scorecardresearch
Voice of the customer: Changing possibilities with AI

Voice of the customer: Changing possibilities with AI

Popularly known as the ‘Voice of Customer’, VOC is an umbrella term that is used to signify the technology infrastructure, analytics and AI capabilities that help businesses realise this very important goal.

‘Customer centricity’ and ‘customer obsession’ are two fundamental principles that almost any enterprise has to live by in order to be successful. ‘Customer centricity’ and ‘customer obsession’ are two fundamental principles that almost any enterprise has to live by in order to be successful.

‘Customer centricity’ and ‘customer obsession’ are two fundamental principles that almost any enterprise has to live by in order to be successful in today’s increasingly competitive market. A stepping stone towards this journey of customer centricity is the ability to ‘listen’ to what customers are saying both explicitly and implicitly about the enterprise and its products and solutions. Popularly known as the ‘Voice of Customer’, VOC is an umbrella term that is used to signify the technology infrastructure, analytics and AI capabilities that help businesses realise this very important goal.

Listening to customers and analysing their feedback is not a new concept. Organisations have been doing this for multiple decades in various ways. In the past, some of the most popular ways of capturing feedback from customers were through primary research activities such as targeted surveys and focus group discussions. However, these are expensive, not real-time, and prone to sampling bias. In general, these methods have proved to be difficult to scale in today’s digital world where there are multiple new-age channels of interaction and touchpoints. Technological advances over the past few decades and the advent of artificial intelligence (AI) specifically have brought new possibilities to the table.

In today’s digital world, customers interact with enterprises and their products through various channels and mediums. These can be broadly classified as internal and external channels.

•    Internal channels are generally offered and controlled by enterprises as a direct line of communication between the customer and the enterprise. Some of the most used internal channels of customer interactions are incoming phone calls, virtual assistant chats, live agent chats, and e-mails. Apart from these, customers also provide explicit qualitative and quantitative feedback through mediums such as surveys. The Net Promoter Score (NPS) survey is one such widely used tool in the industry for capturing customer feedback.
•    External channels are outside of the direct control of the enterprise and therefore typically open to everyone in the ecosystem. Social media platforms such as Twitter, Facebook, Instagram, and Reddit are good examples here.

Technological advances powered by AI have made it possible to log and store all these types of customer interactions for further analysis. Some of the key technological advancements which have enabled this include real-time voice-to-text transcription and the creation of enterprise data lakes for storing both structured and unstructured data.

Voice of Customer (AI Workflow)

A typical new-age AI workflow that helps translate vast amounts of interaction data into actionable insights is shown below:


Data Capture and Ingestion

All customer interactions both in internal and external channels are logged and stored. This results in vast amounts of unstructured data being in the form of text, audio, video, and the like. All this data is stored in usable formats through multiple stages of Extract-Transform-Load (ETL) processes. This data is typically retained in an enterprise data lake (EDL), which is a central repository for all structured and unstructured data within an enterprise. 

Data Pre-processing

Data in its raw form (text/ audio/ video, etc.) cannot be consumed by downstream AI algorithms.  Multiple pre-processing and featurization steps are required to make the data consumable and useful to AI algorithms,. Some examples here are tokenisation, phonetic hashing, lemmatization, and others. The emergence of deep neural networks has brought in a host of embedding techniques that are often used in this space.

Data Augmentation

Various natural language processing (NLP), speech recognition and image-processing algorithms are used to derive useful metadata from raw data. These metadata elements form a useful feature set for downstream AI algorithms. Some examples include named entity recognition, part-of-speech tagging, tonality tagging, and the like. 

Task-Specific Algorithms

A suite of task-specific AI algorithms can be used to derive actionable intelligence from data. The scope here is often decided based on a combination of business priorities, organizational operating models, and the maturity and ingenuity of AI and data science teams. Some of the key task-specific algorithms can be seen in areas such as theme detection, text summarization, sentiment analysis, intent determination, and emotion classification.

Consolidation and Normalisation

Consolidation and normalization of AI model output across multiple channels of interaction is a vital aspect of any successful VoC program. The idea is to break silos that exist amongst multiple channels and bring up a true omni-channel view of customer pain points, needs and asks. 

Analytics and Insights

Analytics and insights teams provide the last-mile connectivity between the AI described above and the business  or product teams who manage the final customer experience. 

This is therefore a critical piece that determines the success or failure of the entire program. 
The role of the analytics and Insights team is to use the output generated by the AI models and present the most insightful and actionable pieces of intelligence to the decision makers in the business. The tools and techniques used in this space range from simple slicing and dicing of data on spreadsheets to the creation of sophisticated data visualizations and dashboards.

Finally, it might be already evident that the workflow described above needs to be supported with a very strong and matured technology platform and infrastructure setup. This includes a robust data strategy in the form of an enterprise data lake, an analytics platform strategy in the form of right databases, messaging platforms, visualization tools, and more, and an AI platform strategy in the form of AI infrastructure and toolsets.

It must be evident by now that a successful VoC program requires significant technology and manpower investments. However, above all, it requires an unwavering commitment to the values of ‘customer centricity’ and ‘doing right by the customer’. The technology landscape around VoC has seen massive changes in the past decade and is bound to change even further in the days to come. However, one aspect of any successful business that will probably never change is the importance of ‘listening’ to the customer in a meaningful and actionable manner.

Views are personal. The author is a Director of Data Science at Fidelity Investments India.