Joseph Selle, a specialist in enterprise A.I. transformation with IBM, relates how the company has embraced the notion of a "cognitive enterprise," and tells how it has transformed business processes across the organization.
Q: Just what is the “cognitive enterprise”?
Selle: It’s more than a buzz phrase. A cognitive enterprise is a company that has decided to invest in data and technologies that will allow its workers to do their jobs better than before. Artificial intelligence is another synonym for cognitive. An A.I. enterprise is one where there are vast quantities of data available for workers to manipulate, with tools that allow for extraction of information from that data.
Q: Instead of mimicking the human brain, it sounds like it's doing something that humans are incapable of doing because of the massive amounts of data involved. Is that the case?
Selle: It’s a little of both. It’s providing humans with the ability to handle and manage data in quantities that they couldn't have before. That's an enhanced capability. It's like taking a person and making them a super professional. If you’re a supply-chain planner, for instance, and you are operating within a cognitive enterprise that has been investing in data and the right tools, you're going to be twice as good in your job.
If you're a procurement professional, you might be four times better as a negotiator because you know things about price movements in the future. You have models at your fingertips that are telling you whether the prices are going up or down, and whether or not you should be a hard bargainer. Whatever your job is, it's not only just supply chain. It’s finance and human relations and asset management and all of the functions of the company. If those functions have A.I. or cognitive capabilities around them, the people doing the jobs are doing them faster and more accurately.
Q: The early days of A.I. were about creating what were then called expert systems — in other words, replicating the experience of a human expert. What you're talking about seems to go beyond that.
Selle: It does. The expert system idea was big 20 years ago. Now we think about an A.I. system as a digital assistant. It resides on my smart device, and I can ask it questions like “There’s a hurricane that's about to hit North Carolina. Are any of my suppliers at risk of being shut down?" And the system will respond, "There is a category-four storm coming towards North Carolina. It has a such and such probability of becoming a category three and a half by the time it makes landfall. You have three suppliers there, and they are preparing to ship parts that will be incorporated into a certain order." I can connect all of those databases and get a vision of what the impact to my customers will be, which is really what I care about.
Q: Would the proper term for such a system be cognitive analytics?
Selle: Yes, although analytics might not be the most current term for it. The hierarchy of terms is visibility, prediction, and prescription. Visibility is getting the information to a place where you can draw a conclusion from it. Prediction is using probability to understand where a price is going to go, and prescription is telling the procurement person what to do about it.
Q: At the prescription phase, what do the humans do? Do they blindly go along with what the system is telling them?
Selle: What we’ve found is that people don't trust the computer, especially when it gives them a challenging answer. Say you're a good old-fashioned supply-chain person who’s planning safety stock, and you thought you needed eight days of stock on a certain part to make sure you're never going to run out. Through some very sophisticated modeling, we’re able to tell that person that they only need two days of safety stock, and that this willl save hundreds of millions of dollars in inventory holding costs.
Q: That's a scary thing to say.
Selle: It is. They have to be convinced. We've learned that you need to include the probability confidence in the answer, and the reason why the system came to that conclusion. When you give that background information, it turns out that the prescriptive advice is more closely followed.
Q: The term machine learning implies that it becomes more accurate in its recommendations over time. Does the system get better?
Selle: It does. Machine learning is the ability of the computer to understand and see patterns in data. If you give the computer a block of data this big, it's going to see certain things. If you give it a block that's twice as big, it's going to see more. You're teaching the system with more historical data, or greater breadth of data.
Q: Tell me about IBM's implementation of a cognitive enterprise “data lake.”
Selle: Think of it as a pyramid, where the bottom layer is the data. It's structured, unstructured, and batch data, streaming from the internet of things, consisting of devices that you’ve deployed across your network. In the middle layer are tools that organize the data and bring structure to it, because the computer can't ingest unstructured data. And on top are the applications and solutions that the users touch. As for the data lake, think of it as the big repository of data that the stack is built upon, and at the top is the application that needs that data.
Q: To what degree is all this aspirational versus reality? How much is actually in practice and happening right now?
Selle: I am privileged to be working at a company that has been investing in this area since the nineties. We created the Deep Blue computer that eventually beat Garry Kasparov, the chess grandmaster. The next challenge we tackled was the Jeopardy game, which allowed us to go into some new areas of natural language understanding and speed of processing. We've been pouring research dollars into this question for a long time. Now we have a data lake, the ability to structure that data, and a portfolio of solutions that are aiding the jobs of human resources professionals, procurement and the supply chain.
Q: I understand it that you’ve expanded your brief beyond supply chain into additional functions throughout the organization. How wide-ranging is this?
Selle: Although supply chain and procurement were the first adopters, it has since expanded to 19 or 20 domains within the back office of IBM, such as finance, sales and marketing — any of the normal functions that make up a business. An example is our cognitive pricing recommendation engine. We’ve given it to our salespeople in the field, and it lives on their smartphones. When they're discussing a bid with a client, they can put the factors of the bid into the system and it will return them an optimal price. As a result, our salespeople can be much better bargainers. We’ve deployed the system over the last two years across multiple geographies and products. Wherever we put it in, we find that our win rate and profit dollars improve. And customers like it because it compresses the timeline of the whole process.
