monthly budget.
The script allocates the remaining budget to each campaign according to the allocation ratio calculated above.
It calculates the next day’s daily budget by dividing the allocated budget by the number of business days left in the current month.
The script applies a minimum rule to bump the allocated budget above the minimum if necessary.
It handles traffic budget by copying the budgets over and blanking out the budget for non-traffic campaigns.
The script calculates the Salesforce Item ID-level pacing compliance percentage by dividing the MTD spend by the prorated total budget.
It checks if the DataFrame is empty before performing the calculation and assigns the calculated pacing to the corresponding column.
The script generates the output DataFrame by selecting the changed campaigns and renaming the columns.
It prints the shape and first few rows of the output DataFrame if there are changes detected. Otherwise, it returns an empty DataFrame.

Vitals

Script ID : 807
Client ID / Customer ID: 1306927165 / 60270223
Action Type: Bulk Upload
Item Changed: Campaign
Output Columns: Account, Campaign, Daily Budget, Rec. Daily Budget, SBA Allocation, SBA Budget Pacing
Linked Datasource: M1 Report
Reference Datasource: None
Owner: dwaidhas@marinsoftware.com (dwaidhas@marinsoftware.com)
Created by dwaidhas@marinsoftware.com on 2024-03-13 19:18
Last Updated by dwaidhas@marinsoftware.com on 2024-03-14 14:39

> See it in Action

Python Code

#
# SBA Campaign Budget Pacing - Minimize Lost IS (Budget)
#
# Allocates according to:
# * Remaining budget for each Salesforce Item ID Budget Group
# * Remaining weekdays in month
# * Historical spend and spend potential
# * Campaigns with spend in lookback period
# * Minimum daily budget
#
# Author: Dana Waidhas
#
# Created: 2024-02-26
#

##### Configurable Param #####


MINIMUM_DAILY_BUDGET = 10


##############################

########### START - Local Mode Config ###########
# Step 1: Uncomment download_preview_input flag and run Preview successfully with the Datasources you want
download_preview_input=True
# Step 2: In MarinOne, go to Scripts -> Preview -> Logs, download 'dataSourceDict' pickle file, and update pickle_path below
# pickle_path = ''
pickle_path = '/Users/mhuang/Downloads/pickle/avb_marketing_datasource_dict_1702622906522.pkl'
# Step 3: Copy this script into local IDE with Python virtual env loaded with pandas and numpy.
# Step 4: Run locally with below code to init dataSourceDict


# determine if code is running on server or locally
def is_executing_on_server():
    try:
        # Attempt to access a known restricted builtin
        dict_items = dataSourceDict.items()
        return True
    except NameError:
        # NameError: dataSourceDict object is missing (indicating not on server)
        return False

if is_executing_on_server():
    print("Code is executing on server. Skip init.")
elif len(pickle_path) > 3:
    print("Code is NOT executing on server. Doing init.")
    # load dataSourceDict via pickled file
    import pickle
    dataSourceDict = pickle.load(open(pickle_path, 'rb'))

    # print shape and first 5 rows for each entry in dataSourceDict
    for key, value in dataSourceDict.items():
        print(f"Shape of dataSourceDict[{key}]: {value.shape}")
        # print(f"First 5 rows of dataSourceDict[{key}]:\n{value.head(5)}")

    # set outputDf same as inputDf
    inputDf = dataSourceDict["1"]
    outputDf = inputDf.copy()

    # setup timezone
    import datetime
    # Chicago Timezone is GMT-5. Adjust as needed.
    CLIENT_TIMEZONE = datetime.timezone(datetime.timedelta(hours=-5))

    # import pandas
    import pandas as pd
    import numpy as np

    # other imports
    import re
    import urllib

    # import Marin util functions
else:
   from marin_scripts_utils import tableize, select_changed
   print("Running locally but no pickle path defined. dataSourceDict not loaded.")
   exit(1)
########### END - Local Mode Setup ###########


