Power of Statistical Analysis in Problem Solving - Six Sigma Way!

It is always interesting and insightful to do Process analysis using Hypothesis Testing in a Transaction Process: I encountered a scenario when I was trying to resolve Lead time reduction in “Customer Call Resolution” across multiple locations.

The challenge was unlike in manufacturing where the process parameters and other factors are standardized largely, here the process was personal dependent and to add more to the complexity I had to deal with multiple locations and team felt they all do the same process. But the data showing otherwise in terms of lead time of call closure showing variation between locations.

But I had a strong hunch ( to test though) that they were all doing the process differently and wanted to use hypothesis testing to check the how much of the variation is Statistically significance and get back to the team to unearth which location is dealing with the process differently ( since they kept saying that they all do similar way to deal with the call resolution)

I wanted to use the power of statistical analysis relevant to the context and help the team to appreciate, learn and resolve the problem at hand.

Here the story unfolds:

Problem Case: Investor complaint / call Resolution timely. Multiple locations to cater to the investor complaints ( 4 major metros). Average call resolution time in terms days approx. 100 days.

Goal: To reduce Lead time to close the complaint within 70 days.

Note: Too much of Variation and there were cases resolved more than 120 days

Box plot to show the multiple regions perspective in one visual chart:

Note: Analysis of Box-Plot

• 1.      observe the huge variability in the process between regions and too many extreme values.
• 2.      Kolkata Region-Less variation in handling the cases
• 3.      Chennai Region- Extreme data variation is less
• 4.      Mumbai and Delhi region- Extreme data variation is high compared to other regions

 Now comes the Hypothesis:

I just did two sample t Tests : Mumbai and Chennai one set and another one Kolkata and Chennai.

Mumbai Vs Chennai 

Null Hypothesis :

                          Both Mumbai and Chennai TAT same - no difference

Alternate Hypothesis :

                             Mumbai and Chennai TAT not same.

Two-Sample T-Test and CI: Mum TAT, Chn TAT

               N                        Mean StDev     SE Mean

Mum TAT  300    33.2    23.5            1.4

Chn TAT  300    39.3     31.8             1.8

Difference = mu (Mum TAT) - mu (Chn TAT)

T-Test of difference = 0 (vs not =): T-Value = -2.65 P-Value = 0.008   

P value is less than 0.05 hence reject the Null and conclude that  there is a difference between Chennai and Mumbai TAT. How much is the difference: Estimate for difference: -6.03- means Mumbai is doing 6 days earlier than Chennai

 Kolkatta Vs Chennai  

Null Hypothesis :

                          Both Kolkatta and Chennai TAT same - no difference

Alternate Hypothesis :

                             Kolkatta and Chennai TAT not same.

 Two-Sample T-Test : Kolkatta and Chennai results:

                            N       Mean      StDev  

Kol      300          23.7                  18.7     

Chen  300         39.3                    31.8    

Difference = mu (Kol) - mu (Chen)

95% CI for difference: (-19.70, -11.35)

T-Test of difference = 0 (vs not =): T-Value = -7.30 P-Value = 0.000  

Note:

•  Look at the p value which is less than .05 that means Kol and Chn TAT are not the same. They are different.
• 2.      Hence there is a statistical difference between Kolkatta and Chennai and Estimate for difference: -15.52 – which means Kolkata resolving 15 days earlier than Chennai
• Lets look at the practical difference between Kolkatta and Chennai

What I had just proved that there is a statistical significance. Then need to proceed to Practical significance as below.

Kolkata:

• Initial stage meeting won’t happen without further documents
• Delay in getting the response for the docs and other clarifications
• Manpower changes in complaint handling
• Ready to arrange for a meeting immediately by the members
• Pretty much same kind of complaint

Chennai:

1. We don’t wait during the initial stage – meeting happens without all documents in place
2. Fairly ok to get the docs better than Kolkata
3. Scheduling the meeting with the members delay
4. Critical and complicated cases

Solutions Broadly: Since enough data analysis done and insights i got, now moving to solutions is easy as below:

Control Stage: Monitored the results for couple of month post implementation and results collated as below.

Note: You can notice the before and after the implementation in terms of no. of days taken to resolve the complaints from customer. Observations as below

• Though the average remains same with few days of reduction the variability has been substantially reduced from 100 plus days to less than 80 and mostly the data points are around 70 days ( as per the target defined at the beginning of the project)