The Bots Are Coming, and Fast
When examining their options, firms need to be like Goldilocks, holding out for systems that are just right, according to Ian Aitchison, CEO of the Asia Pacific region at COPC, a global consulting company.
Businesses do not want to start with too simple of a business process, he says, because the results will be insignificant. Nor do they do want to take on too complex of a task because the project might fail. Furthermore, if the current business process is weak, the bot will not necessarily strengthen it; often, firms might have to revamp their operations before they can begin to automate them.
Next, companies will need to build up their systems’ reasoning. Bots typically rely on knowledge that the companies deploying them make available to the systems. When customers ask questions or provide information, bots select the closest-matching responses from preprogrammed repositories and then use that knowledge to craft the appropriate responses. But mistakes are possible, especially if the data on the back end is inaccurate or out of date.
And then companies should never lose sight of the fact that bots are still an evolutionary technology, one involving a lot of trial and error. “When we started, we tried too hard to be perfect right away,” Autodesk’s Spratto admits. “Instead, I would recommend that companies get something up fairly quickly and then focus on improving it.”
Complicating things further, both the front end and the back end of companies’ systems have become more complex. Customers can interact with companies in so many ways, from phone and email to social media. Bots need to support all of these possible user options. And the software also has to be tailored to various interfaces. “We found developing with Watson was easy, but tying the software to the back systems was difficult,” Spratto says. Autodesk relies on inContact, SAP, Salesforce.com, and Tibco for its back-end processing, and each required its own integration. As a result, deploying these systems costs time and money. Autodesk started with a team of about half a dozen IT pros, and that number has grown fourfold since then.
For many businesses, though, justifying such expenditures could be a real challenge. “Right now, few metrics or benchmarks are available to help businesses understand bots’ impact,” COPC’s Aitchison says. But, he points out, such a lag is typical when any new technology first emerges.
Going forward, many vendors expect to add sentiment to their current bot technologies. Conversational bots, they hope, will be able to act like empathetic humans, understanding not only the literal translation of the words that people speak but also the consumer’s emotional state. In this case, the bots’ expressions and responses could change dynamically during one interaction or from one interaction to the next.
But in the meantime, bots continue to be new, emerging technology creating significant buzz. A lot more time, money, and manpower will be needed to build them up if machines will ever be able to interact as effectively as (if not better than) humans.
Paul Korzeniowski is a freelance writer who specializes in technology issues. He has been covering speech technology issues for more than two decades, is based in Sudbury, Mass., and can be reached at paulkorzen@aol.com or on Twitter at #PaulKorzeniowski.
AI and Machine Learning Aren’t the Same
What are the differences in how humans and machines think? Historically, the distinctions were easy to discern. But with IBM’s Watson able to get the better of humans on the TV game show Jeopardy!, the dividing lines have become murkier. Those lines will blur even further as artificial intelligence and machine learning continue to evolve.
AI is a generic term for the “smarts” exhibited by machines. The field operates under the assumption that human intelligence can be described so precisely that a machine could be programmed to simulate it. Typically, a machine mimics the cognitive functions that humans associate with reasoning, such as problem-solving and learning. Programmers outfit systems with intelligent agents that can make assumptions and take actions that maximize the chances of reaching predefined goals.
AI’s recent resurgence follows advances in computing power, which now enable developers to use larger data pools and build more complicated programs.
AI can have a positive impact on customer service. The software is being used to help contact center staff more effectively interact with customers. With contact volumes growing, many enterprises have trouble keeping pace with service requests. Currently, AI offloads simple, tier-one support requests from agents so they can spend more time on complex tasks.
Machine learning, a follow-up to AI work, is a method of data analysis that further improves system reasoning. Relying on algorithms that monitor interactions and iteratively learn from them, this approach enables computers to perform functions without explicit programming. Machine learning differs from AI in that as the system is exposed to new data, it adapts independently. AI systems, by contrast, have to be told what to do.
At the moment, AI and machine learning are relatively unsophisticated, automating only the most routine interactions. Eventually, the technology is expected to mature and perform more intelligent, human tasks, such as recognizing that a caller is upset and taking steps to defuse his anger. —P.K.