You are on page 1of 10

HR Analytics at ScaleneWorks

Behavioral modeling to predict renege

Case Report

Advanced Business Analytics

Submitted on February 15, 2017


HR Analytics at ScaleneWorks – Case Report

Table of Contents
Introduction ................................................................................................................................ 2
Firm situation and context....................................................................................................... 2
Data set .................................................................................................................................. 3
Analysis ..................................................................................................................................... 5
Conclusions ............................................................................................................................... 7

1|Page
HR Analytics at ScaleneWorks – Case Report

Introduction
Firm situation and context
ScaleneWorks People Solutions LLP (ScaleneWorks) commenced its operations in Bangalore

the summer of 2010, providing talent acquisition services to its customers. The business started

when three HR Practitioners: Sanjay Shelvankar, Ashish Tiwari and Sharon George decided to

work together to create one of the best talent acquisition companies in the world (Rahuk &

Dinesh, 2016). The three co-founders came from different backgrounds: tech, consulting, talent

acquisition, and marketing. Sanjay is the CEO, and is looking at ways to improve operations of

the company.

As the company started operations, a problem became apparent. The recruiting process was

starting to cost the company a lot of resources, as a considerable number of applicants where

reneging their offers. In some cases, 15-35% of applicants deciding to renege their offers

(Rahuk & Dinesh, 2016). This was a major issue for ScaleneWorks because at 30% renege

rate, 12,000 offers roller out every year with 15-man hours per candidate, resulted in 54,000

man hours gone to waste (Rahuk & Dinesh, 2016). Talent acquisition lifecycle involves a large

number of agencies, and minimizing it would be very beneficial to the company.

Recruiting follows the traditional talent acquisition lifecycle (TALC). The process runs as follows:

Sourcing, screening, selection, fitment & offer, and post offer follow-up (POFU). During the

sourcing process, candidates are identified from a range of: networking circles to

vendors/consultants. The screening process is done by HR executives that look for: notice

periods, gap in education, previous companies of employment, etc. Selection involves

performing several rounds of interviews with the candidates. Fitment and offer is a calculation

based on the interviews in the selection process, if the candidate deemed fit, they are given an

offer. POFU is the final stage and involves the HR executive to contact the selected candidate

2|Page
HR Analytics at ScaleneWorks – Case Report

and ensure that candidate joins the company after completion of the notice period served to the

parting company.

In the case a candidate renege’s their contract, the customer has a chance of not receiving their

client. This can be problematic because a candidate has already been promised to the client,

and preparations for the candidate joining their firm has begun. It isn’t in ScaleneWorks best

interest to tell the client the bad news, so a replacement has to be found. This creates a

scenario where a not ideal candidate might be selected for the job, and might affect next time

the client looks for new recruits. In either scenario, ScaleneWorks future contracts from that

client will likely be negatively impacted.

Sanjay is looking to use a data set of candidates to look at preventing future renege. He wants

to analyze key drivers that influence a candidate joining a firm. He wants to look for rules that

can be implemented to predict the renege. He also wants to create a predictive algorithm that

calculate the probability a particular candidate will not renege on their offer, post acceptance

stage.

Data set

ScaleneWorks has collected 12,000 records of candidates over a year. The data covers the

candidate reference number, duration to accept offer, notice period, offered band, percentage

hike expected/offered/difference, joining bonus, gender, candidate source, relevant years

experience, line of business, candidates date of birth, joining location, candidate relocation

status and final joining status of candidate.

3|Page
HR Analytics at ScaleneWorks – Case Report

The data set consists of 12,333 observations, and some columns of data’s have not applicable

(NA’s), which might affect the results. The duration to offer has 2,719 NA’s which translates into

22% of missing data. Percent hike expected, offered and difference have between 596 - 851

NA’s which translates into 4.8% - 6.9% of missing data. This is minimal missing data, and the

regression should have no problem.

Looking at the histogram below, on the horizontal axis 0 indicated a candidate joining the

company and 1 indicating renege. This confirms Sanajay’s problem, as the renege is a large

portion of the data set.

Figure 1: Histogram of renege

4|Page
HR Analytics at ScaleneWorks – Case Report

Analysis
The analysis can be started with the data set being loaded into Rstudio. The train and test data

is generated based on a 75% samples size. Using the test data, the significant coefficients can

be seen in table 1 below (see Appendix-A for full regression).

