Emotional analysis is an extremely useful yet controversial AI tool. It is sometimes controversial due to the privacy infringement that happens as far as the customers are concerned. At the same time, it is especially beneficial in areas like retail where real time engagement matters, helping to improve customer experience, sales, and profit. In an attempt to stay in the ethical path and simultaneously work in a way that benefits business, here we are combining AI techniques to analyse non-confidential information received from customers.
This article aims to explain how this data can be used to improve products, understand pain points and increase sales.
A large retail chain ABC has its customer base having a large social media presence and the customers also connect to the retail call centre for reporting issues. So for retail chain ABC, the customer feedback comes through social media posts appearing on the internet, reviews that are posted on company website or other sites, and through call centre tickets where customers connect to resolve the problems they face.
ABC retail chain is a leading player in retail market, but they work on continuous innovation and they are looking at more ways of keeping their customers happy, providing better products and solutions, improving sales and thus improving their business.
ABC wants to combine AI techniques to address their problem through a holistic approach. While doing this, ABC is planning to initially keep out the more controversial techniques such as facial analysis but non-confidential data can be monitored using all possible ways to arrive at an optimal solution. Multi-dimensional emotional analysis can be done using the below:
- Voice analysis- by analysing the voice of customers who call to report issues to create useful data.
- Text analysis using Natural Language Processing- by scanning across social media posts, reviews, and tickets to extract information and understand customer sentiments.
Introduction of Solution(s)
Introducing voice analysis to analyse calls and videos:
Voice analysis in a call center is the process of using voice recognition and using AI services to analyse recorded voice calls, translate speech to text, and analyse the conversations. We are looking at indications for what makes the customer happy, excited, anxious, angry and looking at how different triggers result in a whole range of emotions. The voice calls will also be studied to see how the customer starts a call, progresses through it and ends it to understand the levels of satisfaction a customer is arriving at for the service provided. The YouTube videos and similar audio channels can also be analysed like this.
Introducing Natural Language Processing to analyse written text:
NLP or Natural Language Processing uses algorithms and computational methods to analyse the written text. This can be used to extract meaningful information from textual data found in social media, surveys, blogs, news stories, reviews, and other sources and use them to predict user sentiment. It gives you a better insight of what real customers think about your company. Sentiment analysis of social media posts, customer reviews and call centre tickets all can be done since these are all public or in the case of call centre tickets- recorded information which can be used by the company and hence cannot be considered as strictly confidential.
Application of Solution(s)
For both voice and text we start with pre-processing the data. When it comes to voice this means removing any disturbances such as noise. The given data is made as clean as possible. For text, we remove incorrect and meaningless words, junk characters and other useless information. In case of text sometimes we need to split given phrase text, remove whitespace etc to get meaningful words.
Once clean-up and pre-processing is done for both voice and textual data, features are extracted and this is what is used to train a model. In Machine Learning we can train a model using various techniques such as Neural Networks, logistic regression, support vector machines (SVMs) etc.
Once the model is trained this is then used to predict the sentiment of text to be positive, negative or neutral, we can identify the more frequently used words, what is trending etc. Similarly voice can help us emotional content, frequently referred products or problems etc.
The final results are then used to address customer issues, increase the attention in a specific area- such as an in-demand product or a pain area which is often referred to.
Application of results from the voice and text analysis using AI can be manifold as follows:
- Focus on add-on sales for the product most in demand. This can be done with the help of call centres by referring to them over calls, rolling the product over a wider range such as a different geographical area or a different customer segment.
- Focus on advertisement of lesser know products. Invest in some good advertising to get the lesser known products to the hands of more people.
- Address the problem areas- Immediately identify and work on solving frequently reported issues.
- Get innovative ideas- sometimes customer base can be the best place where a company can find innovative ideas. We can try to understand what customers are looking for and bring in that innovation.
Our long term focus is at improving the analysis tools by adding historic data to analysis and improve precisión. For instance while analysing the voice of a customer, various insights can be derived from emotional analytics but even better precisión can be obtained if we know the customers inherent ways of speaking. A person who is generally jovial will respond to a situation differently from a laconic one. We are looking at cross checking results of analytics against the individual’s historic records.
Another aspect that we will soon focus on is converting real time data from retail store cameras to useful figures. Keeping facial analysis still out of picture, we can still use store data if the camera captures can be converted to data. We will be looking at details on traffic in each alley, how much time a particular customers spend in each alley, detecting signs of taking as well as keeping back the products and using all this data to conclude what is most popular, what can be the next big thing, what needs to be looked into if customers are constantly doubtful whether they should pick or leave back some particular products etc.
We can conclude that there are several ways by which the multi-dimensional emotional analysis can help ABC and the company can continue working on every aspect of it. We also have a long term plan in which we are planning on working on latest and more innovative ways of incorporating Artificial Intelligence for helping ABC and its customers. Overall, voice and text analysis using various AI techniques is a good start and ABC will focus on giving a better customer experience as well as improving their business.
This article was submitted as part of the SogetiLabs India Hackathon’s blog and whitepaper contest.
Author – Archana Ambali
As a testing manager within the ORO programme for a multinational postal service company, I have been working with Capgemini for over a year now. I have extensive experience across domains, primarily BFSI, and hold a degree in Mechanical Engineering and a PGP in Data Science and Business Analytics. As a certified Agilist and data enthusiast, I bring a unique perspective to my work in testing. In addition to my professional pursuits, I am also passionate about literature, travel, and food.