Metrics are critical for understanding how your customers are doing. Most of them are measured by analyzing the direct feedback from customers through different feedback channels and modelling them. For instance, the Net Promoter Score (NPS) and Customer Effort Score (CES) are direct feedback metrics collected through feedback popups or forms, and calculated based on customer feedback for a product or service they have consumed. Customer Lifetime Value (CLV), another popular metric, is calculated based on the previous purchasing history of a customer that projects the total revenue that an individual can bring to the business.
Sample feedback forms
However, the lack of response or indifference towards rating on these feedback forms means that businesses cannot depend only on the feedback mechanisms to model accurate metrics. In this article, we discuss the limitations of the feedback-based system, and how to overcome them with conversational analytics techniques.
Limitations with Feedback based Metrics:
Most of the feedback-based KPIs are measured based on the structured data collection from the customer directly or indirectly. Conversion, a term generally used in this context, is the process of having the desired action or outcome taken by a customer on the site or product of a business. Due to the increased conversion rate and adoption of automation, most of the direct feedback collection methods are system-driven. Automated popup chats or calls are directly triggered by the system once the conversion has happened. But there are numerous challenges associated with it:
- Unresponsiveness: Even if you have an efficient feedback collection mechanism, data shows that the response rate will be between 5-25%, which means around 3/4th of customers are not responding to the survey. An organization cannot accurately measure the metric due to this trend. For example, happy customers often disregard the feedback request since they are pleased with the services and do not feel the necessity to record feedback. Whereas, unhappy customers are eager to voice their concerns. This leads to data imbalance and bias towards negative customer feedback.
- Illegitimate Feedback: A considerable amount of feedback is not legitimate for different reasons like unfair business competition, automated bot feedback, system-triggered conversion, etc. It creates additional challenges for businesses to filter such data and makes it almost impossible to cleanse it.
- Real-time Survey: Often, enterprise systems are not capable of triggering real-time feedback after the purchase, and it leads to a poor response rate.
What is the alternative?
How could a business understand the customer experience without directly asking for customer feedback?
By reading the customer’s mind. Using Artificial Intelligence (AI).
This is possible with the right set of customer interaction data and artificial intelligence models. When a customer tries to associate with a business, he will find ways to get more details about it through different channels like chat, mail, or call. He may check their official site, log in to their platform, send a direct inquiry mail, or even connect with a customer care executive for inquiries. Enterprise systems are mature enough to collect these data in different stages of the customer journey, right from the initial conversation to the end of conversion. There is enough data about customer behavior, including how they interact with your business, the effort spent to make a purchase, and the intensity of their emotions after purchase.
Here is how we can make this work with AI.
Conversational analytics and its Importance in Customer experience
Conversational analytics, a broader term in AI, is the practice of extracting meaningful full patterns and information from human interaction data using different Natural Language Processing Algorithms. In business, customer interaction happens in multiple ways like inquiry calls, chats, emails, platform-specific actions, etc. This unstructured data is captured at different stages of the customer journey through a variety of channels and is recorded in several formats in multiple applications.
There are many opportunities for enterprises to formulate customer metrics with this data if they can consolidate the data points per customer per conversion basis, which will open up the possibilities of conversational analytics in Customer Experience.
The three key pillars that businesses should consider when designing a conversation analytics platform for CX are:
1. Behavior Analytics
Behavior analytics is comparatively a new field in analytics that collects various user actions on a web or mobile application. It then analyses how a customer interacts with a digital platform. When a user visits an application, he performs a set of actions to fulfil his needs. For instance, for an e-commerce platform, the Customer logs in, enters the required item in the search pane, and adds additional parameters to filter and sort the results to find the exact product he wants. Once satisfied, he may proceed to checkout, which includes another set of actions. Throughout this journey, he stops by a few different sections of the website, calls customer service to have his questions answered, may get stuck somewhere, or may even decides to cancel the login/ subscription in the middle.
Imagine if a system captured all the events in a customer journey, including most elements interacted with, features ignored, pages revisited, time taken?
This is feasible using tools like Google Analytics, Mixpanel, Amplitude Analytics, and HotJar.
How can we derive customer experience from behavior analytics data?