Coefficients: Estimate std. Error z value Pr(>|z|)


(intercept) -1.37E+00 1.22E+00 -1.127 0.259653
Duration.to.accept.offer 6.35E-03 1.53E-03 4.138 3.51E-05
Notice.period -2.43E-02 1.47E-03 -16.464 <2e-16
Offered.bandEl 1.11E+00 2.02E-01 5.504 3.72E-08
Offered.bandE2 9.75E-01 2.21E-01 4.407 1.05E-05
Offered.bandE3 1.18E+00 2.87E-01 4.107 4.00E-05
Percent.hike.offered.in.CTC 2.66E-03 1.29E-03 2.067 0.038771
candidate.sourceDirect 3.52E-01 6.89E-02 5.11 3.21E-07
candidate.sourceEmployee Referral 7.07E-01 9.96E-02 7.095 1.29E-12
LOBBFSI 4.73E-01 1.47E-01 3.213 0.001312
LOBERS 3.18E-01 1.39E-01 2.285 0.022286
LOBETS 5.05E-01 1.64E-01 3.088 0.002016
LOBINFRA 8.71E-01 1.51E-01 5.786 7.20E-09
Age 3.60E-02 9.63E-03 3.743 0.000182

Table 1: Significant coefficients

To determine the validity of the data assessment of training and testing error was done. After

performing a cut of 0.5, the training error resulted in 22% and testing error in 57%. These results

demonstrate that there is some over-fitting in the testing data.

Key Questions
Using the data above, Sanjay’s key questions can be answered. Sanjay wants to identify key

drivers for candidates joining a firm, rules that can be implemented to predict renege, and he

wants to create a predictive algorithm to predict the renege.

Analyzing the data in Table 1 shows several key indicators for someone that will likely renege

their offer. Key indicators are: Duration to accept offer, Notice period, Offered band 1-3, Percent

hike offered by company, Resume being sourced from direct/employee referral, Line of

businesses BFSI/ERS/ETS/INFRA, and the age of candidate. Duration to accept offer can be

5|Page
HR Analytics at ScaleneWorks – Case Report

explained by the candidates having more time to look at other options, and possibly reneging

his initial offer. Notice period can be explained by the fact that having a longer period allows the

candidate more time to re-negotiate with his current employer and go back on his new offer. The

three starting offer bands are the best indicators, as they have the highest coefficient. Pay

bands can be explained by the candidate being easier tempted with a higher offer elsewhere. A

higher percent hike in cost to company (CTC) can be explained to higher renege, because CTC

may be more generous than the candidate’s current job, but the candidate might feel they are

worth more. The sources of new candidate through direct and employee referral can be

explained by biased opinions of the recruiting individuals, and could lead to flawed candidates.

Line of business in BFSI, ERS, ETS, and INFRA can be explained to be a higher competitive

environment, where it might be common to have more bargaining. Finally age can play a big

part as more senior candidates will have more choices and have a higher chance to change

their mind if a better offer is presented.

To help identify renege, rules can be implemented in the recruitment process. A new system

can be implemented that weights candidates more heavily based on the criteria outlined in the

key drivers. In the initial stage of sourcing, candidates that are found directly or through

employee referral can be flagged to go through extra hard screening and weighed more

unfavorably. When going through the screening process, the line of businesses from BFSI,

ERS, ETS, and INFRA and higher age, can be more thoroughly processed, and attaching a

heavier weight in the new system. Moving into the fit and offer stage, the three lower bands of

offers can be removed, so the initial starting band is category 4, the candidates weighing heavily

with key indicators should be removed by this point. In the stage of POFU, a reduction in time to

accept the offer and reducing the notice periods for the candidate to leave their current

employer would help reduce renege.

6|Page
HR Analytics at ScaleneWorks – Case Report

Sanjay can use the following formula to determine potential renege: y= -1.372 +

0.00635*(Duration.to.accept.offer) - 0.02427*( Notice.period) + 1.10900*( Offered.bandEl) +

0.97450*(Offered.bandE2) + 1.17700*(Offered.bandE3) +

0.00266*(Percent.hike.offered.in.CTC) + 0.35230*(candidate.sourceDirect) +

0.70660*(candidate.sourceEmployee Referral) + 0.47270*(LOBBFSI) + 0.31840*(LOBERS) +

0.50510*(LOBETS) + 0.87100*(LOBINFRA) + 0.03603*(Age).

Conclusions
In conclusion, Sanjay and his team have a few options on how they can process potential

candidates. Firstly a new weighing system can be implemented to help the screening stage,

where factors such as how the candidates was found and from which kind of industry they are

from. Once the candidate gets to the selection stage, the factors can be weighed heavily against

them. In the fitment and offer stage, the three lowest and the percent hike offered should be

lowered, to maintain happier candidates and reduce renege. Latstly, the POFU stage can take

extra effort in increasing duration to accept offer as well as reducing the notice period the

candidate must give to his old employer.

