How To Avoid A #ChatbotFail

While brands have scrambled to launch Facebook Messenger chatbots since the social media behemoth opened up the channel for development last year, the early results haven’t been particularly promising. Facebook is seeing a 70% failure rate among those 35,000 or so bots when it comes to understanding user requests. To combat this poor performance, Facebook is making some changes to Messenger, including adding a persistent menu that will allow users to choose from a number of requests or statements instead of using natural language and risking stumping the bot entirely.

There’s no question that AI will play a huge role in the future of retail, but in these early days of chatbots and virtual assistants, how do you reap the benefits while avoiding the pitfalls of this emerging technology? We caught up with Linc engineer Alessandro Sanchez to talk about the potential weaknesses in current chatbots and how smart brands are creating a chatbot experience that beats the odds and delivers great service. He says brands that want to see chatbot success should focus on three best practices.

Define the use case

Facebook is struggling with its own conversational agent, M, Alessandro says because M is an all-purpose AI, without defined parameters. “Not all chatbots are created equal. The most successful chatbots are going to be ones that operate in specific domains, where it feels natural to interact with them,” he says. The post-purchase experience is an excellent example of where chatbots excel, Alessandro notes, because shoppers have a limited range of specific intentions (finding out when their order shipped, getting assistance with a return) that a bot can be programmed to identify and respond to. He points to order tracking as one area in which Linc’s clients have seen great success with bots:

“Tracking itself is an easy, step-by-step process, where the majority of shoppers simply ping the bot every once in a while for a status update. The most popular button is our ‘refresh’ button, where a shopper can have an incredibly easy and natural interaction to stay updated on their package outside of the push notifications they already receive. For example, 90% of an apparel customer’s shoppers don’t ever require human assistance because of this feature. When a shopper is not satisfied with just a status update, they can ask a range of natural language questions that we’ve anticipated and can readily handle due to our natural language understanding capabilities. Our bot can field questions like Where is my package?, When is it arriving?, How do I return my item?, etc.”

The message is clear — focus on having your bot do one thing (i.e., deliver post-purchase customer service) and do it exceptionally well. Specialization is a step towards success.

Provide context to learn and grow

It seems like an obvious point, but the more data a bot has about a shopper, the richer and more relevant it’s able to make an interaction with that shopper. “We use a range of technology to detect the intent of a shopper and make heavy use of the circumstances surrounding that particular shopper and their request at that moment, in order to determine intent. We have data and context about each shopper, when they ordered the item, when it’s coming, what they’ve purchased before, their return history, when they attempt to interact with customer service, etc. We know who they are and their reasoning. Our bot will constantly update the shopper profile and change behavior based on it,” says Alessandro.

In order to be truly useful to customers, your bot needs to have the capacity to learn from its interactions with them, both individually and at scale. A successful bot is able to draw on the customer data you’ve already collected and analyzed and also add to it and adapt its behavior based on this new data. If you haven’t figured out how chatbots and conversational channels integrate into your quest for a single customer view, you aren’t setting yourself up for success.

Make the human-to-bot hand-off seamless

“The problem with most chatbots is if they don’t understand what you’re saying, they don’t do a good job of transferring your issue to a human agent to be solved,” says Alessandro. This results in customers getting frustrated with the bot and either looking for another channel through which to engage with your brand or, worst case scenario, deciding not to bother engaging further. “At the present stage of the technology, we’re looking at some sort of hybrid experience between humans and chatbots. A chatbot should be smart enough to know when it doesn’t know and to escalate to a human,” Alessandro notes. Linc’s own bot has a mechanism to allow a human agent to take over the conversation, which is triggered by specific cues that indicate a shopper is not getting what they need — the bot is unable to understand a human’s messages, the shopper indicates a response is not helpful through a CTA attached to most of the responses, or it does detect the shopper’s intent, but knows a human agent can best solve the problem. “The key is to be sure to forward the inquiry as soon as we detect the shopper would be better served by a human agent, there’s avoiding any frustration and avoiding the common stigma associated with dysfunctional and ‘useless” bots,’ says Alessandro.

At their current stage of development, chatbots are not a 1:1 replacement for human customer service professionals, so relying on them to handle interactions that require a nuanced, personal touch is simply unrealistic. A successful bot should, by specific embedded cues or questions, understand when the hand-off to a human is necessary and expedite that process before the shopper on the other end of the interaction becomes frustrated or impatient. The best bots help the agent solve the customer’s inquiry more quickly by providing additional insights with the hand-off, as Linc’s platform does.

What takeaway would Alessandro offer to brands that are wary about reports of the less-than-stellar success rate of the current generation of chatbots and eager to sidestep the major pitfalls of this new technology?

“Take the articles with a grain of salt because they are talking about the mass of bots on the platform. Success depends on the domain-focus and the problems the chatbot is trying to solve. We’re seeing a lot of potential and success in post-purchase engagement and customer service because it’s a natural interaction to be handled within a conversational channel. For example, if 60% of post-purchase questions are about order tracking and returns and I build a chatbot that only handles tracking and returns, it’s already handling the majority of customer questions. That’s a huge win. It’s about the nature of the questions and post-purchase inquiries are easy to tackle when you have the expertise to do so and the data,” he says.

In short, ‘keep it specific and build it smart’ is the chatbot mantra you should be embracing in order to see ROI and success this year.

To learn more about chatbots and how to leverage them in 2017 to reduce customer service costs, and drive additional revenue through upsells, check out The Definitive Guide to Conversational Commerce. It offers a thoroughly-researched look into these emerging channels and answers the most common questions and points of confusion.