Most of today’s biggest companies have begun experimenting with chatbots, and many others are poised to follow suit. A recent survey found 56% of service leaders were actively looking for ways to integrate artificial intelligence, including chatbots. This surge in interest reflects the significant benefits this technology can offer: bots can save human agents time by providing quick and accurate answers to routine queries. They can also serve as an efficient air-traffic controller of sorts, by gathering initial information from customers (who they are, why they are reaching out) then routing them intelligently to the right representative. The best bots do this in a way that provides an excellent experience and enhances the customer’s impression of a brand by either answering their question quickly, or routing them to the person who can.
But not all bot deployments go so smoothly in reality. We’ve all heard of major bot fails like Microsoft’s Tay, which the company shut down within a day after the bot “learned” to spout racist statements. More common are mundane annoyances, like bots that fail to understand simple commands (see CNN example) provide unhelpful responses, make extraneous small talk or just don’t understand the customer well enough to actually be helpful. So why is it so difficult for companies to get bots right?
Are you interested in receiving more in-depth content on deploying chatbots in enterprises? Subscribe below to receive relevant updates.
Common Roadblocks to Successful Chatbots
Obstacles at these four junctures frequently trip up companies on the path to launching effective bots:
1. Identifying the right use cases.
Most companies can easily come up with a list of the 20 most common issues their customers face. But just because a use case generates a high volume of requests, that doesn’t always mean a bot should take it on. Many service leaders still aren’t sure how to determine which topics are the best fit for bots, and which should be handled by humans. Generally speaking, bots have a greater chance of failing when dealing with topics that are emotionally charged or require long, complicated back-and-forth interactions. Instead, it’s a better idea to automate straightforward topics that can be resolved quickly like replacing a credit card or checking status on an order.
Getting use cases wrong can have a huge negative impact on customers and your brand. Because the stakes are so high, companies would be wise to draw on expert guidance to determine which topics to assign a bot, and start with very limited use cases out of the gate, then build over time.
2. Designing flawless conversations.
Some companies treat bots purely as a cost-saving and automation tool. They fail to grasp that designing an elegant conversation that customers will find satisfying is just as important, and that starts with considering the way chatbots presents themselves to customers. I’ve heard of companies who name their lead generation chatbot “The Sales Bot.” How many customers would enjoy talking to that guy? When considering how chatbots communicate, companies should prioritize the customer’s needs rather than their own.
The way conversations flow depends on the use case case and brand identity. A company selling a millennial entertainment product will want the conversation to be a fun and modern. A healthcare chatbot, on the other hand, shouldn’t crack jokes given how potentially serious the topic at hand could be, which would obviously offend customers and undermine the brand. Failing to get conversations right means customers will stop interacting with your bot, and be left with a negative experience of your brand. Leaders who want to achieve adoption and impact with their bots should be testing and monitoring conversations with real rigor and vigilance to ensure they aren’t having an adverse effect.
3. Securely deploying data to train smart bots.
Like other AI tools, chatbots require lots of data in order to learn how to handle requests effectively. Training intelligent bots means exposing them to a multitude of sample conversations on the topics you want them to address, as well as clean sentences that express what customers ask and how to respond. The challenge is that data on past interactions with customers often contains sensitive personal information, including names, contact details and credit card numbers. So companies need to develop methods of segmenting out personally identifiable data from the aggregate data, so that the aggregate data can then be used to train the bots.
Many service leaders don’t have a secure technology solution they can trust to segment out personally identifiable data, then export the aggregate data that’s required to build a bot. Even if they succeed at this step, companies face challenges with finding a way to automate bot training. Many assume they need to hire expensive data scientists to get bots off the ground and end up indefinitely postponing projects due to budget concerns.
4. Deciding whether the channel or the chatbot comes first.
Most brands today recognize that they need to offer service where customers are, whether that’s a traditional phone call or interactions on Facebook Messenger, Twitter, web chat or SMS. And many assume that they can simply begin offering customer service on these channels and add bots simultaneously. Unfortunately it doesn’t work that way. Companies need to first establish themselves on the channel so they can determine what answers actually work, and amass enough data that can then be used to train the bot to behave appropriately and perform effectively.
Bots are Just Part of the Team
At the end of the day, there’s no getting around it. Companies that want to scale to meet customer demands for real-time, personalized service must figure out how to succeed with chatbots. Getting started means setting realistic expectations — AI-powered bots aren’t going to solve all your problems immediately.
Service leaders should think of chatbots as agents in the contact center: they have to teach them how to behave, how to represent the brand and how to provide customers with a good experience. They need to be integrated into the overall service workflow, able to answer common questions directly and reroute to human agents when necessary rather than being an isolated, external entity. Companies who are just starting to experiment with chatbots should prioritize quality over quantity. It’s better to provide a good experience to a small number of customers on a limited set of topics than a bad experience to many.
And remember, bots aren’t a silver bullet, but they can absolutely help service organizations get to the next level when they put customer satisfaction first.
Enjoy this article? Sign up for more updates on chatbots.
We’ll let you know when we release more educational content on deploying chatbots in the business setting.