Use text analytics to get deeper insights into employees' minds

It is a given that organisations want to know and understand what their employees are thinking and feeling. Or at least they ought to. Most organisations depend on yearly surveys for gaining insights into the overall mood of the workforce. While this annual exercise brings together both ratings data and textual data, the former usually on a 5-point Likert scale and the latter as comments or suggestions from the employees, organisations typically tend to focus on only the ratings. This is perhaps understandable because it is easier to comprehend and compare ratings against say, the previous year's data. Textual feedback, for the most part, is left to the individual managers to interpret. Given the paucity of time that managers have, they typically tend to glance at such feedback, cavalierly dismissing that which doesn't confirm their opinion of themselves and giving way too much credence to that which does.

However, this need not be the case. In this day where most data is in a digital format, much more can be done with the textual data to enhance an understanding of what the employees are telling an organisation. The field of text analytics comprising of such esoteric disciplines as Natural Language Processing (NLP) has seen a boom over the last ten years with advanced algorithms able to make meaningful insight from all the clutter. The culmination of this is obviously in what is now being called cognitive computing that involves self-learning and adaptive systems that utilise machine learning and pattern identification to mimic human behaviour. IBM's Watson analytics, for example, has a cognitive computing web service that enables users to identify the characteristics of a writer from textual input fed to it. The model for this has been trained on millions of social media texts and while this may not be completely applicable to an organisational setting (without having to rebuild the model), it does provide us with a wonderful example of the way this whole area is moving. While readers might scoff at using such systems currently in their organisations, there is much that can be done even without resorting to cognitive models for textual data.

One of the easiest ways to make sense of textual data is to visualise it in the form of simple wordclouds. A wordcloud is a collection of words organised in a random manner with the size of the word indicating the frequency of its use and the colour of the word indicating the positive (green), negative (orange/red) or neutral (grey) tenor of the words as shown in the figure given below.

Such a visualisation can help managers to immediately understand the dominant topics that people are thinking about such as feedback, projects, exposure and outlook in the example shown. The colour coding also shows that people are worried about their pay but do appreciate the opportunities for working and would recommend this company to others.

While this gives a sense of what people are talking about, it would still be great to encapsulate it as a number and to follow its progression on a regular basis. Enter the sentiment score. Methods now exist using machine learning approaches to extract the dominant sentiments being expressed in the text. Aggregating this at team, functional, divisional or even organisational levels can provide additional, tangible and actionable insights into the overall mood of the organisation.

In addition to this, techniques and algorithms can also extract dominant themes around which conversations are happening within the organisation. While the wordcloud gives an overall sense of the words being used, themes can be extracted using more advanced algorithms. With this arsenal of insights, managers and employees can have frank conversations about how to move the needle on organisational culture. And that, as the Bard says in Hamlet, is a consummation devoutly to be wished for.

The writer is founder & CEO of HR analytics start-up nFactorial Analytical Sciences