How the NPS Prediction Model can improve Customer Satisfaction

Net Promoter Score (NPS) is a measure of customer loyalty and serves as an indicator of customer satisfaction. It depicts how customers respond to the company’s customer service initiatives and engage with the brand as a whole. However, the issue with NPS is that the score might not always consider all the factors that contribute to a good or a bad customer experience. 

It might not accurately represent the agent’s performance during the interaction. Customers tend to give a bad score based on other factors as well ranging from network or product quality to general dissatisfaction with the brand. Often agents end up getting blamed for matters beyond their control. 

Xdroid’s solution to this issue was to develop our very own NPS Prediction Model. 

Xdroid’s NPS Prediction Model

In May 2023, we launched the NPS Prediction Model. This NPS predictive model can leverage data on customer interactions, experiences, and feedback to predict NPS, CSAT/other customer satisfaction scores. The model is trained with call and sentiment tags, it identifies the use of certain words and phrases to study how the customer is responding to interaction with the agents. Based on all the factors it predicts the scores and also classifies the customer as a promoter, detractor, or neutral. The predicted NPS is more telling of how the call went, and whether or not the interaction left the customer satisfied. 

The analysis is enhanced by studying customer’s sentiments at the beginning of the call, and towards the end of the call to check whether or not they are satisfied with the resolution offered to them. It tracks the change in the customer’s attitude (if any) during the call. For instance, if the customer goes from disgruntled (“I am very disappointed, my product is still not here!”) to satisfied with the solution to their problem (“Thank you so much for addressing my issue!”). Such a change along with agent behaviour, call quality, and other parameters is studied comprehensively and a score is predicted by the model. 

For every interaction, the model predicts if the customer will be a detractor, neutral, or promoter, and gives the Net Promoter Score (NPS) on a scale of 0 to 10. The agents will receive feedback on their performance and the managers/supervisors can access reports for the calls. The company can review what worked or didn’t work during the interaction. They can check how stressful it was for the customer to get their issue solved, and the amount of time taken and categorise it as high, medium or low effort. The company can work on and implement operational changes to improve CX and reduce customer effort. 

How does the NPS Prediction Model work? 

Xdroid’s NPS Prediction model can be used as a descriptive model. The model is ready to use and will be an asset to your customer service department. The more you use it, the more you’ll be able to benefit from it. Once calibrated based on your business needs, you will have access to scorecards and Zero-day KPIs for the model. With data regarding every aspect of the call, you can also compare the actual NPS to the predictions and discover what factors influence customer satisfaction levels the most. 

You can study the impact of agents’ behaviour on the customers, their response to factors like the ‘success’ or ‘failure’ of the call, the duration of the interaction, and how much time it would take to get the issue resolved. The Model interprets this data and reveals how satisfied the customers have been with the company’s services. 

Feeding the model with historical NPS data further transforms it into a predictive model which comes with its advantages. You can also easily customise the model with actual customer-rated calls with real NPS data for specific projects and objectives. This data could give the model accuracy a further boost. It will learn the actual sentiments of the customers, what matters to them, and how they behave. It can identify patterns to forecast future consumer behaviour. 

The predictive model becomes more robust and versatile as it ingests more real-life data over time, enabling it to handle a wider array of use cases. The model can ultimately proactively help the company make more strategic business decisions. 

Case Study: NPS Prediction Model

The implementation of Xdroid’s NPS Prediction model has already begun. Our clients have already started using it to track their customer satisfaction levels. Xdroid had the opportunity to implement the Model with one of our biggest BPO clients. Our client fully utilised the model and analysed, trained and tested it out on 80000 customer-rated calls. 

Across these calls, they measured 25 emotions at different points in the conversation to track the evolution/change of emotions from the beginning of the call to the end of the call. 

Through the calculation of the MAPE (Mean Absolute Percentage Error), the model displayed an accuracy of 93% in predicting the Net Promoter Score between 0 and 10 that would have been given by the customer.

In subsequent tests on thousands of verified customer interactions, the model also functioned as a descriptive tool to decipher the actual events during the call, maintaining its 93% accuracy in deducing the exact number. The goal extended from simply predicting the customer’s numeric rating to gaining a comprehensive understanding of the interaction. This spawned a new index that we call the Voice of the Customer Index or In-Call NPS, which extends beyond merely gauging customer opinions. It incorporates subjective criteria that affect customer perception, organized into three main pillars that cumulatively shape the entire experience.

The Pillars are:

  1. how the emotion changes during the call
  2. how the agent’s behavior impacts the success of the call 
  3. how much time it takes the customer to solve a problem

With the fine-tuned model, our client had access to knowledge about NPS for different call categories with valuable consumer insights. They could indulge in 100% full call sampling and compare it to the response bias of the customers as they fill out the customer survey. This allowed them to identify factors that have a lasting impact on the customer’s attitude towards the company/brand, what they prioritise during their interactions, and what makes a call a success/failure for them.

The model gave our client the ability to differentiate between the consumer perception towards the brand and that towards the customer service/the contact center agent. It provided clarity in figuring out what was contributing positively towards their customer experience and what wasn’t. Ultimately it enabled them to make data-driven insightful decisions regarding their customer service initiatives, and other operational business strategies. 

Benefits of Xdroid’s NPS Prediction Model

Here are several ways a predictive NPS model can be used to improve customer satisfaction:


  • Identifying Areas for Improvement: By modelling NPS scores, businesses can identify which areas have the most impact on customer satisfaction, even for the interactions without any ratings. With this knowledge, resources can be directed towards these areas for focused initiatives.
  • Preventive Action: A predictive model could potentially identify dissatisfied customers before they give low NPS. You could study the correlation between certain phrases/patterns to weaker calls with low NPS. Businesses can proactively address these issues before it becomes a problem.
  • Personalization: The predictive model can also provide insights into what different customer segments value most, allowing for more targeted and personalized service or product improvements.
  • Strategic Planning: It is important to keep track of what is working and playing to one’s strengths. By predicting future NPS scores, businesses can forecast the likely impact of different strategic decisions on customer satisfaction and plan better. 
  • Benchmarking: The model can help businesses compare their performance against industry standards or competitors in terms of NPS to drive improvements. They can further aggregate sentiment data and study its evolution during calls and Cluster calls using additional metadata provided by the customer. 
  • Customer Retention: With a predictive NPS model, companies can identify detractors and anticipate churn rates. The model will highlight the stress levels of the customers, how they reacted to certain aspects of the call, how the duration of calls affects them etc. Targeted initiatives to combat these issues can boost customer retention.
  • Resource Allocation: By understanding what influences NPS scores, businesses can allocate resources more efficiently to maximize customer satisfaction. They can also design training programs to address issues that lead to low customer satisfaction.
  • Testing Improvements: With a predictive model, businesses can model the impact of proposed changes (like a new product feature or customer service protocol) on NPS scores before implementing them.

By continually using these insights to improve, businesses can make significant strides in improving customer satisfaction, loyalty, and overall business growth. 

Do you want to be part of the NPS revolution?

You too can take on the endeavour of ensuring 100% customer satisfaction with Xdroid’s NPS Prediction Model. Collaborate with Xdroid to join the likes of industry leaders and elevate your game in the CX Industry. Request a demo today and discover the power of AI here:


Latest News