Whether you know it or not, enterprise AI, or artificial intelligence is boosting the intelligence of various business processes, from the supply chain to customer service.
The age of enterprise artificial intelligence has emerged. Increased adoption of enterprise AI brings a wave of expectations in businesses and among customers who buy from them. Not all of these hopes about the potential of enterprise AI can be realized yet. So the question is: What capabilities can AI deliver today?
Enterprise AI (AI) is sprouting in various corners of companies. AI has generated smartbots—automated software programs that can conduct a variety of human tasks, such as scheduling meetings or ordering products online. While smartbots proliferate, machine learning (ML) has brought new capabilities and flexibility to enterprises. At the same time, it can be confusing to try to separate what AI and ML can realistically deliver from what seems more like science fiction. Enterprises have troughs of data to analyze, but finding the application of AI that will deliver true business value is difficult.
"Most people don't see the reality of current AI," said Jerald Hughes, chair of the Information Systems Department at the College of Business and Entrepreneurship at the University of Texas.
Enterprise AI's growing foothold is promising, he said, but user expectations may be overhyped, and some AI products already in the marketplace are ridiculous.
AI performs certain tasks quite well—and much better than human beings. One such task that has gained momentum is fraud detection in financial services.
The financial industry now uses AI to identify behavioral patterns that unearth cybertheft, and then monitors for those patterns.
"The cost/benefit balance here is effective,” Hughes agreed. “The impact of credit card theft is quite high, so we don't mind getting a high proportion of false positives," Hughes said.
But AI is not as adept at solving other kinds of problems. As machine learning algorithms mature and pick up behavior patterns, they can detect them in human interaction. Performance then opens the door to other desirable applications of AI. Still, complex tasks are ill-suited for AI.
Self-driving cars present an example of the challenge of complexity. Many cars sold today are equipped to detect unsafe driving, either within the car or in relation to nearby cars; this kind of vehicle “awareness” is a step toward autonomous navigation. But as reliance on AI in cars increases, so does the potential for the false positives, such as alerts about an impending collision that may be false.
"This would cause problems in self-driving cars," Hughes said. "If my car is always slamming on the brakes because it gets a false positive that there is a blockage or danger ahead, then it's going to bring traffic to a standstill."
AI is making inroads in the service industry using voice-activated devices. Chatbots take consumer behavior pattern detection to a new and interesting level. Chatbots already handle service calls from customers 24/7, and these bots can follow the patterns in a customer dialogue, detect changes in those patterns, and shift focus to stay on track in identifying a solution. This application of AI has, overall, reduced service desk costs and increased customer satisfaction rates. According to recent research on connected customer experience and the use of AI, 56% of organizations are actively looking for ways to use AI, but only 39% have a defined AI strategy.
But Hughes said that chatbots’ unequivocal success is still limited to contexts where inquiries are low level and product knowledge is simple to impart to a chatbot or instances where interactions that can rely on purely algorithmic functions.
“Paying your electricity bill over the phone to a chatbot is OK;” Hughes said. “God forbid you have some wrinkle that is not addressed by the algorithm, because at that point the AI becomes worse than useless."
One of the most visible applications of enterprise AI has been in natural language processing (NLP). Voice-activated devices have had enormous impact on consumer perception of machines that speak.
Machines understand human beings well enough today to respond to simple voice-generated requests, as in, “Who won the Red Sox game last night?” or “Add coffee to my shopping list.” This capability inspires expectations that AI will be able to do more. Consider voice-activated dictation that is more context-driven and iterative, such as, “Let’s plan our vacation in Europe this summer.” Now that machines understand human speech and text, real-world application can be increasingly rich in capablities.
Hughes said that while NLP offers promise with AI, there is nonetheless hyped expectation. True machine understanding of speech is still immature, Hughes said, because language is so driven by context.
"As humans using natural language, we have a built-in recognition and tolerance of the ambiguity inherent in language usage,” he said. “We use thousands of contextual cues all the time to process natural language, without realizing it. Phrases like, 'I don’t see it' can be figurative or literal, and AI doesn’t yet know how to understand the difference.
"Those systems have no capacity to deal with ambiguity,” Hughes said. “Even the best AI today simply does not have that background of experience that enables humans to create meaning in the presence of ambiguity."
Another use of enterprise AI that is becoming commonplace is in C-suite decision support. Many enterprise software platforms, from customer relationship management to customer experience and enterprise resource planning, now include built-in analytics, with machine learning easily accommodated.
This is an important step forward for companies seeking competitive advantage, because they can effectively use data as the foundation for future decision making. Data-driven decision making has raised expectations about how business decisions will be made in the future.
"Consumers will expect smart, flexible, customized responses to everything they do,” Hughes said. “Producers will hope for smart, flexible, customized business operations in every aspect. We see this tension already in new businesses like Amazon and Uber, and in the notion of a gig economy generally.
In some industries, AI is bringing about fundamental changes: not merely improving operations and efficiency, but also enabling entirely new possibilities. Perhaps the most significant of these, from the perspective of benefit to society, is healthcare.
AI has been applied in healthcare in several profound ways in recent years, from clinical diagnostic support to data-collecting chatbots to pattern detection in claims analysis. These applications already produce cost savings, earlier detection of health problems and more accurate diagnosis.
"AI is already doing amazing things in reading X-rays and scans,” Hughes said. “It can call attention to things that could be missed otherwise, helping doctors make better decisions. It can seek out patterns in what would otherwise be hopelessly large and complex bodies of data."
New enterprise AI-enabled applications can help shape human behavior by tracking diet and exercise regimens and even warn proactively about harmful behavior. AI could analyze too much sedentary behavior or too many hours at the office, even identifying unhealthy food choices.
According to Hughes, what we experience today signals just the beginning of how AI can track, analyze and guide human behavior toward better decision making in the future.
"The great thing about AI is that it can pay attention [to our behavior] in a narrowly targeted way, without ever getting tired," he said. "There are thousands of narrow-slice AI applications here, waiting to happen."
Scott Robinson is an enterprise architect and AI consultant with a 25-year history in business intelligence, analytics, and content management in the healthcare and logistics industries. He is currently CIO of the GlenMill Group, a research consortium providing new AI technology and infrastructure for enterprise applications and services.