# Analysis of call center employee performance

This code basically calculates a few metrics based on numerators and denominators and then it bins the data into quintiles. This is a common operation that I perform for various data sets. My issue is that I always have to rename the metrics and columns.

I have a feeling that I could create functions to make this more general perhaps. The common operations I am doing are:

1. % of total numerator
2. delta from a target
3. attainment (actual divided by the target)
4. impact (the weighted delta from the target, how much a person is increasing or decreasing the total score)
5. The quintiles (I have them labeled so the end-user understands what they are. These files are used in Excel for pivoting how the top and bottom employees perform)

I would like to have it more general: input a numerator and denominator. Instead of global variables, perhaps they could be input into a function?

csat_target =  0.75
fcr_target = 0.75
nps_target = 0.25
aht_target = 5.25

csat_weight = 0.30
fcr_weight = 0.30
nps_weight = 0.40
aht_weight = 0.00

df['fcr'] = df['final_hits'] / df['total_hits']
df['csat'] = df['csat_score_weight'] / df['survey_count']
df['nps'] = (df['promoters'] - df['detractors']) / df['survey_count']
df['aht'] =  (df['total_time'] / df['total_calls'])
df['goal_attainment'] = ((df['nps'] / nps_target) * nps_weight) + ((df['csat'] / csat_target) * csat_weight) + ((df['fcr'] / fcr_target) * fcr_weight)

df['fcr_volume'] = df['total_hits'] / df['total_hits'].sum()
df['fcr_delta'] = df['fcr'] - fcr_target
df['fcr_impact'] = (df['fcr_delta'] * df['fcr_volume'])

df['survey_volume'] = df['survey_count'] / df['survey_count'].sum()
df['nps_delta'] = nps_target - df['nps']
df['nps_impact'] = df['nps_delta'] * df['survey_volume']
df['overall_nps_impact'] = np.where(df['survey_count'] > 0, df['nps_impact'], 0)

df['csat_delta'] = df['csat'] - csat_target
df['csat_impact'] = -(df['csat_delta'] * df['survey_volume'])
df['overall_csat_impact'] = np.where(df['survey_count'] > 0, df['csat_impact'], 0)

df.replace([np.inf, -np.inf], np.nan, inplace = True)

df['call_volume'] = df['total_calls'] / df['total_calls'].sum()
df['aht_delta'] = df['aht'] - aht_target
df['aht_impact'] = (df['aht_delta'] * df['call_volume'] )

df['fcr_weight'] = np.where(df['total_hits'] > 0, fcr_weight, 0)
df['nps_weight'] = np.where(df['survey_count'] > 0, nps_weight, 0)
df['csat_weight'] = np.where(df['survey_count'] > 0, csat_weight, 0)

df['fcr_dfile'] = pd.qcut(df['fcr_impact'], 5, labels = ["D1", "D2", "D3", "D4", "D5"])
df['nps_dfile'] =  pd.qcut(df['overall_nps_impact'] , 5, labels = ["D1", "D2", "D3", "D4", "D5"])
df['csat_dfile'] = pd.qcut(df['overall_csat_impact'] , 5, labels = ["D1", "D2", "D3", "D4", "D5"])
df['aht_impact_dfile'] = pd.qcut(df['aht_impact'], 5, labels = ["D1", "D2", "D3", "D4", "D5"])
df['aht_dfile'] = pd.qcut(df['aht'], 5, labels = ["D1", "D2", "D3", "D4", "D5"])


Yes, functions would be a splendid idea. However a lot of the calculations aren't repeated, so it's hard to see which parts could be factored out usefully.

Apart from that I'd suggest putting the *_target and *_weight variables into a dictionary or another data frame so that they can be referred to by name, making it more easy to use them from functions:

parameters = pandas.DataFrame({
'target': [0.75, 0.75, 0.25, 5.25],
'weight': [0.30, 0.30, 0.40, 0.00]
}, index = ['csat', 'fcr', 'nps', 'aht']).transpose()


With the data frame above it's easy to select just the parameters for "csat" for example, using parameters.csat.target or .weight.

That also makes certain queries more generic:

df['goal_attainment'] = sum(df[name] / parameters[name].target * parameters[name].weight
for name in parameters if name not in {"aht"})


A couple of frequently used queries can also be reused, e.g. df['survey_count'] should likely be cached, similarly for the comparisons:

survey_count = df['survey_count']
positive_survey_count = df['survey_count'] > 0


(Maybe take a closer look to the Pandas API, e.g. how the data could be structured with nested names and multi indexes.)