As I discussed in an earlier blog (Applied vs. Generalized Artificial Intelligence: what’s the difference and why it matters for enterprise IT), all AI is not created equal.
Just as applied AI is currently delivering more business value than generalized AI, natural language processing (NLP) is a key differentiating characteristic of AI systems.
To establish trust between employees and AI based self service options, the AI system must not only return faster responses than traditional support channels, it must also demonstrate an understanding the requester’s issue.
Today, one of the biggest challenges in AI is how to account for the complexity of natural language.
Getting on the same page
Human service and support agents are no strangers to the task of deciphering ambiguous and cryptic descriptions in help tickets, forcing them to spend considerable time screening and routing them.
Similarly, traditional knowledge bases require users to search for information using specific words or terms. However, it’s highly unlikely that every employee will search knowledge base articles using the same keywords that creators used when writing them. Likewise, expecting knowledge base creators and end users to use a standard syntax when creating and searching articles is unrealistic.
The same challenge can be found when leveraging scripted chatbots, robotic process automation or simple web-based forms for self-service. These also rely on a linear dialog that is programmable and predictable. It turns out humans are neither! The inability to understand context, express empathy and learn in real time makes these tools incompatible with natural conversation flow.
Can you hear me now?
Static knowledge base articles coupled with linear rules-based automation has resulted in the loss of credibility and stalled usage of employee self-service. In fact, employee experience data (related to internal IT support) from Happy Signals shows the self-service portal with the lowest score of all service and support channels and one that greatly contributes to lost productivity. Even physically calling the help desk is nearly an hour faster than attempting to use self-service.
To live up to the promise of making employees more productive, self-service needs to more closely emulate human-to-human customer service interactions. For instance, AI systems should be able to perform sentiment analysis to understand how a business user is feeling (i.e., dissatisfied, frustrated, and happy) and modify recommendations or courses of action accordingly, in real time.
Sentiment analysis must also understand the contextual meaning of words, allowing the AI system to choose the best definition of a word that can have multiple meanings and uses. For example, with a ticket description like “Salesforce isn’t running”, NLP can be used to determine that Salesforce is a business application and “isn’t running” indicates the application is failing to load properly.
Going beyond the script
Disparate and scripted automation technologies, like rules-based chatbots, not only create disconnected silos of automation, but also result in static and impersonal self-service processes and interactions.
Gartner predicts that this year, 75 percent of enterprises will have more than four diverse automation technologies within their IT management portfolios, up from less than 20 percent in 2014. Milind Govekar, research vice president at Gartner, said “the most current use of automation in IT involves scripting…scripts are more fragile than agile.”
Applied AI platforms that leverage advances in NLP and semantic analysis guide users through sophisticated business and IT processes using language in the way it’s spoken and written. Additionally, NLP-driven automation allows service management staff to focus more on creating high quality, personalized self-service experiences and less on guessing how employees will search for knowledge articles.