Four Things Every Leader Should Know About Applying Artificial Intelligence To Business
IPsoft is, in many ways, an unusual entrant into the crowded, but burgeoning, artificial intelligence industry. First of all, it is not a startup, but a 20-year-old company and its leader isn’t some millennial savant, but a fashionable former NYU professor named Chetan Dube. It bills its cognitive agent, Amelia, as the “world’s most human AI.”
It got its start building and selling autonomic IT solutions and its years of experience providing business solutions give it a leg up on many of its competitors. It can offer not only technological solutions, but insights it gained helping businesses to streamline operations with automation.
Ever since IBM’s Watson defeated human champions on the game show Jeopardy!, the initial excitement about AI has led to inflated expectations and often given way to disappointment. So I met recently with top executives at IPsoft to get a better understanding of how leaders can successfully implement AI solutions. Here are four things you should keep in mind:
1. Match The Technology With The Problem You Need To Solve
AI is not a single technology, but encompasses a variety of different methods. In The Master Algorithm veteran AI researcher Pedro Domingos explains that there are five basic approaches to machine learning, from neural nets that mimic the brain, to support vector machines that classify different types of information and graphical models that use a more statistical approach.
“The first question to ask is what problem you are trying to solve.” Chetan Dube, CEO of IPsoft told me. “Is it analytical, process automation, data retrieval or serving customers? Choosing the right a technology is supremely important.” For example, with Watson, IBM has focused on highly analytical tasks, like helping doctors to diagnose a rare case of cancer.
With Amelia, IPsoft has chosen to target customer service, which is extraordinarily difficult. Humans tend not to think linearly. They might call about a lost credit card and then immediately realize that they wanted to ask about paperless billing or how to close an account. Sometimes the shift can happen mid-sentence, which can be maddening even for trained professionals.
So IPsoft relies on a method called spreading activation, which helps Amelia to engage or disengage different parts of the system. For example, when a bank customer asks how much money she has in her account, it’s a simple data retrieval task. However, if a she asks how she can earn more interest on her savings, logical and analytical functions come into play.
2. Train Your AI As You Would A New Employee
Most people by now have become used to using consumer facing cognitive agents like Google voice search or Apple’s Siri. These work well for some tasks, such as locating the address for your next meeting or telling you how many points the Eagles beat Patriots by in the 2018 Super Bowl (exactly 8, if you’re interested).
However, for enterprise level applications, simple data retrieval will not suffice, because systems need domain specific knowledge, which often has to be related to other information. For example, if a customer asks which credit card is right for her, that requires not only deep understanding of what’s offered, but also some knowledge about the customer’s spending habits, average balance and so on.
One of the problems that many companies run into with cognitive applications is that they expect them to work much like installing an email system — you just plug it in and it works. But you would never do that with a human agent. You would expect them to need training, to make mistakes and to learn as they gained experience.
“Train your algorithms as you would your employees” says Ergun Ekici, a Principal and Vice President at IPsoft. “Don’t try to get AI to do things your organization doesn’t understand. You have to be able to teach and evaluate performance. Start with the employee manual and ask the system questions.” From there you can see what it is doing well, what it’s doing poorly and adapt your training strategy accordingly.
3. Apply Intelligent Governance
No one calls a customer service line and asks a human to talk to a machine. However, we often prefer to use automated systems for convenience. For example, when most people go to their local bank branch they just use the ATM machine outside without giving a thought to the fact that there are real humans inside ready to give them personalized service.
Nevertheless, there are far more bank tellers today than there were in before ATMs, ironically due to the fact that each branch needs far fewer tellers. Because ATMs drastically reduced the costs to open and run branches, banks began opening up more of them and still needed tellers to do higher level tasks, like opening accounts, giving advice and solving problems.
Unfortunately, because cognitive agents tend to be so much cheaper than human ones, many firms do everything they can to discourage a customer talking to a human. To stretch the bank teller analogy a little further, that’s almost like walking into a branch with a problem and being told to go back outside and wrestle with the ATM some more. Customers find it incredibly frustrating.
So IPsoft stresses to its enterprise customers that it’s essential that humans stay involved with the process and make it easy to disengage Amelia when a customer should be rerouted to a human agent. It also uses sentiment analysis to track how the system is doing. Once it becomes clear that the customer’s mood is deteriorating, a real person can step in.
Training a cognitive agent for enterprise applications is far different than, say, Google training an algorithm to play Go. When Google’s AI makes a mistake, it only loses a game, but when an enterprise application screws up, you can lose a customer.
4. Prepare Your Culture For AI As You Would For Any Major Shift
There are certain things robots will never do. They will never strike out in a little league game. They will never have their hearts broken or get married and raise a family. That means that they will never be able to relate to humans as humans do. So you can’t simply inject AI into your organizational culture and expect a successful integration.
“Integration with organizational culture as well as appetite for change and mindset are major factors in how successful an AI program will be. The drive has to come from the top and permeate through the ranks,” says Edwin Van Bommel, Chief Cognitive Officer at IPsoft.
In many ways, the shift to cognitive is much like a merger or acquisition — which are notoriously prone to failure. What may look good on paper rarely pans out when humans get involved, because we have all sorts of biases and preferences that don’t fit into neat little strategic boxes.
The one constant in the history of technology is that the future is always more human. So if you expect to cognitive applications simply to reduce labor, you will likely be disappointed. However, if you want to leverage and empower the capabilities of your organization, then the cognitive future may be very bright for you.
An earlier version of this article first appeared in Inc.com
Wait! Before you go…
Choose how you want the latest innovation content delivered to you:
- Daily — RSS Feed — Email — Twitter — Facebook — Linkedin Today
- Weekly — Email Newsletter — Free Magazine — Linkedin Group
Greg Satell is a popular author, speaker, and innovation adviser who has managed market-leading businesses and overseen the development of dozens of pathbreaking products. Follow Greg on Twitter @DigitalTonto. His first book, Mapping Innovation, was selected as one of the best business books of 2017 by 800-CEO-READ.
NEVER MISS ANOTHER NEWSLETTER!
I still remember the day we got the email telling us that the in-person version of the 2021 Consumer…Read More