March 2019
Text by France Baril

Image: © metamorworks/

France Baril, owner of Architextus, is a DITA/XML documentation architect who helps organizations analyze their content and processes, select tools, learn about DITA and XML, manage the change process, and develop supporting material (from DTDs or schemas to XSL transformations).




Do we need to pump the brakes on chatbots?

In the last decade, chat has become the most popular platform for customer and technical support. Support is a huge cost for business. With the hype around Artificial Intelligence and deep learning, thereís a lot of excitement about the possibility of replacing live support with chatbots. But a deeper look shows the technology has a long way to go.

Technical communicators, especially those who work with structured content or other tools that support multi-channel publishing, can play a big role in providing high-quality content for chatbots to draw answers from. But before we jump headlong into the wonderful world of chatbots, it's important to take a step back. We need to understand two aspects of this technology in order to proceed successfully: Why people love live chat, and how chatbot support fits into their experience as a whole.


Live chat is well-loved

The following stats show clearly that live chat is growing in popularity and adoption:


  • Chat adoption rates have risen from 38 percent in 2009 to 58 percent in 2014, which is a compound annual growth rate of 8.8 percent over that time. (Forrester)
  • 53 percent of customers would prefer to use online chat before calling a company for support. (Harris Research)
  • Chat has become the leading contact source within the online environment, with 42 percent of customers using chat versus email (23 percent) or other social media forums (16 percent). (J.D. Power)
  • 62 percent of customers expect live chat to be available on mobile devices and, if available, 82 percent would use it. (Moxie Software)
  • 63 percent of millennials prefer to have their basic customer support questions answered by chat versus traditional channels. (Software Advice)


Why business is excited about chatbots

Businesses have been responding to the popularity of live chat by including it as a channel for customer service and technical support Ė but at no small cost.

Contact centers (once known as call centers) represent huge costs. The idea that live support could be replaced with chatbots has business salivating over cost savings. Analysis firm Juniper Research estimates the technology could save business US$8 billion by 2022.

Advances in Artificial Intelligence and deep learning are spurring interest in the development of chatbots. But in the excitement of new technology and the benefit to business, we seem to have overlooked the user experience.

A closer look at what users like about live chat will help make sure we donít break whatís working in the rush to replace live chat with bots.


What people like about live chat

There are many reasons people prefer chat over email and phone. As we task bots with taking on more of the workload, we need to consult these reasons:


  • 79% Questions answered immediately
  • 51% Can multitask
  • 46% The most efficient method
  • 29% In control of conversation
  • 29% Better information than email
  • 22% Don't like talking on the phone
  • 21% Can chat from work

Source: eConsultancy 2013

When chatbots work, itís because they deliver on those customer preferences. But they donít always work as planned.


Chatbot fails

Because chatbot technology is still in its early phases, thereís not a lot of data to look at. It is possible, however, to look at real-life experiences to see whether bots are delivering on all the things users love about live chat. Letís take a look at how chatbots have worked for me and others in the last year.


Scenario 1: The bossy bot that shuffles you along

Having recently moved from Canada to France, I needed a new mobile phone. I went online to buy phone and internet services from Red by SFR.

After hitting an obstacle with payment options in self-service, I clicked on the chat icon and was connected to Red Bot. Red Bot told me it was here to answer my questions. Great! But then Red Bot proceeded to ask me yes or no questions.



Bot: Hi, I am RED bot. Let me guide you and answer your questions.

Bot: Do you want to order a RED box?

Me: Yes.


Next, Red Bot redirected me to a live rep by asking me to enter my phone number.


Bot: Very good, I will put you in contact with one of our representatives.

Title after picture: Your Team RED representative.

Text: Enter your phone number so we can call you starting now. Your number:

Placeholder in field: Enter your number here.


This was a problem because I didnít yet have a French phone number. After all, this was the reason why I was trying to get phone service in the first place! I tried my Canadian number, which they couldnít call.

Next, I had to go out and buy a prepaid card from another provider so I could speak with a customer rep.

I entered the phone number and waited for ten minutes.