RPT_COL_CAMPAIGN = 'Campaign'
RPT_COL_DATE = 'Date'
RPT_COL_ACCOUNT = 'Account'
RPT_COL_PUBLISHER_NAME = 'Publisher Name'
RPT_COL_STRATEGY = 'Strategy'
RPT_COL_DAILY_BUDGET = 'Daily Budget'
RPT_COL_CAMPAIGN_STATUS = 'Campaign Status'
RPT_COL_PUB_COST = 'Pub. Cost R$'
RPT_COL_CLICKS = 'Clicks'
RPT_COL_CONV = 'Conv.'
RPT_COL_IMPR_SHARE = 'Impr. share %'
RPT_COL_LOST_IMPR_SHARE_BUDGET = 'Lost Impr. Share (Budget) %'
RPT_COL_LOST_IMPR_SHARE_RANK = 'Lost Impr. Share (Rank) %'
RPT_COL_SBA_STRATEGY = 'SBA Strategy'
RPT_COL_SBA_CAMPAIGN_BUDGET = 'SBA Campaign Budget'
RPT_COL_SBA_ALLOCATION = 'SBA Allocation'
RPT_COL_REC_DAILY_BUDGET = 'Rec. Daily Budget'
RPT_COL_SBA_BUDGET_PACING = 'SBA Budget Pacing'
RPT_COL_SBA_TRAFFIC = 'SBA Traffic'
RPT_COL_PROGRAM_END_Date = 'Program End Date'


BULK_COL_ACCOUNT = 'Account'
BULK_COL_CAMPAIGN = 'Campaign'
BULK_COL_DAILY_BUDGET = 'Daily Budget'
BULK_COL_SBA_ALLOCATION = 'SBA Allocation'
BULK_COL_SBA_BUDGET_PACING = 'SBA Budget Pacing'
BULK_COL_REC_DAILY_BUDGET = 'Rec. Daily Budget'


COL_SPEND_FULL_POTENTIAL = 'spend_lookback_full_potential'
COL_SPEND_FULL_POTENTIAL_CAPPED = 'spend_lookback_full_potential_capped'
COL_SPEND_MTD = 'spend_mtd'
COL_SBA_ALLOCATION_NEW_FLOAT = RPT_COL_SBA_ALLOCATION + '_new_float'
COL_SBA_ALLOCATION_NEW = RPT_COL_SBA_ALLOCATION + '_new'
COL_SBA_STRATEGY_BUDGET_REMAINING = 'SBA_Campaign_budget_remaining'
COL_SBA_BUDGET_PACING_NEW = RPT_COL_SBA_BUDGET_PACING + '_new'
COL_BUDGET_REMAINING = 'budget_remaining'
COL_DAILY_BUDGET_NEW = RPT_COL_DAILY_BUDGET + '_new'
COL_REC_DAILY_BUDGET_NEW = RPT_COL_REC_DAILY_BUDGET + '_new'
COL_DAYS_REMAINING= 'weekdays_remaining'
COL_DAYS_TOTAL= 'weekdays_total'
COL_PACING_CALC = 'pacing_calc'

outputDf[BULK_COL_DAILY_BUDGET] = "<<YOUR VALUE>>"


today = datetime.datetime.now(CLIENT_TIMEZONE).date()
print("inputDf shape", inputDf.shape)
print("inputDf dtypes", inputDf.dtypes)


# change back to percent string
if inputDf[RPT_COL_SBA_ALLOCATION].dtype == "float":
    inputDf[RPT_COL_SBA_ALLOCATION] = round(inputDf[RPT_COL_SBA_ALLOCATION] * 100.0, 0).astype(str) + '%'
if inputDf[BULK_COL_SBA_BUDGET_PACING].dtype == "float":
    inputDf[BULK_COL_SBA_BUDGET_PACING] = round(inputDf[BULK_COL_SBA_BUDGET_PACING] * 100.0, 0).astype(str) + '%'

# coerce Program End Date into Date type
inputDf[RPT_COL_PROGRAM_END_Date] = pd.to_datetime(inputDf[RPT_COL_PROGRAM_END_Date], errors='coerce')


inputDf = inputDf.set_index([RPT_COL_SBA_STRATEGY])
group_by_salesforce_item_ID = inputDf.groupby(RPT_COL_SBA_STRATEGY)


