Current AI Landscape - Prescriptive Models in the Age of AI Assisted Analytics

Ravindran Venkidapathy, Vice President, Xavient Information SystemsUntil the turn of 20th century, there were three types of lies ‘Lies, Damned lies, and Statistics’. From Mark Twain and A Disraeli to Sir Charles Went worth Dilke and Robert Giffen, this was a colloquial saying, even among the rich and the opulent. But the dawn of 21st century seems to have reversed this saying into ‘Truth, Damn Truth and Statistics’. Thanks to big data and AI assisted analytics, the growing skepticism ‘Nothing is predictable here’ and the breath taking degree of ignorance has slowly started fading away.In my experience, this is becoming true for media/telecom clients who are bent on arresting their customer churns and building ‘true lift’ models.

Age of Sci-Fi Fantasy?
But the general Purpose AI that can think and communicate like humans, what computer scientists call as Strong AI still remains a Sci-Fifantasy. However, AI that is trained to excel in a narrow focus or a Weak AI, has turned out to be a versatile technological solution to many vexing problems.Using an elegant technique called ‘deep learning’, Weak AI is what Facebook uses to tag names in its uploaded photographs.
Enter everyday shopping, Amazon today has become Walmart on steroids and Netflix is the logical extension of our corner video shop, but the basic act of shopping is not different from what existed 1000s of years back when goods travel took years to cross continents. Now, it is just hours and simply gets faster and better, all because of AI.

Straight From AI Labs
Imagine a scenario when kids come home and ask, ‘Daddy, is it really true that when you were young, if you came home and you said something to your TV— did it really just sit there and ignore you? That's just so rude of the TV’.

Driven by this digital outlook and the fact that todays’ digital stack is being built on AI ONLY architectures, leading organizations have stepped up their investments and intensified focus on AI through dedicated
state of the art AI labs. From speech to machine vision, driving Bio Metric based Authentication and from home automation by re-imagining the future of TV to control of devices, ours is the era of thrilling things.

The AI landscape is vast, and hence it is a crucial time to roll out image, x-BOTS, biometrics, video, fraud, and machine vision based analytics to enterprise and consumer clients

Some where during Mar 2016, just about 15 months back, I still vividly remember trying out Google speech API and you won’t even believe the results I got. I was taken aback. It was like a drunken man blabbering; I spoke something and I got some weird text out which was completely out of context. It was insane. So that’s when I thought STT (Speech to Text) is a million miles away for practical use.

At present, not even 15 months later, we have come a long way, and are almost at par, almost as accurate as human decoding. This is certainly going to disrupt many areas, especially the CX industry, customer experience and customer service industry. Platforms such as AMPLIFY take customer service seriously and are especially for those who want to take their customer service experience to the next level. Built on LVCSR platform, the Large Vocabulary Continuous Speech Recognition platform, such platforms can recognize voice vocabulary as we do, as humans do and are intelligent enough to detect the intents. In fact, they can even suggest the next best action, in a live set up, while the call is in progress. They can re skill themselves, if we let it learn the new vocabulary and are capable of doing analytics on top of by applying proper weight age, quantifying an issue, a problem that the customer encounters and then suggest an appropriate action. This can help increase service levels, reduce call times significantly.

But this is just one side of the equation and we all know Cambridge Analytic a, a UK based company which precisely predicted 2016 U.S. election outcomes and where U.S. media Moghuls including CNN failed miserably in that aspect. Here we cannot afford to ignore the reasons cited by CA especially from a business perspective, because they factored in sample sizes, scale and data representation into their predictions.

However, the very models that we are building today for customer churn, roll to pay, uplift and so and so forth or any customer analytics model for the matter, do not have the right data representation from voice, neither scale nor sample size. In this age where negative feedback goes viral, most enterprises take a chance and conveniently ignore 56 percent of all customer interactions that are from voice and are factoring in just one or two percent for predicting customer behavior; all because of the high cost of conversion.

So this is where such AI platforms step in, to balance both the worlds essentially extending help to build better models, like Cambridge models as opposed to CNN models for customer analytics with its low cost and high accuracy conversion capabilities. Not just voice, it’s believed that the AI landscape is vast and hence, it is a crucial time to roll out image, x-BOTS, biometrics, video, fraud and machine vision based analytics to enterprise and consumer clients.