Finally, the rep came on the line and asked if I wanted to buy their services. Yes, I said, but I have a few questions about payment options. With a tone of annoyance, he told me that heís just here to take orders, not answer questions.

While the repís rudeness wasnít the fault of the bot, the interaction underlined a clear failure to align channels.

Needless to say, I did not buy my phone from Red by SFR.

Here is a summary of the failures I experienced that could have been prevented in a live chat:


  • Donít like (or canít) talk on the phone but redirected there anyway
  • My question is not answered immediately (or ever)
  • Not in control of the conversation


Scenario 2: The bot thatís just another smaller, less powerful search engine

If your customers wanted to find answers from a search engine, they would have started there! In fact, they probably did.

Hereís a bot attempting to help someone buy clothes. The chat-like interface is really just a thin imitation of a faceted search Ė a filtering down of options until the desired target is reached.

Credit: Image from


The problem is the chatbot doesnít have the capability or real estate of Amazon. Does the customer want dresses, tops, or bottoms? Uh, what if sheís looking for a hat? Is that considered a top?

Itís also forcing the customer to use filters that might not be important to her. What if she doesnít care about price?

The meta issue in this example is that the chatbot is not actually chatting. Its search is disguised as chat. This trick comes up a lot. Businesses know that users like chat and so offer the interface for functions that arenít chat. This bait-and-switch only adds to usersí frustration. If youíre going to use a talk bubble, make sure you can talk.

Summary of failures compared to a live chat offer:


  • Not in control of the conversation
  • Question not answered immediately


Scenario 3: The bot that needs technical support

Humans, unlike bots, donít glitch. Where does the customer turn for help when the chatbot isnít operating as designed?

In the example below, using a service in France, the chatbot is answering half in French, half in English, with my phone preference set to Spanish. ŅQue?

On top of that, every message is repeated, and variable data isnít filled.


Larger boxes: Service unavailable. Try again later or contact customer support at 353.

Placeholder: Write an SMS from Lycamobile.

This experience, like the previous example, is also only chat-like. The user is not allowed to ask questions. In fact, they have to get the codes for questions from the support web site.

Summary of failures compared to a live chat offer:

  • Question not answered immediately
  • Inefficient
  • Not in control of the conversation


Scenario 4: The bot that only gets in the way

After getting a new phone number, I needed to update my bank profile. First, I logged onto the bankís website hoping that I could do it myself. In the global user experience, this is self-service customer support. From the user perspective, this is level 1 support.

No luck. The field for the phone number was grayed out (disabled).

Next, I looked for answers in the help section of their website. I was now at level 2 customer support. I used the search and found some instructions that told me to go to the profile page I had just visited. My problem still wasnít resolved.

I went on to the chat Ė level 3. I got a bot that directed me to the same article I had already read. The bot asked if this answered my question. No, it did not.

The bot directed me to a form for email support. To me, thatís attempt number 4 for support.

I received an automated email saying that someone will answer my question in 24 to 48 hours, but in case it was helpful, it suggested that I read that article AGAIN. The content architect in me smiled at the fact that this business nailed content reuse, yet I still didnít have my answer.

After some back and forth through email with a real person, they finally changed the phone number for me, after validating my business registration. For security reasons, the phone numbers of business accounts can only be changed by a customer representative. Mine was an exceptional problem that couldnít be solved through information already available.

Summary of failures compared to a live chat offer:

  • Question not answered immediately
  • Information was not better than email
  • Inefficient


Chatbot successes

Weíve discussed some of the glorious ways chatbots in customer service and technical support fail. But thatís not to say they are a total loss. The chatbot potential is real and, in time, they may be able to replicate the benefits of live chat.

But for the time being, their success is limited to three distinct areas:

  • Entertainment and simple questions
  • Upsell
  • Information gathering before connecting to a human


Entertainment and simple questions

Chatbots are well-suited to serve as a vehicle for entertainment and providing simple benefits to users. A good example of this was a bot that is sadly no longer with us called Poncho the Weather Cat.

