A key value of chatbots is that their artificial intelligence means they can provide answers and be ready to use right out of the box. However, while bots are smart, the reality they learn over time, so many organizations are realizing bots should still be trained by the organization to perform at their optimal levels.
“Chatbots are just virtual tools and depending on how they’re built, they can actually reside in multiple virtual spaces,” said B2B Marketing Consultant Pam Didner. “Some depend on what features have been added to it. They can have a very basic function that you have to manually set up, others have an AI functionality built into them where you can actually train the tool.”
The key differences between a chatbot and an AI assistant or AI-powered chat tools are:
- A chatbot is used to interact with humans via text messaging for tasks such as customer support to answer basic questions around product information, to set up meetings or share related content. Chatbots are not configured with AI; they rely on a rule-based system, where the user can pre-configure it to answer common questions. Other types of bots trigger a live person to engage with a customer via text.
“Chatbots are very conditionally-logic driven, which means they have defined rules that say, ‘If you do this, then do that’” or ‘If you like this topic, go to this web page,’” said Rashmi Vittal, CMO of Conversica. “You could think of it is as an intelligent router.”
- AI-powered assistants and tools are much more intelligent and can be used to support (and mimic) human activities with artificial intelligence and natural language processing.
- Interactions with a chatbot are less conversational than an engagement with an AI assistant. With AI assistants, the recipient may not even realize they aren’t interacting with a real person.
“Virtual assistants are really about repeatedly touching and engaging in a two-way back-and-forth dialogue between itself and the contact,” said Vittal. “These asynchronous conversations are similar to the normal concept of how someone would conduct an email conversation.”
AI applications and bots are constantly evolving, but it’s up to the user to “train” the assistant to ensure proper engagement. Currently, B2B businesses are leveraging them to enhance customer support on their websites and answer customers’ questions or share promotions and content. Other organizations such as the telecommunications company CenturyLink are taking it beyond just a chat function on their landing pages to help them close deals, much like a real sales representative.
The company has had great success with its AI-assistant named Angie. Prior to implementing the Conversica Intelligent Virtual Assistant, its lead optimization and lead routing cycle were all done manually.
“We’ve got Conversica plugged into our Marketo instance so we’re able to get people through generic nurture to a certain point and send them over to Angie,” said Kate Cindric Federhar, Sr. Manager of Marketing Operations at CenturyLink. “It allows us to contact a large volume of customers and prospects to get them to get people in the door, in the funnel and start to talk to sales quicker.”
Scott Berns, Director of Marketing Operations at CenturyLink added, “It simulates your ability to reach out and speak to customers and prospects in a cost-efficient manner. And it learns as it goes. It gets smarter as we get smarter.” Berns noted that for every dollar the team spends on the assistant, they get about $20 in return.
Evolving Bots & Assistants Requires A Human Touch
Whether it’s a simple chatbot or a more intelligent assistant, there is some manual work to be done on the backend to “train” the technology to have relevant and authentic conversations with the contact.
For basic chatbots, users must build in a workflow for the bot to know how to engage with a contact, according to Didner. They have to know their audience and predict the types of challenges and situations they are in so they can configure the bot to answer their questions.
“On the B2B side, a lot of stuff that we do is predictable,” said Didner. “For example, when people come to our website, they probably want to find out something about a product. They want to know how to use it better or they want to talk to a salesperson. There are predictable scenarios that you can build a workflow around to facilitate that conversation.”
However, Didner emphasized that this is not a “one-and-done” type of approach. It’s important for marketers to often go back and update how the bot engages with contacts based on previous conversations and new data and insights about their audience.
“Think of email campaigns, for example, and how B2B marketers check open rates and conduct A/B testing to test subject lines and different content,” said Didner. “We send different samples and we’ll go back to see the results and then we’ll modify it. That same analogy applies to AI training. You need to set up parameters that help the AI to learn. Marketers should come back after a while and see how they are learning so they can feed the right information based on the right questions. It’s kind of like training a dog.”
More intelligent, AI-powered assistants require similar backend work, but down the line, they continue to learn with every passing conversation they have with a contact. With the AI already built-in, the bot is capable of having natural conversations with contacts through email and SMS texting.
“It takes natural language understanding and deep machine learning capabilities to not only be able to decipher what’s being said but for it to learn from the interactions it has,” said Vittal.
Platforms such as Conversica have help centers in place to direct the AI to continue to learn, as well, because experts agreed that all AI solutions should have a human involved in some way, shape or form.
“[The training desk] looks at the exchanges of messages that our assistants are having with the contacts,” said Vittal. “And should AI not be confident in how it’s either interpreting a message or how it should respond to a message, it sends an alert to the desk team to get the guidance it needs to know how to respond. Once it receives that guidance, it learns from it and therefore, the next time it happens, it gets better and smarter.”
Tasking Marketing Teams To Own The Bot For Continued Training
Given the technical components of chatbots and AI assistants, it’s quick to assume that IT should own and manage the technology. But Didner strongly believes it should be the marketing team that owns and monitors it. This allows the team to really get a strong grasp on how the technology is working, as well as learn how contacts are engaging with it. Ultimately, it enables users to better train the assistant down the road and even identify gaps in content offerings.
“I always tell my clients that if you want to train a bot, you’re marketing manager should be responsible for going back and seeing how it is performing,” she said. “Don’t push it to IT or an analyst, because if you are monitoring, you are seeing the results and it will help you learn. In theory, IT can do the job. But marketing should take the initiative to actually monitor the AI thinks and the questions that come back [from the contact]. Those questions can easily be turned to content you can create in the future.”
