For or against conversational bots for customer service?

That was the topic of a panel I attended a few weeks ago in Montreal. Being a vocal proponent of bots and especially bots integrated to the contact center, you can easily guess what my answer was.

But it doesn’t matter, what I think. You can’t go back in time. Bots & intelligents assistants, whatever you call them, are here to stay. To me, being for or against chatbots for customer service is like being for or against the web in 1996. Yes, it was ugly. It was slow (remember those 56k dial-up modems?). You couldn’t do much except consult very basic web pages. There were no standards, both in terms of protocols or architectures, or even in terms of UI design. But look at where we are today. We’ve come such a loooooong way! The web is so pervasive now. It just works. It’s the same for bots. We’re at the beginning of a new era.

One of the arguments I got from the other panelist was that most chatbots do not perform well, they are not well designed. Am I surprised of that? Not at all! In fact, I cannot agree more! Most chatbots are ill-designed, they are brittle, you can break them easily. But is it because someone buys a cheap pair of shoes that breaks easily that everybody must be against shoes? Nonsense. This tells more about the kind of industry we are trying to build, I think. Building great chatbots is not easy. It takes more than 10 minutes to build one, unlike what so many companies would like you to believe. I think the problem lies more with the fact that many mobile/web developers started developing chatbots like they were doing web sites or mobile apps. The skills required to build great conversational UX are totally different and you need to hire the right team.

Is NLP technology ready?

In a follow-up article by Les Affaires, it was said that the panelists (me included) believed the technology was not there yet. Well, that was not exactly what was meant. At least not what I meant. Due to advances in deep learning in recent years, NLP technologies have become much better at tasks like detecting intents and extracting information (entities) from text.

But unless you are just doing a very simple proof-of-concept, or some simple button-based chatbot (is it really conversational, then?), the fact remains that to use these technologies effectively, some expertise and experience are required. Even for a PoC, you will need a rigorous approach to ensure your project meets the expected performance. Otherwise, how can you assess the effectiveness of your NLP technology provider? You need tools to benchmark your NLP and your dialogue as it evolves. Now imagine doing a full-fledge task-oriented self-service chatbot development project…

NLP is far from being the only obstacle to building an effective CUI. A chatbot needs to properly manage the conversation, understand the customers’ needs, and find the relevant solutions to the customers’ requests. But we’re not there yet, that’s why chatbots are far from replacing human agents today and having close collaboration between humans and chatbots is key.

Lack of integration

I think the real challenge for customer service departments is whether the technology can be easily integrated into the rest of the contact center operations. Most chatbot platforms make the integration with the rest of the contact center rather difficult, if not impossible.

What kind of integration to a contact center should you expect? First, when your customer wants a human agent to join the conversation, the whole history of the conversation should be pushed to the agent’s desktop. Or a least a summary of the relevant parts of the conversation so the customer does not have to repeat herself. You also want the hand-off to the agent to use the same routing mechanism as the other channels (voice, chat, email, etc.).

And the converse is also true. If your customer starts the conversation on the chatbot or an intelligent assistant and later calls your contact center, the agent should be able to dig the whole journey to better understand the customer’s needs. If a piece of the conversation is missing, this will inevitably irritate your customer.

If you can’t have a seamless integration between your chatbot technology and your contact center platform, you may end up with two categories of agents: those who use the contact center agent desktop, and those who use the chatbot platform’s tools. The latter won’t be able to log as contact center agents, they will only be seen by the chatbot platform.

This is a situation you want to avoid at all costs because you will need to consolidate two sets of historic reports, possibly use two different workforce management solutions, integrate two systems to your CRM, etc. In other words, you will operate in complete silos. How can you expect to deliver an exceptional customer experience in this context?

The point here is that chatbot projects should be considered as part of a complete omnichannel strategy, not just as yet another disconnected channel, just for the sake of having a chatbot.

For or against, then?

Of course, yes! Chatbots and intelligent assistants (IAs) have the opportunity to provide real benefits, both for the customer and the organization:

  • They provide advanced self-service and automation capabilities on text-based channels already used by the customers, without requiring the install of a new app or changing habits.
  • Customers can express themselves in their own words, instead of having to decipher the organization’s verbiage.
  • Customers have access to the whole history of the conversation, for their own records.
  • Intelligent assistants are also very different from simple search widgets where you only get the documents matching the words you entered. IAs can ask disambiguation questions whenever required, and provide answers personalized to the customer. Imagine having access to a FAQ, but with answers adapted to you!
  • Like with normal chat sessions, the organization has access to the raw intents/requests. It can learn from requests that cannot be processed by the system and further train/update the chatbot to answer them in the future. All this without requiring an app update on the customer side.
  • A chatbot can follow-up on the conversation at a later time, on the same channel where it started. Say you chatted with a human agent on Facebook Messenger, a survey can be sent to you in the next 24 hours to know your satisfaction level, also on Messenger. Why send you an email?

Moreover, chatbots can also alleviate human agents from the most mundane and repetitive tasks, letting them work on more complex and challenging problems. The ones requiring problem-solving skills, empathy, etc. And doing so, your agents will feel more useful to the organization, retention will be higher, which will ultimately save costs.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.