DOE - Design of Experiments

WHAT IS DESIGN OF EXPERIMENTS?

A methodology proposed by Ronald A. Fisher, in his ground breaking book The Design of Experiments (1935). DOE is a systematic approach to establish relationships between factors affecting the output of a process. In a layman’s language it is about finding the causal relationships of factors with the process and its output. The information derived out of this methodology can be used to manage process outputs.

IDENTIFYING THE MOST APPROPRIATE DOE

Keep this flow chart or questioning technique in mind when about to tread on the DOE path.

DOE PROCESS IN A NUTSHELL

HOW TO CONDUCT DOE?

Design of Experiments: Using Conversation Analytics output to create an experiment
Real-time Case Study

BUSINESS QUESTIONS TO BE ANSWERED

• What are the various factors driving high Non-Talk?
• What specific topics to be addressed in order to reduce the Non-Talk?

(X) FACTORS PROVIDED BY THE SPEECH ANALYTICS SOLUTION

• Call Topics
• Member Sentiment (this gives additional understanding if Negative experience in driving high talk-time)

(X) METADATA FACTORS

• Call Volume
• Hold Count
• Hold Time

(Y) METADATA RESPONSE

• Non-Talk (Absence of speech)

IDENTIFYING THE VITAL FEW FACTORS (OPTIONAL)

This step is purely “optional”. However, it certainly lends itself in selecting the factors that have strong relationship with the response which helps to create more targeted and robust experiments. Therefore, a basic correlation coefficient can be derived by performing the correlation analysis.

So, based on the above thought I performed the correlation analysis and came up with a set of strongly related factors for the experiment.

ANSWERING THE BIG QUESTION

In order for me to answer the “Big Question” which factors are driving or influencing the Non-Talk. I decided to go with the “Full Factorial” design with “2-15 factors” and was able to draw some interesting insights out of it and design the mitigation plan.

Findings:

• It was pretty evident from the Standardized Effects plot that “Topics” that represents the “vital few” call types had the maximum effects on the Non-Talk
• A combinatory effect of Topics with Call Volume and Topics with the Call Duration was also observed
• This experiment also highlighted that Factors ABC and see together has no significance

BENEFITS OF DOE

• These experiments can identify the key influencing factors to control a process and improve it
• DOE yields a lot of information at a low cost and effort
• DOE can help the process reach Robust performance
• These experiments are effortless and precise for studying multifactor effects
• DOE helps in identifying the influencing factors to maximize performance and reduce defects
• DOE gives more information compared to one at a time experiments
• Basic DOE can be conducted without having significant statistical knowledge and understanding

CONCLUSION

DOE results can be more telling and effective if used with the Speech Analytics related factors. These experiments will provide 360 degree insight into a business problem by not only focusing on the data but also considering the voice of the customer as an important factor that can help companies to design customer oriented products and services. Today there are various statistical software available to perform DOE which can help any organization to save millions of dollar by avoiding factors that can drive existing and prospective customers away from their products or services.