Huawei SmartCare and NPS: A Powerful Synergy

 Introduction
Using results from a recent project, this paper demonstrates how Huawei’s SmartCare CEM solution can drive your NPS (Net Promoter Score) improvement program. The key factors which influence NPS can be identified, prioritised and incorporated into a model which accurately predicts NPS results. The insight provided will allow your organisation to focus on those changes which will most positively impact NPS for your entire user-base, including the silent majority who never respond to surveys.

NPS: A key industry metric
NPS is increasingly being adopted by service providers to evaluate their customer-facing performance. Based upon survey responses to the question “On a scale of 0 to 10, how likely are you to recommend us to a friend?” NPS is calculated by subtracting the percentage of detractors (those responding with 0-6) from the percentage of promoters (those responding with 9-10). This is shown diagrammatically in figure 1.

The resulting score, which can range from -100 to +100, has been shown to correlate well with business results, leading many service providers to align a part of staff bonuses with NPS results. This motivates the organization to seek to improve NPS, and with scores across the telco industry typically ranging from -50 to +40, there is often large scope for improvement. However, there are two big challenges when seeking to improve NPS: Where to begin, and what about the silent majority of your users who never respond to surveys?

Generating Actionable Insights for NPS
In addition to supplying the service provider with PSPU measurements of every end-user, a key benefit of Huawei’s SmartCare CEM solution is that it also enables corresponding subjective feedback to be sampled, allowing each user’s true perceived experience to be determined. This allows operational decision s to be driven by the true end-user experience, rather than internal, inside-out facing KPIs. SmartCare can flexibly accept many types of subjective feedback. For example Huawei’s M2 solution allows subjective feedback to be easily and conveniently collected via the user’s mobile device. Crucially, NPS survey results can also be used; by comparing a users’ NPS survey responses with the associated PSPU measurements it is possible to determine and isolate the key factors which influence NPS, which in turn gives the service provider the insight they need to improve NPS.

Modeling NPS in Practice
Huawei has recently adopted this approach to improve the NPS result of a tier 1 service provider by focusing on network performance aspects.

Many non-network aspects can impact NPS, such as brand perception, retail experience and customer care. For this project, to isolate the contribution of the network performance, two alternative questions were asked in the NPS survey: “How likely are you to recommend us to your friends based upon your experience of using our network for voice/data?” Of course, the responses would still be partially influenced by external factors, but these were accounted for as error terms in the subsequent analysis.

For a large sample of users, the responses to the “network” NPS questions were compared with their recent network service performance based upon PSPU measurements. Using a combination of regression and decision-tree based analysis, together with Bayesian network algorithms, a model relating network service performance to network NPS response was created. The resulting model allows NPS scores to be predicted based upon the user experience received. Figure 2 shows this approach diagrammatically.
 

Referring to figure 2, the top row sequence shows th at a user’s NPS response is a function of their rational and emotional reaction to the actual experience they received. This is modeled on the second row: PSPU measurements allow the objective experience to be accurately determined. This data is then fed into the NPS prediction model to determine the likely NPS score for each user, and in particular to determine whether they are likely to be a Promoter or a Detractor.

Encouraging Results and Meaningful Insights
Once in place, the model was used to predict the likely NPS response of every end-user, including the silent majority who would never normally respond to NPS surveys. The results achieved were very encouraging. Tables 1 and 2 below show a comparison between the actual NPS score of newly sampled users with the predicted score of those same users. Accuracy rates of 76% and 78% were achieved for predicting Promoters and Detractors respectively.  

 
A key point from Table 2 is that 76% of detractors can be identified. NPS related research shows that detractors are very damaging to a business, so the ability to readily identify them is vital. With this new insight, the service provider can take a business decision as to how to handle them.

Predicting the NPS response is just one aspect of the SmartCare NPS model. More importantly, it can also show the key factors and thresholds which drive the responses, allowing remedial action to be taken. Unsurprisingly, the results from this project showed that NPS is impacted by both the network performance and what the user was trying to do at the time. For example, disconnections while downloading large web-pages were seen to be more impactful on NPS than interruptions of small web-pages. Furthermore, the model has revealed important thresholds for KQIs, as shown by the examples in table 3. A value o f Web_Page_Browsing_Delay greater than 12000ms tends to result in detractors, while a value of 7040ms or lower is required to maintain promoters.
 

Using this insight, targeted network optimization was conducted, specifically focusing on those KQIs which would lead to an increased number of promoters and/or decreased number of detractors. Following a 12-month program, network NPS was significantly improved, as shown in figure 3 below.
 

More importantly, the improvement in network-focused NPS also resulted in a 10 point improvement in overall NPS, as shown in Figure 4:
 

Building on Success
Huawei SmartCare can already predict network NPS for all end-users, as well as identify the key influencing factors. The next step, working with our ecosystem partners to measure the user’s end-to-end lifecycle experience, is to develop a model which predicts overall relationship NPS. This will offer service providers unprecedented insight, removing the guesswork from NPS improvement programs.

Author:  Jonathan Hopkinson