# ## Calculate Full-Potential Spend
# * Adjust Historical Spend by _Lost Impression Share due to Budget_ (see [Formula](https://docs.google.com/document/d/1EbCQ5z9Up8TZ6GISEeCaRSB3Fc15vCPCfeIydree23M/edit#bookmark=id.5fsx7jlseze6))
#

adj_ratio = 1 + (inputDf[RPT_COL_LOST_IMPR_SHARE_BUDGET] / (1 - inputDf[RPT_COL_LOST_IMPR_SHARE_BUDGET]))


inputDf[COL_SPEND_FULL_POTENTIAL] = round(inputDf[RPT_COL_PUB_COST] * adj_ratio, 2)


# ## Remove Date Segmentation
# * Calculate MTD Spend

# SUM Series with Date index and only includes current month
def current_month_sum(x):
    x = x.sort_index()
    mtd = x[ (x.index.month == today.month) & (x.values > 0)]
    return mtd.sum()

groupby_cols = [ \
    RPT_COL_SBA_STRATEGY, \
    RPT_COL_STRATEGY, \
    RPT_COL_PUBLISHER_NAME, \
    RPT_COL_ACCOUNT, \
    RPT_COL_CAMPAIGN, \
]

agg_spec = {
    RPT_COL_CAMPAIGN_STATUS: 'last', \
    RPT_COL_DAILY_BUDGET: 'last', \
    RPT_COL_SBA_CAMPAIGN_BUDGET: 'last', \
    RPT_COL_SBA_ALLOCATION: 'last', \
    RPT_COL_REC_DAILY_BUDGET: 'last', \
    RPT_COL_SBA_BUDGET_PACING: 'last', \
    RPT_COL_SBA_TRAFFIC: 'last', \
    RPT_COL_CLICKS: 'sum', \
    RPT_COL_CONV: 'sum', \
    RPT_COL_PUB_COST: 'sum', \
    COL_SPEND_MTD: current_month_sum, \
    COL_SPEND_FULL_POTENTIAL: 'sum', \
    RPT_COL_PROGRAM_END_Date: 'last', \
}

inputDf[COL_SPEND_MTD] = inputDf[RPT_COL_PUB_COST]


df_campaign_agg = inputDf.reset_index() \
                         .set_index(RPT_COL_DATE) \
                         .groupby(groupby_cols) \
                         .agg(agg_spec) \
                         .reset_index() \
                         .set_index(RPT_COL_SBA_STRATEGY)

# ## Only allocate budget for recently trafficking campaigns
# * Exclude Campaigns that are:
# ** not ACTIVE
# ** without spend in lookback period
# ** Program Date is in the past


inactive_campaigns = (df_campaign_agg[RPT_COL_CAMPAIGN_STATUS] != 'Active') & (df_campaign_agg[RPT_COL_PUB_COST] == 0)
expired_campaigns = df_campaign_agg[RPT_COL_PROGRAM_END_Date].notnull() & (df_campaign_agg[RPT_COL_PROGRAM_END_Date] < pd.to_datetime(today))
df_campaign_agg = df_campaign_agg.loc[ ~(inactive_campaigns | expired_campaigns) ]



# ## Calculate Budget Allocation Ratio
# * Cap full potential spend at 2X (don't spend twice as much as before)
# * Compare full potential spend for each campaign to total spend within same SALESFORCE_ITEM_ID budget group


df_campaign_agg[COL_SPEND_FULL_POTENTIAL_CAPPED] = df_campaign_agg \
    .apply(lambda row: min(row[COL_SPEND_FULL_POTENTIAL], 2 * row[RPT_COL_PUB_COST]), axis=1)


# use transform to calculate sum for each SALESFORCE_ITEM_ID and make it available on every row
# note: no need to build aggregate DataFrame and JOIN back to original


df_campaign_agg[COL_SBA_ALLOCATION_NEW_FLOAT] = 100.0 * \
        df_campaign_agg[COL_SPEND_FULL_POTENTIAL_CAPPED] / \
        df_campaign_agg.groupby(RPT_COL_SBA_STRATEGY)[COL_SPEND_FULL_POTENTIAL_CAPPED].transform('sum')


df_campaign_agg[COL_SBA_ALLOCATION_NEW] = round(df_campaign_agg[COL_SBA_ALLOCATION_NEW_FLOAT],0).astype(str) + '%'

