May 2017
By Alberto Ferreira

Image: Robot Romeo © Sandro Salomone/Softbank Robotics

Alberto Ferreira is a user experience researcher and globalization services expert with over ten years of hands-on experience in UX on the client and agency side with some of the biggest companies in the world, including Sony, BBC, and Mars. He is the author of Universal UX: Building Multicultural Experience, and writes and speaks frequently on topics ranging from Agile to persuasion design.




Rebooting with bots: The future of (assisted) technical documentation

In recent years, bots have come a long way and many organizations are successfully using them to assist customers. But how will they change the landscape of technical communication?

The fear of robots taking over the world, or at least our jobs, is an ongoing concern in the media. Business magnate Elon Musk recently suggested in a CNBC interview that robots can and will take away jobs from their human creators, and that governments will have to subsidize the unemployed population. In 2013, an Oxford University study set the number of U.S. jobs to be replaced by automated technology in the next decade or two at 47 percent.

This might seem a distant vision, but the fact is that Artificial Intelligence (AI) is already a reality. A number of industries actively use AI to complement their activities: Medical services use it to interpret medical images, reducing the error margin and biases in human diagnosis. Banks routinely use algorithms to detect fraudulent claims and operations, which are then analyzed by human counterparts. Travel companies use AI to predict fare trends and air control requirements. But can AI be leveraged for simple communication at a human level? And, more importantly, can we use it to satisfy our increasing hunger for information?

From human conversation to bots

Australia-based technology company FastBrick developed Hadrian X, a bricklaying solution that can deploy up to 1000 bricks in one hour without human intervention.

In Japan – a hotbed of robotic technology – over a hundred SoftBank Mobile stores use an endearing robot named Pepper to welcome customers. The robot is able to read facial expressions and engage in conversation, actively learning traits and responses with each interaction.

The use of AI to replace human contact has also spread to the online world, with conversational interfaces. These types of interfaces are termed "naturalistic", as they emulate the way that humans communicate most frequently: verbally. Direct verbal dialogs remain the easiest way to get intent or wishes across. But written conversation, transmitted at the speed of light via a growing number of platforms and services, is becoming more and more important in the age of instant messaging and social media. According to a Radicati Group study from 2015, there were over 3.2 billion instant messaging accounts worldwide, not including mobile messaging. The number of messages exchanged every day is certainly mind-boggling.

These figures lead to differing customer expectations when it comes to communicating with services: Consumers expect companies to be available and reachable through instant messaging – a changing paradigm that becomes evident when we look at the usage of social media products like Skype, Facebook, and WhatsApp. And that’s where the little "digital helpers" called bots come in: They provide chitchat and quick information regarding product updates, technical assistance, and support, acting as colorful characters in the product’s ecosystems.

Bots are also useful in attending to purchasing decisions, a method Asian businesses have already used for years. Ninety-two percent of luxury brands use WeChat for marketing in China, and luxury handbag giants like Prada and Versace are already preparing to sell their items on WhatsApp, making it possible for customers to buy products with only a few messages. For years, companies like Alibaba have successfully combined bots and humans in customer support.

These interfaces have also started to become ubiquitous in the West. Facebook Messenger has a canopy of bots integrated by default, ranging from pearls like MeditationBot – a personal coach for your zen needs, to Hipmunk – a travel and flight personal agent. Skype is following suit with a variety of help bots that provide anything from healthcare advice (Baymax) to an accurate description of your face (Your Face). These bots are only available in selected countries, but will soon expand to other territories.

Virtual assistance for real-world problems

Apart from these novelty applications, conversational bots called chatbots are experiencing broader adoption in various commercial applications. Their potential is still being untapped, and the future holds few restrictions, but the present already offers companies some alluring solutions. Chatbots are most commonly used in customer support. Automated online assistants can take on the roles of customer care representatives in order to respond directly to customer queries in a chat box.

Personal assistant bots like Amy,, and Kasisto Kai are also becoming more mature. They can automatically schedule meetings and manage personal schedules. With the rise of mobiles, the IoT, and messaging platforms, PA bots are becoming ubiquitous.

Figure 1: Pepper, the robot, greets customers with a literal plastic smile.
Source: © Jake Curtis/Softbank Robotics


Chatbots and PA bots are usually of two types: text-based or voice-based. Both types of assistants rely on natural language for communication, but use different means. They use complex natural language processing (NLP) engines designed to analyze the input and interpret it depending on its structure, tone, language, and context. This technology is not the exclusive property of industry tech giants. The consumer digital industry has already implemented multiple variations of it, in the form of completely artificial voice interfaces like Apple Siri in the Mac ecosystem and Microsoft Cortana in Windows.

These systems generally consist of a combination of voice recognition and speech-to-text layers: a single system recognizes the vocal patterns of the user and transcribes the words spoken into information for the interaction layer. Although commercially available NLP engines are on the rise, AI research is expensive and slow, and only a few companies are able to develop frameworks to use it. The biggest proponents in the field are arguably the usual suspects:

  • IBM, with its Watson engine, which encompasses NLP and artificial learning solutions in one cognitive technology platform. It includes products like Watson Virtual Agent for customer support chatbots, and Watson Explorer, which can analyze any type of data to find trends and relevant information for any customer query.
  • Google Natural Language API applies machine-learning capabilities by extracting relevant information in any type of text document, including sentiment analysis, to understand how customers feel about your company.
  • Microsoft's Language Understanding Intelligence Service is a component of Microsoft Cognitive Services, and can be integrated in the Bot Framework also offered by the company.
  • Facebook uses its framework, acquired in 2015, to augment its bot technology. The framework is freely available and allegedly used by over 65,000 developers.

