AT&T is investing in both chatbots and artificial intelligence (AI) to support its business customers and Sorabh Saxena, president of business operations at AT&T, knows the drill. “It’s important for chatbots to be personal, professional, accurate and friendly,” he said. “It is also necessary to be able to escalate to a real person quickly when they cannot solve the customer’s issue.”

There’s more of course, he continued. “AI needs to not only understand the products and services that AT&T provides, but also be able to put them in context of the customer’s entire solution.” For example, Saxena said, a customer may buy connectivity from AT&T, cloud services from another vendor and security services through yet another. A “‘conversational experience’ then needs to integrate all of this knowledge to provide a support model which will encompass all of the customer’s needs quickly, accurately and with a personal touch,” he said.

The company knows that providing a stellar conversational experience can be a daunting task. “To achieve a best in class conversational experience, our tools must be accessible, in a variety of formats, and easy to use,” Saxena said. “As customers want their questions and issues resolved quickly, it is imperative that we have an accurate knowledge base for responses. Additionally, nothing replaces the personal experience. We must have the ability to go deeper in understanding the context of the issue or question and provide solutions or recommendations accordingly.”

As it sets out to achieve all this, AT&T is quite aware of the challenges that lie ahead. The solutions and networks created for business customers are increasingly complex, Saxena said. Also, “integrating data on a customer specific level to provide accurate responses is difficult, but necessary to provide a positive digital experience.”

AT&T is not alone in its assessment of the path ahead for providing a customer conversational experience. The experts we spoke with agree that it's a difficult one to some degree and furthermore, few companies have yet to achieve the ideal scenario of an easy back-and-forth between the company and its customers. There are many reasons for this.

Related Article: What Is a Conversational Experience?

The Limits of Bots, AI

There are difficulties with AI and natural language processing technology that underpins the process, said Ajeet Kushwaha, director of bots and automation at customer engagement company Freshworks. “In order to build a true conversational experience, bots need to understand the intent correctly and then engage the user contextually,” he said. “Understanding intent and maintaining context require a lot of machine learning/AI and natural language processing/natural language understanding capabilities which are still evolving.”

One specific roadblock right now is that bots require humans to say very specific keywords in order for the automated machine to link it to an intention. This means that the user cannot be truly conversational in how they talk to the bot, said Diana Lee, a conversation designer with customer experience platform provider WIZELINE. “Think about it like this: if you showed two dresses to a friend and said ‘which one do you like better?’ what would you say? You would most likely point to one dress and say ‘that one.’” Currently, the way a bot handles this conversation requires that you to use the product name because it doesn't hold the context that a human brain naturally holds, she said.

Learning Opportunities

A voice bot has the same challenge of needing to be specific about the product to trigger the voice assistant picking up keywords, continued Lee, “far more than you would need to be if you were to ask a store clerk.”

Related Article: Calling All Linguists: The Messaging Bots Need Help

Changes for the Agents

There can also be difficulties on the agent side due to the changes in process and measurement needed to provide conversational experiences, Kushwaha continued. There may typically be more agent hand-offs for conversations that happen over longer periods, for example. There may be more creative ways of setting customer expectations on response times and systems that know when to bring in new team members automatically. Also, he added, the metrics used to measure transactional experiences are not relevant or are in direct contrast to those based around conversational ones. “You can't look to minimize ‘average handle time’ when you want to allow your customer the convenience of engaging on their schedule,” he said. “You may need to redefine when it is OK to move a conversation to a closed or resolved state.”

Timeliness is another necessary feature of conversational experience and that too can cause difficulties for companies trying to implement it, said Katie Wilson, founder and CEO of direct advertising platform TapOnIt. “It can be challenging to keep multiple one-on-one conversations going simultaneously, especially as your customer base grows,” she said.

It gets only worse if you’re maintaining various conversations across multiple channels, said Shannon Howard, a senior digital marketing strategist at Overit Media. “If you're not using software that centralizes communications, and you're individually going into each social platform email, text, etc. to respond to inquiries, there's a really strong chance that something is going to fall through the cracks.”