Telecommunications

Predicting Customer Relationship using Transactional Behaviour

Step 1: Introduction to the Problem Statement

{The conversation starts with our client who is a telecommunication B2B industry introducing the problem that was being faced regarding the use of TNPS}

The customer experience program of our client is designed to have two levels or types of surveys, on level one, transactional surveys are sent to the customer, in response to an activity that the customer has completed with our client, it may be ordering a service, resolving a query, etc. across four areas of service. On level two, a relationship survey is sent to the customer to measure the overall relationship the customer has with the brand. In the earlier program, at both levels NPS was used as the CX measure, the transactional NPS and the relationship NPS.
It does not make sense to ask the customer to recommend our brand on the basis of one transaction, hence asking NPS in a survey that is sent to the customer, in response of a single activity done with us, is immaterial.
There were a lot of hurdles in introducing a new Business measure in transactional surveys, like, convincing the board, finding the appropriate new measure, how will the new measure be related to the Relationship NPS and switching a new measure would mean discarding the historic transactional NPS data and much more.

Step 2: Why we chose NES?

Earlier, the customer experience survey of our client consisted of asking all the 3 metrics (NPS, NES and CSAT) for the 4 business areas SALES/Delivery/Netops/CRM. The transactional CX program needed one prime metric which would be tracked across all business areas. At the time, three metrics were being asked in transactional surveys, NPS, customer effort, customer satisfaction and the survey was lengthy for the customers. So, to simplify the survey we needed to select that one prime metric and find out which of the 3 correlates better with all the drivers being asked after the 3 metrics in customer experience survey.

The Model:

As a first step, we applied machine learning algorithm on the 2 year survey data. We constructed 3 models for each of the 3 metrics as dependent variables and the drivers as independent variables to check which model gives the better results.
We setup linear regression analysis models to establish a relationship between the drivers and the metrics for each of 3 metrics and to study which metric explaining the model better.

The Result :

After running 3 regressions, we looked at the results and found that the metric which explain the regression analysis better amongst the others was the customer effort score, which is very close to NES (R square was more when it was the dependent variable).
Also, we looked at number of significant drivers and found that again regression ran on NES has more significant drivers when compared to other models.

Conclusion :

The above steps were repeated for all 4 business areas and based on our results for each business area, NES was the leading metric which shows better results. Hence, we concluded that NES should only be incorporated in our transactional customer experience surveys for all business areas to improve the survey engagement.

Theoretically :

It makes more sense, at a transactional level, to ask the customer about the ease with which they were able to complete the particular transaction rather than asking if they would recommend the brand. Since people are most likely to recommend a brand when they are loyal to it and such a relationship develops over time and over multiple transactions, which can only be measured in a relationship survey.

‍

Step 3: Relationship between NES and NPS

{In this section, we will talk about the relationship between NES and NPS}

The Net easy score is a business metric that allows us to measure the ease with which the customer was able to complete the transaction with us.
At the transactional level, it makes more sense to ask about the ease of transaction rather than how loyal the customer is to us, as loyalty is something that is based on the firm’s relationship with the customer overtime and with multiple transactions. Therefore, we set forward this approach to measure the NES at every transaction and NPS while analyzing the overall relationship the firm had with the customer.
The challenge that now remained was to devise a model that could establish a relationship and use the present NES to provide guidance about the future NPS. This would give us an opportunity to predict the relationship NPS using current NES and take measure to ensure good relationship with the customer.

Step 4: Predictive model

01 Fortunately, all of the historic transactional data did not go obsolete as, along with TNPS there was also a question in the old survey related to ease of transaction which we could effectively use. Therefore, with roughly 13 quarters of our client’s historic data in the bag, we started framing the basic framework of the study. Fortunately, all of the historic transactional data did not go obsolete as, along with TNPS there was also a question in the old survey related to ease of transaction which we could effectively use. Therefore, with roughly 13 quarters of our client’s historic data in the bag, we started framing the basic framework of the study. Fortunately, all of the historic transactional data did not go obsolete as, along with TNPS there was also a question in the old survey related to ease of transaction which we could effectively use. Therefore, with roughly 13 quarters of our client’s historic data in the bag, we started framing the basic framework of the study. Fortunately, all of the historic transactional data did not go obsolete as, along with TNPS there was also a question in the old survey related to ease of transaction which we could effectively use. Therefore, with roughly 13 quarters of our client’s historic data in the bag, we started framing the basic framework of the study.

‍

02 At its core, this analysis is the answer to two simple questions,

How much does transactional behavior impacts the relationship of customer with the brand?
How long does transactional behavior take to impact the relationship of customer with the brand?

These questions led us to the requisite model that helped predict the forthcoming Net Promoter Score.

03 Since transactional and relationship are two separate surveys, in the initial phase of approach setting the main goal was to find a medium that could become the bridge between NES and the NPS surveys. By design, the relationship surveys were framed in such a way that after the NPS question, respondents were asked a set of drivers. Each of these drivers corresponded to each service area for which transactional surveys are issued. For example, if delivery is a service area with NES surveys, there was an attribute present in the NPS survey that asked respondents to rate us on the satisfaction in regards to their delivery experience.

‍

04 After identifying the attributes corresponding to different survey areas, it was time to find out the impact each of these attributes had on the Net Promoter Score. In other words, we had to find out how much do transactional activities in separate service areas affect NPS. For calculating such an impact, we used the regression analysis with the 4 drivers as predictors of NPS. This gave us the answer to our 1stquestion and we found out that each service area has a different strength in influencing NPS. For instance, it was found that the area that manages customer queries and requests has the highest impact in molding the NPS and the sales experience had the least impact.

‍

05 Now we started looking towards the second question where we needed to determine the time period. To achieve this, we started analyzing trends with data that goes back to about 13 quarters. Since NPS is calculated in a quarterly manner, the NES was also looked at quarterly (instead of monthly) so that the two trendlines synch. Trends of quarterly NES were compared to the trends of their respective service area’s attribute in order to spot patterns between the two. We used cross correlation technique to find out, the correlation between the attribute and the NES at a gap of different quarters and the gap at which highest correlation existed was chosen as the lag between NES and NPS or the time period that the transactional behavior takes to affect the relationship measure. Just like impact the lag was also different for different service areas. For example, we found out that the service area that deals with fixing connections and resolving issues had highest correlation with its respective attribute at a lag of just 1 quarter, whereas sales had a lag of 4 quarters.

‍

06 With all the findings up to this point, we were equipped to calculate a Net Easy Score, which would be a combination of NES of all the service areas and take into consideration the lag and the strength of impacting NPS for respective areas. Overall NES = Ʃ business area NES (with lag)*Impact of mapped NPS attribute on NPS Since we had adjusted for the lag, the overall NES had the highest correlation with a 0 quarters lag between NPS and calculated NES which somewhat validates the overall NES in respect to NPS.

‍

07 After calculating an overall NES that had a 0 quarter lag with NPS, we turned to a simple linear regression model with overall NES as the predictor variable for NPS. This model formulated an equation (with the slope and intercept) that uses the overall NES value for a quarter to predict the NPS for that quarter.

‍