Chatbots are a powerful new channel to increase sales and assist more customers. They are are also a minefield of mismatched expectations.

Inconsistencies in the customer experience hurt a brand’s ability to develop fruitful, long-term customer relationships. Within the field of customer communications, the bar is set high, and we’ve worked hard to gain consumers’ trust in instant, real-time chat interactions. The evolution of customer service, stemming from call centers and services disguised as chat, created an unfriendly, disappointing connotation that cast a dark cloud over the customer service experience.

Eventually we found a way to bring the friendly, personable chat experience to a business-to-consumer interaction. The results are in the data. Many consumers prefer chat over phone and email. According to a 2016 Forrester study called “The Future of Customer Service,” 53 percent of U.S. online adults are likely to abandon their online purchase if they can’t find a quick answer to their question; 73 percent say that valuing their time is the most important thing a company can do to provide them with good online customer service. They prefer chat because of the expectation they have for it — fast responses, a human interaction, and the answer or solution they need.

While chatbots seem like the natural progression of this chat experience we’ve come to know and love, I see hints of an interactive voice response (IVR) system shoehorned into a messaging app.

I was encouraged recently when I read Sebastian Krumhausen’s guidelines for building a chatbot. There are a lot of things to consider when building a fully functional, automated experience, and Sebastian did a nice job of blending requirements for both the builder and the consumer.

But the challenge of providing great customer service within the chatbot experience isn’t entirely technical — it’s finding the right mix of technology blended into the customer service organization to help customers where and when they need help.

In his article, Krumhausen talks about creating a “hybrid experience” for chatbots, and I want to focus on this because it’s the most vital part of advancing the chatbot experience, especially for customer support.

Within the customer interaction

There are certain elements of the visitor experience that should be present to ensure an optimal customer experience, as follows:

1. Bots should be transparent

The number one question we get on live chat is “Are you a real human?” It’s to a point where customers don’t (or can’t) believe a real human can be fast to respond, or actually available to talk.

It’s true that a customer’s trust goes up and frustration goes down when they get a quick response. As Krumhausen pointed out, “Saying ‘Sorry, I did not understand you’ is better than ignoring the user.”

But if there’s one thing a customer hates more than a slow response, it’s being lied to. A bot should admit that it’s a bot at the outset of a conversation, and not try to trick the customer into thinking it’s a real human. Transparency will be key if businesses want customers to trust bots.

2. Bots should offer an OPERATOR option

Have we not learned anything from years of infuriating IVR systems? The first thing we do when an IVR answers a support phone line is yell ‘OPERATOR’ over and over until we’re finally transferred to a real human being.

Look, I understand that chatbots will eventually be more intelligent than IVRs, but there is still a limit to natural language processing (NLP), and eventually a customer is going to reach that limit. It happened to one of our writers recently. We sent him out to buy pants using only chatbots and he got very close to completing the task, but the conversation fell apart when he tried to ask a question about the pant inseam.

If at that point he could have requested to speak with a human — or better yet, if the bot had recognized a human was needed — someone could have saved the sale! So give your users the option to opt-out of a chatbot conversation if they want.

3. Bots should treat each customer as a unique individual

Messaging apps might seem less personal than a phone call or an email, but the fact is some of your customers have intimate relationships over messaging. There is a level of trust that comes with letting a business message you on a personal mobile device.

Chatbots have the power to instantly tap into customer records to customize each conversation, or at least remember a customer’s name. The smarter chatbots get, the easier it will be for them to personalize each interaction and know order history, remember preferences, and even predict products they’ll be interested in next. If humans can do it over chat now, chatbots should definitely be able to accomplish this.

Within the customer support organization

Every chatbot interaction is an opportunity for your support team to learn more about your customers and their tendencies. So chatbots should also be designed to provide your team, and the bot itself, with data that will improve performance and customer satisfaction.

1. Record the conversation

The power of chat and messaging with customers is in the qualitative data it provides. The hundreds of thousands of conversation transcripts that are collected provide a unique look at how your customers communicate and what they really want when they’re on your site.

Oftentimes, though, it takes an operator command to send a conversation transcript to a CRM (customer relationship management) system or customer database.

As we move forward with chatbots, they should be trained to capture customer conversations based on certain key words and know how to assign them to the correct customer record. These conversations can be used for training chatbots, training other support operators, informing marketing collateral, and improving the overall customer support experience.

2. Allow for feedback

Chatbots shouldn’t be solely transactional — that is, completing a sale or task and moving on. They should offer the user a chance to rate their experience. This is a standard feature with almost all chat platforms, although again a survey is typically initiated by an operator command at the end of a conversation.

While messaging conversations don’t technically ‘end,’ a chatbot that is designed to augment the customer support experience should know, or learn, when to ask for customer feedback. This is data that admins can use to evaluate the bot’s performance and customers’ overall satisfaction after a bot interaction.

3. Transfer a conversation

As we mentioned above, chatbots aren’t going to get it 100 percent right every time. Eventually they reach the end of their processing capabilities and should know when it’s time to get a human involved.

When that time comes, it should be possible for the chatbot to do one of two things:

  1. Alert an online operator that it can’t answer a customer’s question and open the conversation for an operator to override the conversation. When it does this, the chatbot should hand off the conversation history, so the operator can hit the ground running.
  2. If no operators are online to transfer the customer to, the chatbot should attempt to end the conversation but promise the customer it has flagged the discussion for operator follow-up. It should also provide an idea of when a response can be expected.

We’re just scratching the surface with chatbots, and that’s exciting. But a shiny new technology shouldn’t mean a degraded customer experience, especially since we’ve all worked so hard to gain consumer confidence. With these considerations in mind, we can keep working toward a satisfactory hybrid experience that benefits both businesses and their customers.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.