How to make sure chatbots provide good customer service

Datetime:2016-08-23 03:17:15          Topic: Natural Language Processing           Share

I’m sure you’ve heard the saying, “If I’d asked my customers what they wanted, they’d have said a faster horse.”

While I don’t believe Henry Ford actually said that, I also don’t agree with those that think that “faster horses” is not a helpful answer.

When people say they want faster horses, they are letting you know what the real problem is. They are just not very direct and clear about it. Yeah, sure, it might seem they are asking for an incremental change on the surface. But, if we keep asking “why” to the same people like a five-year-old , we will peel layers of this onion-like problem and discover that speed is the real problem here. They are saying they want to commute faster. Most people don’t really care whether the service providers use faster horses or better vehicles. They just want to get to their work and homes faster.

It’s the emotion and motivation behind the ask that matters.

A while ago, I wrote a piece titled “ Chatbots won’t fix your customer service .” I have gotten into some interesting conversations (and arguments) on that topic.

I wrote:

The user experience while dealing with companies could be much better even without chatbots. And, it could be much worse even with chatbots. It all boils down to what the intent is.

When people say “I hate talking to customer support at company XYZ,” they do not mean they dislike talking to a human or talking on the phone per se. They are implying that they do not like the amount of time and energy that goes into the entire experience for getting a question answered or a task done.

It’s a problem of speed and convenience, not of humans or phones. So thinking that chatbots will just fix that broken experience is the wrong way of looking at conversational products. This mindset is as wrong as thinking bots are the new apps.

Furthermore, most of us have gone through the experience of calling customer service and ultimately ending up on the call with an automated voice. We are already using bots in the loop. And guess what? We hate those monotonous, automated voices. Most of the time, it fails, annoys us, and wastes our time. After spending quite some time holding on that call, we finally end up talking to a real, live customer service representative. And that’s a relief for the most part.

After going through this poorly thought-out experience a couple times, we start to associate calling with a bad experience. This leads us to think that having chatbots on messaging apps will fix the issue because we are removing the calling component from of the equation. We are identifying the right problems (slowness and inconvenience) but are identifying the wrong cause .

Bots and phone calls are merely tools. They do not define what customer support and brand evangelism is all about. They merely help on acting on our intents. A lot of it is about not being a jerk online , as Veronica Belmont puts it . Brands like Product Hunt , Slack , Buffer , and Virgin America have already nailed customer service . They did not need bots to help them with that. Will they be using NLP-powered bots in the future? Most probably. But that’s just a tool that will help them act on their intents better in a cheap, scalable way. Customer service is a culture thing, not a technology thing. Technology under the hood amplifies the culture.

I’m not against usage of chatbots for this problem. I believe asynchronous, one-on-one messaging channels with brands will be great. I am aware of how it would empower smaller companies to provide great user experience at a fraction of a cost . I also know how big a marketing channel chatbots could become.

Currently, most chatbot products can be categorized into two types:

  • scripted, decision tree-based bots
  • natural language processing (NLP) or machine learning (ML)-based bots

ML-based bots require copious amounts of the right data sets to really provide a seamless experience. For NLP bots, the underlying tech just isn’t there yet. Tech giants like Apple and Amazon are working on NLP and ML-based bots. But even Siri and Alexa can handle only simple, structured requests. This will change for sure, but that will take time. The capabilities of scripted bots are limited to the product designer’s ability to imagine all the possible scenarios or paths a user might take and the response iterations in the databases. The failure rate is high in scripted bots because humans tend to anthropomorphize things and ask silly, unpredictable, and out-of-domain questions. The failure rate shoots up even higher when humans use slang and niche-specific terminologies.

Most bot products are scripted as of now. Having textual versions of the decision trees used for phone conversations does not really solve the problem. Well, it does if you are optimizing for the wrong problem.

I do think the future of customer service is going to be all about NLP and ML-powered bots. Until we get there, though, I think we need to take two steps to solve the problem of speed and convenience.

  1. Use scripted bots. But reduce as many steps as possible to get to the spot where a person can actually get their queries answered or problems fixed. This will become simpler once the underlying platforms have mature features like account linking, user profiles, comprehensive data access, security protocols, and personalized experience in place.
  2. Have humans in the loop to reduce the failure rate. If the chatbot fails, the very premise of building and using chatbots as customer service channels gets defeated.




About List