Predictive operational data analytics and NLP

AI and Machine Learning solutions are springing up everywhere, and becoming more and more powerful.

Indeed, there is so much software out that that companies are finding it difficult to know which does what, and how they may fit together. I am writing this article to try to address this problem and I hope it will help at least some people (including me) to think more clearly about the subject.

Here is my current thinking: there are exactly and precisely three things that are worth studying in the world of business: numbers, words, and emotions. That’s it. There is nothing else worth understanding; nothing else that justifies any investments.

And guess what? Predictive Operational Data Analytics combined with the high-value signals that come from NLP provide amazing insights on all three.

By ‘amazing’, I mean they allow us to predict future customer behaviors, and thereby identify both growth-related projects, and possible cost reductions that can be made without negatively impacting customers and customer-related financials.

So what is Predictive Operational Data Analytics about?

Predictive Operational Data Analytics analyzes the thousands of data points that companies typically have in their IT systems for each customer and which represent the operational interactions with each and every one.

The data is available for 100% of customers, on a continuous basis. The analysis provides accurate predictions of things like contract renewal rates, cross-selling, and up-selling. It is safe to think of it as extremely exotic multiple regression analysis.

In standard use cases, there is no analysis of any words contained in the IT systems, and expressions of emotion are absent except as expressed numerically through survey response data. The output may be expressed as words, as Generative AI is now emerging as a way of translating thousands of data points into words that are much easier for humans to understand.

What is Natural Language Processing?

Natural Language Processing has been around for some years. At its best, it is a form of text analysis that uses AI to recognize both the themes that are contained in a batch of words, and the sentiment with which those themes are expressed.

A fundamental concept of the best NLP software is that concepts that use different words but mean the same thing are automatically grouped together. Here is one way to think about text analytics. Themes and theme groups are the key characteristics of NLP.

The great advantage of using themes for your analysis is that there are far fewer of them than there would be if you were just to use words. Phrases that include words that relate to switching providers will all be grouped together into a ‘switching providers’ theme, for example. If you have a lot of survey results, there may be a lot of different themes and they may vary over time, even for the same customers.

How can they complement each other?

Beyond my simple view that one handles numbers, while the other handles words and emotions, I believe there is more complementarity, in at least two forms.

First, each major theme you find in your survey data can be assigned a number. These numbers can be included in the training data you use to build your AI model. The frequency and trend you see for your ‘switching provider’ theme (for example) may have good predictive value for customer retention.

You may therefore be able to use that information to validate the predictions that come from operational data. The only downside here is that there are no themes in your companies day-to-day operational data, as the themes come from surveys in this case.

This reaches an extreme in the case of contact centers where an integrated partnership covering 100% of customers should be possible. Contact center text (whether from chat sessions, voice call transcripts, email exchanges, or from surveys that include open questions) is of immense value in identifying customers who are likely to become Detractors, and of course to define improvement projects that address issues that customers express in words and emotions.

Combining this text with operational data analytics for the same customers will produce continuous insightful predictions of customer defections, with related, prioritized causes.

Pure analysis of the numbers is simply not enough for contact centers in consumer businesses in particular. The difference with B2C is of course that the consumers who reach out to a consumer contact center are the decision-makers. This is rarely the case in B2B. NLP remains highly useful in B2B, both to analyze overall brand-level survey results, and to better understand contact centers, whether for pre-sales or remedial uses.

Conclusion

This whole area is in its infancy, relatively speaking. I would love to be able to provide you some examples of customers that are already using both technologies well together, but I simply don’t know of any. (Using data analytics software and dumping your survey responses into ChatGPT does not count, as the results are simply not all that useful.)

Of course the teams at both Thematic and OCX Cognition would love to hear from you if you want to be at the leading edge.