#
# ## Calculate Remaining Budget
# * For each SBA Strategy budget group, calculate how much Budget is left by substracting SBA Monthly budget from MTD SALESFORCE_ITEM_ID spend


df_campaign_agg[COL_SBA_STRATEGY_BUDGET_REMAINING] =  \
        df_campaign_agg[RPT_COL_SBA_CAMPAIGN_BUDGET] - \
        df_campaign_agg.groupby(by=[RPT_COL_SBA_STRATEGY])[COL_SPEND_MTD].sum()


# ## Allocate Budget
# * Allocate remaining budget to each campaign according to ratio calculated above


df_campaign_agg[COL_BUDGET_REMAINING] = round(df_campaign_agg[COL_SBA_STRATEGY_BUDGET_REMAINING] * df_campaign_agg[COL_SBA_ALLOCATION_NEW_FLOAT] / 100.0, 1)


# ## Calculate SBA Daily Budget
#
# * Calcualte next day Daily Budget by dividing allocated budget by number of Business Days left in the current month

today_numpy = pd.to_datetime(today).to_numpy().astype('datetime64[D]')
next_month_start = (today_numpy + pd.offsets.BMonthBegin()).to_numpy().astype('datetime64[D]')


# for months ending on weekends, use max(1,x) to avoid dividing by zero
days_left = max(1, (next_month_start - today_numpy).astype('timedelta64[D]').astype(int))


df_campaign_agg[COL_DAYS_REMAINING] = days_left
df_campaign_agg[COL_REC_DAILY_BUDGET_NEW] = round(df_campaign_agg[COL_BUDGET_REMAINING] / days_left, 0)


# ### Apply Minimum Rule
# * Bump allocated budget above minimum


allocated_below_min = (df_campaign_agg[COL_REC_DAILY_BUDGET_NEW] < MINIMUM_DAILY_BUDGET)
df_campaign_agg.loc[allocated_below_min, COL_REC_DAILY_BUDGET_NEW] = MINIMUM_DAILY_BUDGET


# ### Traffic Budget
df_campaign_agg[COL_DAILY_BUDGET_NEW] = np.nan


# campaigns to traffic
to_traffic = df_campaign_agg[RPT_COL_SBA_TRAFFIC].notnull() & \
            (df_campaign_agg[RPT_COL_SBA_TRAFFIC].astype(str).str.lower() == 'traffic')
print("Not weekend. Traffic count", to_traffic.sum())


# copy budgets over
df_campaign_agg[COL_DAILY_BUDGET_NEW] = df_campaign_agg[COL_REC_DAILY_BUDGET_NEW]
# then blank out budget for non-traffic campaigns
df_campaign_agg.loc[~to_traffic, COL_DAILY_BUDGET_NEW] = np.nan

# ## Calculate Salesforece Item ID -level Pacing compliance percentage. Ideally should be 100% each day.


# number of elapsed workdays
current_month_start = pd.to_datetime(today.replace(day=1)).to_numpy().astype('datetime64[D]')
total_days_in_month = (next_month_start - current_month_start).astype('timedelta64[D]').astype(int)
df_campaign_agg[COL_DAYS_TOTAL] = total_days_in_month
prorated_ratio = (total_days_in_month - days_left) / total_days_in_month

print("today", today)
print("current_month_start", current_month_start)
print("next_month_start", next_month_start)
print("weekdays_in_month", total_days_in_month)
print("weekdays_left", days_left)
print("prorated_ratio", prorated_ratio)


# Check if the DataFrame is empty before performing the calculation
if not df_campaign_agg.empty:
    # divide MTD spend by prorated total budget
    mask = df_campaign_agg[RPT_COL_SBA_CAMPAIGN_BUDGET] > 0
    df_campaign_agg[COL_PACING_CALC] = round(100.0 * \
                                        df_campaign_agg.groupby(by=[RPT_COL_SBA_STRATEGY])[COL_SPEND_MTD].sum() / \
                                        (prorated_ratio * df_campaign_agg[RPT_COL_SBA_CAMPAIGN_BUDGET]), \
                                        0).astype(str) + '%'
    df_campaign_agg.loc[mask, COL_SBA_BUDGET_PACING_NEW] = df_campaign_agg.loc[mask, COL_PACING_CALC]
else:
    print("DataFrame df_campaign_agg is empty. Skipping calculation on line 365.")

