A Stepped Approach Is Key to Successful Chatbots

Article Featured Image

Although 23 percent of U.S. online adults communicate with businesses at least monthly using live chat or chatbots, chatbots are widely unpopular because companies obsess about the technology and neglect the user experience, observes David Truog, a vice president and principal analyst at Forrester Research.

Chatbots have disappointed many because of unkept promises, Truog argues, noting that developers have overpromised what their chatbots can do from the beginning. Many still configure chatbots to ask open-ended questions like “How may I help you?” only to see the chatbot unable to handle the wide variety of responses that such an open-ended question might invite, he says.

Many organizations have also mistakenly rolled out chatbots to customers prematurely, trusting that artificial intelligence will catch up quickly enough to overcome the initial poor-quality customer service.

Customer frustration aside, chatbots are popping-up on websites, in apps, in text-based messaging channels, and on devices equipped with Alexa or Siri. As that rollout intensifies, organizations will need to adapt conversation design: designing experiences based on conversational artificial intelligence.

“To create a successful chatbot, you need to make smart decisions about what actions it should be able to perform and the intent it supports,” Truog says

To identify intents, Truog recommends researching three sources:

  1. Assumed intents (what you know you are prepared to address). Forrester suggests starting with information sources, like FAQ pages. Though such information is typically not phrased in a conversational manner, FAQ pages and similar informational resources provide a good base for designing chatbots.
  2. Historical intents (what users have been asking you). Such details, though, are typically not available until your chatbot has been in use for a while, so Truog says until then, companies can examine records of what users have said in other channels.
  3. Adjacent intents (other things users would ask if they knew you could help). Troug recommends conducting research to identify unexpressed needs through ethnography, contextual inquiry, and skilled, in-depth user interviewing and observation. 

Truog also maintains that organizations should be prepared to postpone innovation if they are not ready. Most organizations’ motivations for creating chatbots still largely have to do with deflecting questions from human channels while ignoring adjacent opportunities to fulfill unmet needs that they have not expressed because they didn’t think the organizations could help with them.

The next step is to narrow down the list of intents to a specialized subset to avoid the chatbot becoming dysfunctional, Truog cautions. Similarly, he recommends that organizations take a phased approach to deploying chatbots.

The report quotes Jen Snell, vice president of go-to-market strategies for conversational AI at Verint. “Don’t just replicate website FAQs; that’s not solving the problem.”

“You need to decide whether your business goal is to deflect calls, build loyalty, sell more,” Snell says. “But you have to know what users are trying to accomplish by going through a UX design session with them. Companies think talking about their products is so important, but it’s not what users really care about, typically.”

Despite all of the resources devoted to designing and implementing chatbots, there are still issues that should be escalated to humans, Truog cautions. These include interactions that involve highly personal information or high-value customers.

CRM Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues
Buyer's Guide Companies Mentioned