Poncho provided a simple benefit: telling users the weather Ė but with a dash of humor and sassy personality. ("I feel good about the weather. It isnít perfect, but neither is my homemade beer.")

Poncho worked because the service was reliably deliverable. Thereís lots of data about the weather that the bot could draw from. And people seemed to prefer it to a regular weather app on their phone because of the entertainment factor. They got the weather and mild diversion.



Chatbots show strong potential for sales. We know upselling clients with closely related products already works in other channels. Unlike with customer support, upselling doesn't require the bot to answer the customerís specific question. Instead, it can draw from a wider net and make many offers that might be appealing.


Information collection

Chatbots can save time and money as information collectors before the user is connected to a human. In these cases, the bot gathers the kind of information thatís needed for the service to be completed, but that doesnít require human interaction.

Melody by Baidu is a chatbot that helps doctors diagnose medical conditions. Before the patient speaks to a doctor, Melody asks them simple questions such as their age and what kind of symptoms theyíre having. In other contexts, this might be done by a triage nurse.

Itís a type of work chatbots handle well because the goals and scope are simple.


Why do chatbots fail when it comes to support?

Clearly, chatbot technology isnít working in the customer service and technical support fields as well as it could. This is primarily due to two overarching errors in approach: Chatbots are only re-serving information thatís available in other channels. Second, development teams overestimate the maturity of the technology.


Re-serving content

Bots may not be necessary in many cases. If theyíre going to be added to the user story, they must add value, not just be an extra step between the user and their answer. Supporting many channels doesnít mean forcing users through all of them.

The bank in Scenario 4 successfully reused content, but their lack of tracking of the overall user experience made it worse than if they had had fewer channels. In this case, the best solution isnít even providing support content. When I logged in, the system should have been able to see that the account was for a business, and I should have had access to the email form from a button or other control beside the grayed-out phone number field. Help is where people go after they have experienced a first failure.

Redirecting users to other, less preferred channels, as Red Bot did, is a related form of this error. If the bot canít answer the userís specific question, then you might do better not to offer it as a channel.


Overestimating the maturity and capacity of the technology

Implementing real conversation is hard. Thatís why in three out of the four fail scenarios, the bot didnít even allow questions.

Even with the bankís bot that was actually able to chat, at least in the sense that it allowed questions, it was only able to solve a problem that had already been solved, where the solution had been stored in the available data. The bot canít guess at solutions. It can only ask questions that lead it to the answer it has already been provided with.

At the same time, developers and businesses are underestimating the expectations users have of chatbots. By the time a user gets to chatbot support, theyíre expecting a customized resolution to their problem. To give them generalized help is insufficient.



Live chat works because the human interaction allows users to feel in control of the conversation and to receive an immediate and tailored answer. Often, users also get an answer or support that wasn't available in the existing self-service material.

Bots, if they have been trained with the information available in your system, are just another channel. They canít bring users the feeling that they will finally get someone to help with the operation they couldnít perform on their own.

Before you fall for the hype, it's important to consider the role chatbots will play within the global user experience. Are they going to be one more obstacle between users and their answers, or are they really going to support them?

Bots are expensive to develop, train, and deploy. They require new training every time a product or update is released, and they need to learn to decode questions in all the languages that the business supports. Businesses need to carefully weigh the value a chatbot brings against the significant investment of time and money.


Summary: Letís not create the next phone tree

If youíre a certain age, you might remember when phone support worked. You experienced a problem you couldnít solve through the manual or the website and picked up the phone. You got a person right away who listened to your problem and, most of the time, solved it.

Today, business managers and technology experts have figured out ways to automate phone support. Oh, the money they can save! Meanwhile, they have broken what was working. Phone support became the living hell it is now Ė a maze of "press 1ís" that more often than not leads to a dead end. Thatís one of the reasons why people prefer live chat!

As we go forward in the development and design of chatbots, letís consider this cautionary tale. Rushing to automate a system while ignoring what works about it will only lead to its eventual abandonment. Then weíll find ourselves chasing the next big thing all over again.