Beyond these systems, AI engines like Amazon Alexa and also purport to enable access to several devices by voice command alone, and use sensors to react to your presence, tone of voice (or writing), the time of day, and other variables. Fancy telling your smart kettle to make a cup of tea while driving home? The Internet of Things (IoT) makes it possible. However, chatbots are usually much simpler than these large implementations and require much less technical expertise to set up

Figure 2: Kasisto Kai and Revolut are one of several new services in personalized banking, which rely on a text interface to fulfill simple user requests.


Assisted self-service customer support

Chatbots can be exclusively dedicated to a set of "skills", e.g. the service or resources that the assistant has access to. The assistant can have just one skill, such as interacting with a specific device in a simple bilateral relationship, or be "multi-skilled", meaning it can work with various devices in a complex environment. For example, personal banking assistants like Kasisto Kai can assist in simple banking operations like checking your balance or transferring money, whereas Amazon Alexa is the touchpoint to command numerous devices at will.

Companies can use chatbots to specialize in a specific area of customer care. Consider the volume of request calls coming to your company. Busy phone lines, messy case management, a 24-hour brand-bashing hotline. And consider the implications: According to Lee Resources International, for every complaint logged in, there are 26 angry customers that never reported back to the company. This is due to the fact that many customers feel that the time invested in calling is not worth the resolution. A big reason for this level of discontent is the response rate: According to an American Express Survey from 2011, 67 percent of customers gave up on a support call because they were put on hold – steeply decreasing the level of satisfaction with the service.

In the age of instant messaging, where live chat generates a higher level of satisfaction than any other form of customer support (73 percent according to Happyfoxchat), having an accessible chat box for quick inquiries and clarifications is a boon for providing an immediate low-friction channel. For simple matters and operations live chat is an appealing method to increase satisfaction and tackle most basic customer queries. Companies like Nuance provide full-blown chatbots ready for activation.

Creating your chatbot

The following steps will assist you in developing a bot to suit your needs:

1. Check your actual needs

Implementing a customer care bot is not an easy undertaking, technically. Embedding such systems in websites and apps has technical implications on the user experience and information architecture that you must take into account. It is important that your business strategy is aligned with the implementation of a bot. Bots hold a number of opportunities, but also some dead ends. Useful questions include:

  • What is the conversational interface meant to complement – customer support or technical documentation?
  • What support scenarios is it intended to support?
  • What is the back-up plan in case the request escalates? Or should the bot support information retrieval scenarios alone, typical in technical documentation?
  • What sort of information would thus be relevant and how should it be fetched and presented to the customer?

2. Define where the chatbot will be used

On which channels should this chatbot be presented? On the website (mobile version) or desktop, and on which pages? What about your social media channels such as WhatsApp, WeChat, Facebook Messenger, or Telegram? A good rule of thumb is to use the channels that play the biggest role in conversion. If bots are meant to support customer care, start making the bots available during after-work hours, after having an initial period in which it can overlap with your actual support hours to avoid potential support issues in case the chatbot malfunctions.

3. Train your bot

Conversational user interfaces have come a long way in recent years, but they still require constant maintenance and a dedicated training period. The good news is that very little coding experience is necessary in some cases, and conversations can be structured using standard flowcharts and UML or AIML models. You can use decision trees to explore interaction scenarios, e.g. by having the bot reply with a specific response if certain keywords are detected in the user’s input. Alternatively, you can use services like Twyla to train your chatbot for a specific industry or to respond with a specific tone of voice.

Figure 3: A decision tree can be as simple or as complex as the goal of the bot demands. It is infinitely customizable and provides a good tool for constant improvement.


4. Iterate the output

Even after being released, bots should continue to be perfected and improved. New user queries and more experience can lead to new use cases or requests. You can add or improve answers accordingly in order to continuously attend to advanced queries.

Are chatbots the new technical communicators?

While bot technology is usually not able to resolve complex matters – particularly those involving mediation and compromise – chatbots are able to respond meaningfully and even resolve most basic questions posed by customers.

Conversational interfaces are able to supply basic information for customer queries or give information in accordance with the frequent searches and FAQ. Bots can be programmed to reply to standard questions like "When are you open?" or "What is your warranty policy?" with direct links to pages or supply information directly during the interaction.

This makes particular sense for retail and e-commerce websites, where the search box is one of the first points of contact. Users can simply type in a question instead of searching for an answer, making the process more akin to an actual conversation and more comfortable for the user.

Bots can also assist in providing on-demand reference material. Questions like "Where is the specifications list for refrigerator Beraton XM-125?" can be interpreted accurately by most chatbot engines, and the relevant document (in PDF format or a webpage) can be fetched on demand. The more specific the use case, the more efficient chatbots can be. They are not a one-size-fits-all solution but their purpose is inherently fit for users who ask specific questions and want concrete solutions.

Despite the buzz that these systems are generating in the market, current applications of bots are actually more limited than they seem. They can support customers in-context on websites, assisting them through an e-commerce funnel or support journey, and be integrated in your company’s app in order to supply a chat box for customers to submit feedback or suggestions. They can be multilingual and permanently switched on. However, customers still require the adaptability and flexibility of a human with the creativity to improvise solutions and the rationale to break down complex problems. In addition, users are wary of new automated technology. As the uncanny valley hypothesis predicts, human-like robots can elicit feelings of revulsion. The technology behind virtual agents is improving every day, and they are here to stay, but mass job migration is still some way off.