When building chatbot software, keep it simple and user-focused

Deploying software robots that are truly helpful, and not just annoying or useless, takes some savvy. Experts share early lessons learned about this still-emerging technology.

For companies considering implementing chatbot software on their ERP systems to streamline workflows and increase productivity, it may be tempting to create a human-like bot that does everything but make coffee. However, experts advise focusing on specific use cases -- and letting chatbots sound like bots. This could be the best way to provide value internally.

A variety of companies, including Kore and Slack, have partnered with vendors such as Oracle and SAP to create chatbots for almost any use case. In the enterprise, chatbot software is gaining traction with human resources and procurement applications, and other systems are sure to follow as the business case is made for more self-service.

"Chatbots are essential for fostering discovery of information," said Mark Kurtz, chief growth officer and vice president of new media at Gage, a Minneapolis-based marketing agency.

The process of using a bot differs from searching for information. For example, an employee may search for something specific in the ERP system, but the chatbot may discover related information that the user didn't know existed. This could lead to new ways of solving business problems, Kurtz said.

Build chatbot software for specific use cases

Before chatbot software can be built and deployed, there are some caveats to keep in mind. Like the old saying that a jack of all trades is a master at none, so are chatbots.

"Instead of building a general purpose bot, like Siri, that answers everything, focus on a specific use case," said Kriti Sharma, vice president of bots and artificial intelligence at Sage, a U.K.-based maker of ERP software. "Even if the bot does just one thing right, it's good enough for users."

Continuing in that vein, Sharma advises enterprises to focus on tasks that users deal with frequently, rather than those they see once a year or once a quarter.

"If the bot is doing something like ... end of the year taxes, it's difficult to create traction [for the bot]," she said.

Chatbot software also doesn't have to be purely reactive, according to Sharma. It can be proactive, and do such things as sending reminders about overdue invoices.

Focus on the user experience

Natural language chatbots could be overrated, according to Boaz Hecht, CEO of San Francisco-based SkyGiraffe, which makes enterprise mobility platforms.

"When I think of SAP-related bots, I don't care as much about the conversational part of it; I care about the function of it," he said. The value of bots lies in their function, not their conversational capabilities.

To that end, overdoing the natural language interface could result in unnecessary steps for users. Hecht advises companies to choose frequently asked questions for the chatbot software to answer.

On the other hand, giving a chatbot a sense of humor can improve the user experience, according to Sharma. For example, a chatbot for the accounting department can use accounting humor to make the experience more pleasant.

The idea is to take the users through a journey: creating an expense report, answering questions about spending in a particular category and introducing capabilities, Sharma said. From the very first design phases, organizations need to gather user feedback to build a roadmap for the journey, such as asking for feedback after paying an invoice.

"The standard principles of product design change because users are giving feedback on what they want to be built," Sharma said.

Leverage AI and automation, but don't forget to involve humans

Creating a positive user experience also requires integrating artificial intelligence engines, which allow the bots to learn a language the more it is used, according to Gage's Kurtz. However, a human will still need to train the chatbots to match user intent with answers.

Additionally, when building chatbot software, it's important to pay attention to the domain. The bot needs to have enough automation behind it to pull data from systems, and to only ask users for further clarification when necessary, according to Sharma.

"The bot should be able to pull information based on the user," she said.

Don't forget, it's a bot, not a human

Experts agree that trying to pretend a chatbot is a human being will only detract from the user experience. The bot should introduce itself as a bot and explain its capabilities, such as assisting with report creation, Sharma said.

"We're strict on designing a bot as a bot," she said, noting that while some pretend to be able to process natural language queries, users usually end up disappointed.

For enterprises gearing up to add chatbots as a self-service option, experts agree that keeping it simple, leveraging artificial intelligence and automation, and not pretending it is human are critical for bots to provide a positive experience. Keeping the user first and aiming for frequently performed tasks will help the chatbot software gain traction and continue to build the business case for future development.

Next Steps

Learn about chatbots for SAP

Understand how chatbots can help HR

Read a review of the Talla bot

Dig Deeper on ERP and AI