This analysis has had some draw backs. During the training and testing error validation, a

difference between the two demonstrated over fitting. Another analysis with a different mix of

train data with a new regression might solve that problem. Another short coming was the

missing data for duration to offer and percent hike expected, which might have skewed our data.

7|Page
HR Analytics at ScaleneWorks – Case Report

Bibliography
Rahuk, K., & Dinesh, K. (2016). HR Analytics at ScaleneWorks. Indian Institute of Management
Bangalore, 12.

Appendix
Call:
glm(formula = y ~ DOJ.Extended + Duration.to.accept.offer + Notice.period +
Offered.band + Percent.hike.offered.in.CTC + Pecent.hike.expected.in.CTC
+
Percent.hike.offered.in.CTC + Joining.Bonus + Candidate.relocate.actual +
Gender + Candidate.Source + Rex.in.Yrs + LOB + Location +
Age, family = "binomial", data = train)

Deviance Residuals:
Min 1Q Median 3Q Max
-1.64581 -0.68898 -0.49849 -0.00012 2.72058

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.372e+00 1.217e+00 1.127 0.259653
DOJ.ExtendedYes -7.247e-02 6.606e-02 -1.097 0.272609
Duration.to.accept.offer -6.346e-03 1.534e-03 -4.138 3.51e-05 ***
Notice.period 2.427e-02 1.474e-03 16.464 < 2e-16 ***
Offered.bandE1 -1.109e+00 2.016e-01 -5.504 3.72e-08 ***
Offered.bandE2 -9.745e-01 2.211e-01 -4.407 1.05e-05 ***
Offered.bandE3 -1.177e+00 2.866e-01 -4.107 4.00e-05 ***
Percent.hike.offered.in.CTC -2.656e-03 1.285e-03 -2.067 0.038771 *
Pecent.hike.expected.in.CTC 1.890e-03 1.452e-03 1.302 0.193080
Joining.BonusYes 2.078e-01 1.494e-01 1.391 0.164325
Candidate.relocate.actualYes -1.718e+01 1.763e+02 -0.097 0.922371
GenderMale 1.444e-01 8.154e-02 1.771 0.076569 .
Candidate.SourceDirect -3.523e-01 6.893e-02 -5.110 3.21e-07 ***
Candidate.SourceEmployee Referral -7.066e-01 9.959e-02 -7.095 1.29e-12 ***
Rex.in.Yrs 3.886e-02 2.091e-02 1.858 0.063150 .
LOBBFSI -4.727e-01 1.471e-01 -3.213 0.001312 **
LOBCSMP -2.814e-01 1.670e-01 -1.686 0.091886 .
LOBEAS 1.132e-01 1.865e-01 0.607 0.544092
LOBERS -3.184e-01 1.393e-01 -2.285 0.022286 *
LOBETS -5.051e-01 1.636e-01 -3.088 0.002016 **
LOBHealthcare -5.376e-01 2.758e-01 -1.950 0.051221 .
LOBINFRA -8.710e-01 1.505e-01 -5.786 7.20e-09 ***
LOBMMS -1.782e+01 1.768e+03 -0.010 0.991959
LocationBangalore -8.638e-01 1.165e+00 -0.742 0.458303
LocationChennai -7.641e-01 1.164e+00 -0.657 0.511423
LocationCochin -8.159e-01 1.578e+00 -0.517 0.605085
LocationGurgaon -9.036e-01 1.184e+00 -0.763 0.445403
LocationHyderabad -9.712e-01 1.172e+00 -0.828 0.407415
LocationKolkata -9.576e-01 1.190e+00 -0.805 0.421092
LocationMumbai -1.183e+00 1.185e+00 -0.998 0.318237
LocationNoida -1.215e+00 1.163e+00 -1.045 0.296132
LocationOthers -1.724e+01 1.679e+03 -0.010 0.991808
LocationPune -6.602e-01 1.226e+00 -0.538 0.590239

8|Page
HR Analytics at ScaleneWorks – Case Report

Age -3.603e-02 9.628e-03 -3.743 0.000182 ***


---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 8216.2 on 8681 degrees of freedom


Residual deviance: 7058.7 on 8648 degrees of freedom
(567 observations deleted due to missingness)
AIC: 7126.7

Number of Fisher Scoring iterations: 17

9|Page

You might also like