The 5 Fundamentals of Bot Deployment
The chatbots making headlines in recent years have become famous for one thing—failing. There was Coca-Cola’s #MakeItHappy campaign, in which Gawker tricked the company’s bot into tweeting lines from Hitler’s Mein Kampf. Then came Tay, Microsoft’s infamous AI-powered Twitter bot that learned to spew racist comments in less than 24 hours. Equally unforgettable was InspiroBot, which was designed to generate inspirational quotes for posters and ended up producing such memorable messages as, “Cats are going to fill the world with pain” and “Try to tell yourself that you are horrible.”
This string of high-profile flops isn’t surprising. The technology behind conversational bots is still relatively new, and many companies have raced to innovate without necessarily thinking through possible flaws and loopholes. Despite the potential for AI-powered bots to go awry, they are here to stay. In a world where customer expectations are skyrocketing and companies are increasingly competing with each other based on how well they treat customers, automation and AI are necessary to help companies scale their efforts.
Why Companies Need Chatbots
In the past, companies didn’t want customers calling in. It cost them money and indicated there was a problem. Today, companies strive to increase their interactions with consumers, and customer experience serves as a key brand differentiator. More than half of consumers, and 65% of business buyers, say they would be likely to swap brands if a company didn’t personalize its communications to them. It’s easier than ever for consumers to get in touch, and they demand service experiences that are customized, consistent and immediate.
Chatbots are essential in allowing companies to meet the growing scale and expanding demands of customers. Deployed correctly, bots serve two main purposes:
- Intelligently serving up information (like order status, flight information) and articles that instantly fulfill customer requests and answer questions
- Collecting initial information from customers, then routing them to the most appropriate agent
Despite the benefits of chatbots, only 43% of service teams employ them today. Those figures are growing: Companies expect the use of service chatbots to increase by 23% between 2017 and 2019.
Five Principles of Successful Bot Deployment
If your company plans to launch or expand AI-powered chatbots, these five principles will help you drive real results and avoid becoming the next subject of embarrassing headlines:
1. Set Expectations
All stakeholders must understand chatbots aren’t a silver bullet and can’t outright replace human agents. Clearly define what you would like the bot to achieve, and articulate its capabilities both internally and externally. Since training chatbots is expensive and time-consuming, start with simple use-cases (like checking their reservation status or whether a flight is on time) before expanding to more complex topics (like the need to shift flight times). Be realistic about how quickly you can move: By definition, AI-powered assistants learn over time, so you can’t expect overnight success.
2. Be Transparent
Don’t try to pass off your bot as human. Making sure your customers understand when they’re interacting with a chatbot or a human will build trust and, if needed, give you leeway to work through kinks. And customers don’t mind: They are actually 1.7 times more likely to see service chatbots as desirable than not. But your bot should still have a personality. Think through the tone you want your bot to have (e.g., playful, conversational, serious) in light of your brand’s voice. You also want to speak to the bot as you would to a person when framing questions, so it can learn to handle queries in real-world vernacular.
3. Personalize the Experience
The more you personalize the chatbot experience, and the quicker, the better. Early wins will encourage buy-in from both customers and internal stakeholders. You can provide a personalized experience by knowing exactly what the customer is experiencing during the interaction (e.g. an error message, looking at products) and making sure the bot knows your customer’s history with your company (e.g. past issues, recent orders, returns). Having and acting on this information will allow you to demonstrate to the customer that you know and understand them.
4. Establish Fail Safes
Your bot will fail. Even if you’ve trained it well, it will encounter requests that it doesn’t understand or isn’t equipped to handle. Establish mechanisms that enable the bot to quickly recognize when it’s out of its depth and bring in a human agent to fulfill the customer’s request. Ensuring the bot can quickly escalate such cases will help you avoid putting customer satisfaction at risk.
5. Improve Over Time
One of the basic tenets of AI is that it improves over time as it learns from past experiences. Create a system that allows customers and agents to provide feedback about their interactions with bots and suggest how they can be improved. Ideally, users would be empowered to provide feedback in real time so that the bot can improve after every interaction.
Bots aren’t a panacea, and you should tread lightly if you’re in an industry where customer queries get emotional (e.g., insurance or healthcare) or where customer requests are highly varied (e.g. complex manufacturing). But for most companies, chatbots are a key component of a customer service strategy that can meet the demands of a customer base that is growing in both scale and expectations. If you deploy chatbots the smart way, chances are you’ll become known for impressive results, not dramatic flubs.
About the author: Clement Tussiot is a Senior Director of Product Management for Salesforce Service Cloud. Born in Paris, he now lives in San Francisco, where he works on tools for leveraging artificial intelligence and machine learning to improve customer service. Tussiot describes himself as a customer experience nerd, a robots nerd, and a craft beer nerd. He recommends connecting the first two, then relaxing with the third.
Bots Are Learning Their ABCs, But They’ve Got a Ways To Go
AI-Powered Bots Gearing Up to Serve You
 State of Service survey, Salesforce Research, September 2016. Data not represented in the “State of Service” report.
 State of the Connected Customer survey, Salesforce Research, June 2018. Data not represented in the “State of the Connected Customer” report.