The impact of machine learning on the architecture of enterprise applications and the enterprise itself in the 21st century was the dominant theme of Workday's annual summit with industry analysts held yesterday. Machine learning will be as disruptive as cloud computing, Workday co-founder and CEO Aneel Bhusri told attendees in his opening remarks:
In the same way that cloud defined how Workday was born, machine learning's going to define us over the next ten years. It is the fundamental technology that, if we get it right, will continue to disrupt the marketplace, will continue, importantly, to help our customers make better business decisions.
He added that organizations that don't figure out how to harness machine learning will be left behind by the competition, a theme that was taken up by other speakers during the day. As a result, data science skills are in high demand, he added in later remarks:
If the hot job 20 years ago was a computer science major, today it's a data scientist. Computer science is passé now ... The demand for data scientists is so far ahead of the number being created.
The impact on Workday's own products will be that they will acquire the ability to evolve as people use them, says Workday CTO David Clarke:
By using data, these systems can and do and should change and evolve and get smarter as you use them. This is a flywheel. The more you transact, the more data you generate, the more you can train algorithms, the better the systems can get.
This characteristic of responsiveness and change is something that is unique to modern architectures and is strongly differentiated from what we've seen historically.
Sayan Chakraborty, Senior Vice President of Technology, elaborated on this point and its existential significance for competitive differentiation:
Our job is to keep customers competitive and relevant, [so you can] make decisions as fast as your competitors. If they are able to turn the decision cycle faster than you, they will dictate where your investments go and they will be ahead of you.
Machine learning to keep customers ahead
Workday sees its role as providing an engineering foundation and a trust model for machine learning that uses its scale to keep customers ahead of the game. The goal is to deliver business value to customers that iteratively improves as the system learns from usage over time. Chakraborty explains:
Machine learning's primary characteristic is, the more you use it, the better it does — the better it will get for all of our customers ...
The natural evolution for machine learning and the enterprise [is] turning the decision cycle faster and faster. The key word is continuous — being able to move faster and better.
This changes the nature of an enterprise application from an inflexible system of record and processes into one that is able to suggest changes and modify its behavior based on introspection and feedback, he says:
A [traditional] enterprise application asserts a bunch of facts, and then it asserts a bunch of rules. It's a collection of facts, and a collection of rules around those facts.
What machine learning asserts is that facts are probable facts, rules are probable rules. It suggests potential facts and suggests potential rules. As behavior changes, it modifies itself. You actually have a learning enterprise application that is self-modifying, and the better able it is to deliver the best facts and the best rules.
The first example of this principle in action that Workday expects to bring to market will be a self-modifying user experience, he says.
Your user experience needs to adapt to you as your needs change. [The self-modifying UX] is designed to learn from and adapt to that user. That is the beginning of what is a very interesting arc for Workday.
The aim is not to replace users but to augment their ability to operate safely and make effective decisions, he notes.
What Workday outlined today came across as a mature use of machine learning to help solve business needs in a fast-moving world, taking advantage of the vendor's scale and its ability to harness the public cloud providers' machine learning services in very sophisticated ways. My worry is that organizations are finding it harder and harder to digest new technology and that a constantly self-modifying application just steps up the pace of change even more. They are going to need lot of handholding to digest what's coming.