zapp2photo - stock.adobe.com

Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

Successful chatbot content strategies rely on tools, AI

Chatbots aren't a panacea curing customer support issues. Organizations must hone their content quality, organization and management to create successful bots.

Residents of North Charleston, S.C., can now report a pothole via chatbot. If they are Allstate Insurance customers and running over the pothole disables their vehicle, they can ask a chatbot if their insurance plan covers the damage, right there on the side of the road. After that, they can ask Amazon's Alexa voice assistant to start the dishwasher at home as they wait for a tow truck.

Many organizations will create chatbots in the next few years, but well-organized and comprehensive chatbot content will likely be the best predictor of success.

When chatbots work well, the benefits are substantial. Customers can get quick answers to questions and companies can lower the costs of phone support. Chatbots provide 24-hour service, instant responses and the ability to answer both simple and complex questions.

How chatbots work

Chatbots demand that organizations develop product content for the many ways in which users will consume it.

"Chatbots are hard to do," said Paul Wlodarczyk, former project lead at Earley and Associates, the Chicago company that helped Allstate implement its Allstate Business Insurance Expert (ABIE) chatbot. He is currently digital practice lead at Dakota Systems Inc. "But if you contextualize where the chat is happening and carve the problem up, it becomes more manageable." 

When an Allstate customer requested an FAQ, for example, Wlodarczyk's team needed to chunk FAQ-like content into question-and-answer pairs. His team mined support call logs to understand how agents refer to their insurance card and added the appropriate synonyms to the vocabulary in chatbot content, such as certificate of insurance or insurance ID. Then, the chatbot uses those synonyms to recognize different forms of the question and obtain the answer.

EasyDITA, a content management platform, manages Allstate's authoring and content and then pushes the content to Dialogflow, the chatbot engine. EasyDITA also helps to evaluate the input to the chat engine to refine the chatbot content over time. The website team can also monitor the chatbot logs to add to its understanding of the questions that users ask and how they ask them.

It's not feasible to manually handcraft responses and content for all possible customer questions.
John SprungerSenior technical architect, West Monroe Partners

Content management systems, such as easyDITA, can be helpful, but a chatbot tool itself can also manage the content, said John Sprunger, senior technical architect at West Monroe Partners, a Chicago-based technology consulting firm.

Organizations need more than good content to create a chatbot, however. Well-crafted content is essential, but machine learning will also play a big role, said Melissa Webster, vice president of content and digital media technologies at IDC in Framingham, Mass.

"There's no reason why machine learning can't be harnessed to maintain taxonomies and synonyms -- after all, our terminology evolves," Webster said. "Plus, AI can help by ranking terms 'most likely' as systems try to better understand and anticipate the user's intent."

Support the bot

Other organizations may not have as content-heavy chatbot applications as Allstate does, but their customer support chatbots still must have a library of product support chatbot content, such as documents, images, videos, text-based conversations and audio, that a bot can use.

"It's not feasible to manually handcraft responses and content for all possible customer questions," Sprunger said.

Chatbot success
There are mixed results when it comes to the effectiveness of chatbots for customer support.

HP Inc. in Palo Alto, Calif., sells a lot of products, including printers and all of their parts. To identify which product the user is inquiring about, the chatbot must run a diagnostic first. For example:

User: "Chrome runs very slowly."

HP virtual agent: "Which of these options best describes what you are looking for?"

User chooses: "Certain apps in my computer cause lockups."

After the chatbot successfully identifies the product, it then can walk the user through a complex set of steps to run a diagnostic tool. That tool can identify the problem and begin troubleshooting.

This was last published in January 2019

Dig Deeper on Enterprise content management, collaboration and search strategy

Join the conversation

2 comments

Send me notifications when other members comment.

Please create a username to comment.

How does your organization manage content for a customer support chatbot?
Cancel
Wow such an interesting chatbot use case. Along with these interesting use-cases, we also need to get better on user experience, more business process integrations, and making the interaction models more mature and intuitive. We have started on the journey at Engati.com, do visit us to give us feedback :)
Cancel

-ADS BY GOOGLE

SearchBusinessAnalytics

SearchDataManagement

SearchERP

SearchOracle

SearchSAP

SearchSQLServer

Close