Recently James Taylor, CEO of Decision Management Solutions, was featured on the IBM ExpertTV series “Two Questions About Automation.” Here’s a run down of the segment and the questions he answered.
Host David Jenness asked James, “Why are companies struggling with AI and ML?”
According to James, there are three main root causes. The first is that companies start the project by looking at the data. But “just because something is interesting, does not mean it is useful.” Instead, companies should start with the business problems they have.
The second reason companies struggle with AI and ML is that they think of them as stand alone technologies, whereas they need to be looking at them as part of a transformation stack, a digital business stack.
And thirdly, companies often approach ML and AI projects with a pressure to deliver an immediate strategic impact. That’s not the right approach to create successful outcomes. They should approach these technologies with more of a long term focus and “forget about trying to show up on day one with a big bang at the bottom line of the financial statement.”
“So, what is the correct approach?”
James suggests focusing on operational decisions that are made day to day, such as, “What should I say to this customer when they call to cancel their service? How do I attract this prospect to become a customer? What’s the first product I should offer them? Can I approve this transaction? What discount should I give you?” These are the kinds of questions that work well with ML and generate the valuable data.
In other words, focus on the decision to be made. In particular, focus on repeatable high volume decisions. Also, James says, mix AI and ML with business rules – a business rules management system works particularly well. Capture the guide rails, the regulations to be in compliance with, and best practices, not necessarily with data, so that they can be wrapped around the ML problem, and then you solve the whole problem, not just the ML problem.
Lastly, James says, don’t try to get a home run on day one. “You can’t bite off that much ML in one go. You need to test, learn, experiment. Create an environment for continuous improvement. Do I know how I made decisions using this ML and business rules in combination?” The goal is to gradually and systematically get better over time. In this way, companies are much more likely to get value out of their ML investments.
Post Show Discussion
In the live Q & A session that followed the segment, the question came up: “What is the role of the data scientist in the methodology described? Do they collaborate with the automation team? How does that work?”
Ideally, James says, the data scientists, automation folks, and business people would be in the same meeting, to get a solid understanding of the business decisions. The question on the group’s mind should be, “What would help make the decision better? If only we knew _(blank)_ we would make the decision differently.” Then try to figure that blank out. What models would be useful in the context of the current decision making?
“We’ve found that [business] rules people tend to know the rules. ML people know the data. No one asks the people who actually make the decisions everyday,” says James. “If you don’t understand those things, you can’t write the right rules and models. You gotta go listen to the people who actually approve claims and underwrite policies.”
These are big companies we’re talking about, and the work has to be done thoughtfully. It’s a whole discovery process, called “decision discovery,” that uses a standard notation called the decision model and notation standard. This allows you to visually blueprint what people are telling you, and break a complicated decision down into sub-decisions which are not very complicated. The model allows you to target the pieces more systematically.
“What are some of the more common uses cases for AI/ML?”
“Customer facing stuff,” says James, “next best action, next best offer, account management, what should I say next to this customer to make them a better customer?” Other classic ML targets relate to LTV (life time value) and retention. “What is this customer’s propensity to churn? What data about this customer helps to make that decision?” These are some of the higher ROI projects where AI/ML can be effective.
————————————————-
To watch the segment, go to https://techtv.bemyapp.com/ and register. Once logged in, navigate to the search bar and search “James Taylor” to pull up the segment and the post-show discussion.