Behavior analytics typically works by combining the events captured with user segmentations. User segmentation, a key idea in behavior analytics, is the practice of categorizing users into groups according to factors such as their country of origin, age, lifestyle, gender, education, and other applicable factors that impacts the business. This data is then used to derive various product metrics and provide a 360-degree customer journey. We can further extend the process with a machine learning model and historical data since our goal is to uncover customer metrics. We have 20% of legitimate direct feedback and the captured behavior data. Together, they could be a suitable catalyst to unearth the remaining input. The typical flow would look like the below:
The machine learning model trained with product metrics and historical feedback would then be utilized to extract the expected customer metrics based on the business needs.
How can we implement this?
Let us consider a scenario where 100 users visited an e-commerce platform to buy a product P1. Out of these 100 users, 20 of them have responded to the feedback survey with a range of ratings. If we have recorded all of these users’ behavior for the purchase journey along with the user segmentation, with these behavior events and available feedback data in hand, we could easily model a Machine learning algorithm to derive the other 80 user ratings. Users having the same user segmentation and behavior data might have similar ratings. Over time, by finding a different set of data points across all the user segmentation and purchase Journey, the model would be capable of revealing a wide range of patterns across segments and products. Another added advantage of the approach is anomaly detection. It will help us to find out illegitimate feedback and help the business to filter out those data points in future while formulating the metrics.
2. Call Center Analytics
Call center analytics is another crucial method in deriving customer experience metrics from conversational analytics. During the conversion journey, customers engage with the business in a variety of methods – a direct call or chat with a customer service representative, an email to the service or marketing team, or an interaction with a chatbot in the enterprise system. This conversation history has already been recorded in the enterprise system, which can then be utilized to unearth valuable insights about your customers.
Deriving customer experience metric from call center data
Customers express their emotions, sentiment, or even feedback on your product or service when a real-time one-to-one discussion happens with a customer care executive. They will often express their option, satisfaction, and disagreement with different levels of emotion. These data points along with call center time metrics (metrics like response and hold times, abandoned calls, resolution time, and call transfer rate) can be modelled to derive the customer experience using state-of-the-art Natural language processing and other AI modelling paradigms. The model should be trained in such a way that it can extract the sentiment, emotion, and feedback along with its intensity level. The diagram below shows the high-level architecture of Call center :
User segmentation is crucial for call center analytics as well because, each type of user responds differently, and it is crucial to model the algorithm based on the user segments. Once the model matures enough to predict the right set of intent, map the intensity with a weightage metric. Thus, it would be easier to convert the intensity of customer emotions into a measured entity, from which it is easy to derive the customer experience metric.
Speech recognition is one of the crucial tasks in Customer care Analytics and its accuracy is highly dependent on factors like noise, incorrect word choice, regional language and so on. Currently, sophisticated systems are available for speech-to-text conversion but there are gaps in this area for regional language processing and highly noisy data.
Consider the below sample Interaction Metrics between an agent and a customer. We could derive a set of customer experience patterns from this interaction metric itself. This data will fuel the text analytics platform further to derive the intended Customer experience metrics.
3. Social Media Analytics
Social media influences all facets of life, and businesses position it as a key platform for marketing. Customer decisions are greatly influenced by these platforms, and they express their feedback and suggestions as posts, comments, tweets, or reviews. Using these social media customer voices for text analytics reveals interesting customer experience data points. The underlying AI modelling techniques for social media analytics are more or less similar to call center analytics; the difference is in the data collection part. The diagram below illustrates a typical high-level architecture for formulating customer experience metrics with social media analytics.
Given that the majority of social media data is accessible to the public, we have the freedom to conduct a competitor analysis using this information. It is an added advantage of social media analytics over other methods, which also aids in strengthening the metrics related to competitor analysis data. Data collection would be the key area in social media analytics, and tools like HubSpot, Sprout Social, and Falcon.io, are readily available in the market to serve this purpose.
Bringing it all together
As Digital platforms evolve, the method we choose purely depends on how a customer interacts with the business. Deriving Customer experience metrics from non-feedback channels is not limited to the above three methods. There could be different methods in the market based on how the customer-business interaction evolves. Organizations should take a combined or stand-alone approach based on their CX touch points and the technology stack in use. In the future, as artificial intelligence develops quickly, more and more customer experience metrics will be derived without direct customer feedback.
Author Bio:
Sreekanth Narayanan is a software architect with 12 years of significant expertise in the IT sector and deep knowledge of both full-stack app development and AI&ML solutions. Over the past few years, he has become increasingly fascinated by the fields of AI and Machine Learning. Currently he is offering solutions to various industrial challenges in computer vision use cases, natural language understanding, and conversational AI.