# Debug DF with full details
df_SALESFORCE_ITEM_ID_budget = group_by_salesforce_item_ID[[RPT_COL_SBA_CAMPAIGN_BUDGET]].transform('max').dropna().drop_duplicates()
print("Salesforce Item ID budgets", df_SALESFORCE_ITEM_ID_budget.to_string())




# ## Generate outputDf


# Check for changes
changed = df_campaign_agg[COL_REC_DAILY_BUDGET_NEW].notnull() & \
    ( \
       (df_campaign_agg[RPT_COL_REC_DAILY_BUDGET] != df_campaign_agg[COL_REC_DAILY_BUDGET_NEW]) | \
       (df_campaign_agg[RPT_COL_DAILY_BUDGET] != df_campaign_agg[COL_DAILY_BUDGET_NEW]) | \
       (df_campaign_agg[RPT_COL_SBA_ALLOCATION] != df_campaign_agg[COL_SBA_ALLOCATION_NEW]) | \
       (df_campaign_agg[RPT_COL_SBA_BUDGET_PACING] != df_campaign_agg[COL_SBA_BUDGET_PACING_NEW]) \
    )


print("Changed rows:", changed.sum())


# Debug
debugDf = df_campaign_agg.loc[changed] \
      .reset_index() \
      .sort_values(by=[RPT_COL_SBA_STRATEGY, COL_DAILY_BUDGET_NEW, COL_REC_DAILY_BUDGET_NEW], ascending=False)


# print("debugDf", tableize(debugDf))


# Only emit output for changed campaigns
if changed.sum() > 0:


    # construct outputDf
    outputDf = df_campaign_agg.loc[changed, [RPT_COL_ACCOUNT, RPT_COL_CAMPAIGN, COL_DAILY_BUDGET_NEW, COL_REC_DAILY_BUDGET_NEW, COL_SBA_BUDGET_PACING_NEW, COL_SBA_ALLOCATION_NEW]] \
                      .copy() \
                      .rename(columns={ \
                            COL_DAILY_BUDGET_NEW: BULK_COL_DAILY_BUDGET, \
                            COL_REC_DAILY_BUDGET_NEW: BULK_COL_REC_DAILY_BUDGET, \
                            COL_SBA_BUDGET_PACING_NEW: BULK_COL_SBA_BUDGET_PACING, \
                            COL_SBA_ALLOCATION_NEW: BULK_COL_SBA_ALLOCATION, \
                        }) \
                      .reset_index() \
                      .sort_values(by=[RPT_COL_SBA_STRATEGY, BULK_COL_DAILY_BUDGET, BULK_COL_REC_DAILY_BUDGET], ascending=False) \
                      .drop(RPT_COL_SBA_STRATEGY, axis=1)


    print("outputDf shape", outputDf.shape)
    print("outputDf", tableize(outputDf.head()))
else:
    print("No changes detected, returning an empty dataframe")
    outputDf = pd.DataFrame(columns=[BULK_COL_ACCOUNT, BULK_COL_CAMPAIGN, BULK_COL_DAILY_BUDGET, BULK_COL_REC_DAILY_BUDGET, BULK_COL_SBA_BUDGET_PACING, BULK_COL_SBA_ALLOCATION])

Post generated on 2024-05-15 07:44:05 GMT

13 Mar 2024

« Script 805: Pacing Dimension Value from Campaign Name Script 809: SBA Auto Pause & Re enable Budget via Dim Tag »

MarinOne Scripts Creator's Corner