Machine learning will boost the enterprise user experience

Machine learning at the intersection of application design and predictive data analytics.
Written by Joe McKendrick, Contributing Writer

Many of the applications running today's enterprises are not built to help organizations understand or improve the user experiences of their customers. New types of intelligence -- in the form of machine learning behind the scenes -- may help put data to use in boosting the custumer experience and user experience.

Photo: IBM Media Relations

That's the word from Mike Gualtieri, analyst with Forrester, speaking at IBM's recent machine learning summit in New York. In his presentation, he called for greater fusion between UX designers and data scientists. "Most application designers are so old-fashioned, because when they design an application in an enterprise for a customer, they're looking for the interaction design; they're looking at where buttons should be positioned," he states. "They're not thinking what prediction model should be put in this user experience, or in this business process to make it more efficient."

The internet leaders -- the Googles and Amazons of the world -- have brought together application designers and data scientists to further their user experiences through predictive analytics, Gualtieri continued. "Google and the internet giants think of this all the time. Customer experience professionals can do the greatest job in the world, but they're not thinking about how to create cognitive logic within applications, and they.re not going to create intelligent applications."

The goal of today's enterprises, Gualtieri advocated, is to "make customers feel like celebrities." With up to millions of customers, the challenge is to create a special experience for each, based on "a machine learning regime that is continuously building models. But you need those design teams to implement it in the design."

Machine learning, Gualteri elaborated, is part of artificial intelligence, but differs from traditional notions of AI, he explained. "There are two types of AI: 'pure AI' mimics human beings -- the sci-fi stuff, in any movie you've ever seen. That is not what we're talking about. Pragmatic AI has elements of machine learning, which are its building-block technologies to build a modicum of intelligence into applications. This is built up from data, which is a foundational prerequisite."

Forrester's surveys find about 58% of companies are currently researching AI, while 14% have implementations in pilot or production, Gualteri added.

Machine learning is a probabilistic exercise, he explained, noting that it comprises algorithms that analyze data to find a predictive model." The model then generates the likelihood of an event, such as a customer leaving or an upsell occurring.

While data scientists are key to machine learning initiatives, there are publicly available APIs that can provide such capabilities as well. "There are increasingly some prebuilt models you can access, like Google API, and IBM's Watson Services," Gualteri said. "So application developers, without knowing what a random forest is, without knowing what a neural network is, can use those prebuilt models in their applications."

For enterprises seeking to put machine learning to work, scalability is a challenge, Gualteri said. Every business question generates its own model, "There are tons of these models per customers -- it's not just about creating one or two models," he explained. "The future is about many, many models per customer. Say you have a million customers. And you want to predict 10 characteristics, you want to create 10 behaviors, and 10 needs in real time. That's potentially 30 cognitive models per customer. If all of these models are individualized, you'll probably have a series of models -- potentially 30 million cognitive models. You could have a challenge in creating those models, running those models. It requires a lot of compute, and a lot of productivity from data scientists to maintain those models."

Ultimately, this needs to be addressed through greater automation, he continued, "to improve the productivity of the data scientist. We have a skills gap, and we have to make the data scientists we have 1,000 times more productive. We've done a study and determined it's quite possible with what we call massive machine learning automation."

Editorial standards