This is the last part of our four-part blog series on chatbots. In our previous posts:

You would think that just about wraps it up. Wrong – the work has just begun.

Like employees, chatbots need continuous training. In fact – at least in the early stages – they need even more training, because they are basically machines with limited learning abilities compared to people.

Fortunately, chatbots learn quickly and never forget anything they learn. In addition, bots based on the most advanced technologies can broaden their own skills, leaving people with the task of monitoring incoming messages to ensure that replies are reliable.

Chatbot whisperers play a key role

Chatbot trainers are commonly referred to as bot whisperers. They are ultimately responsible for ensuring that chatbots add the value for which they were ‘hired’. Without interaction by bot whisperers, the development of chatbots would be hopelessly stunted.

Bot whisperers can be thought of as process developers: they follow how chatbots do their daily work, identify areas in need of development and train the chatbots to do their job better and handle completely new issues.

A bot whisperer’s work may be strongly proactive, but ‘active reaction’ is also needed.

For example, trends or new developments in general can lead to chatbots needing to learn new things quickly. Bot whisperers must also consider issues from the end user’s – the customer’s – perspective and realise the importance of the customer experience, while maintaining the appropriate level of humour.

Bot whisperers work with the following concepts:

  • Intent: Intent forms the basis of a chatbot’s information, describing the intentions, needs and goals of end users. It describes what the user wants to happen during the chat. For example, in the case of an online store, the intent may be to order a product, track an order’s status, gain instructions on changing an order, return a product, enquire about delivery terms, ask for additional service details, and so on. The more ‘intents’ the chatbot has been trained to recognize, the better it is able to serve end-users – at least in theory.
  • Expression: Expression is one way of describing intent. The same intent or need can typically be expressed or asked about in several ways. The more expressions of intent the chatbot is trained to recognise, the better it will serve end-users – in practice.

However, there needs to be enough variation in the expressions used: for example, repeating the expression “Can I travel with my pet to destination X” 100 times is not enough for recognition of the intent to ‘travel with a pet on an aeroplane’. Expressions such as “Can I fly with my cat to Y”, “I want to travel with my dog from Helsinki to Berlin. What should I do?” and “I’m booking a flight and my two chihuahuas will board with me. What should I take into account?” should be covered.

There is no maximum number of expressions, but around a 100 for each intent is a good rule of thumb – at least for the most frequently occurring ‘intents’. Training the chatbot to recognize the most common intents enables coverage of large numbers of expressions in a short time.

  • Reply: Replies refer to how intents are responded to. At its simplest, a reply can be a single message (if the question concerns opening hours, for example), but replies often trigger an entire dialogue flow.
  • Dialog flow: Dialog flows encompass the entire recognised intent and the related chat, what we might call the service process. For example, identifying an intent to return a product triggers a dialog flow, which guides the user through the product return process.
  • Annotate: ‘Annotate’ refers to training which pairs end-users’ messages with the appropriate intents. Each message paired with the right intent extends the chatbot’s knowledge of the expressions associated with the intent. This enables the chatbot to understand what the end user wants and respond correctly to messages.
The typical working day of a bot whisperer

The key issue for bot whisperers is to understand the substance — that is, the subject area — so as to enable the chatbot to create added value for end users. For example, if the topic is B2B customer service, the chatbot whisperer should sit in on meetings with the customer service team to understand which issues they are grappling with at any given time.

Properly simplified, the typical working day of a bot whisperer tends to involve the following issues

  1. The whisperer understands and recognises the key intents that the chatbot needs to master.
  2. In terms of quantity and quality, the whisperer accumulates a comprehensive number of expressions for each intent, to ensure reliable recognition.
  3. The whisperer defines and describes a reply and dialog flow which serves the end user.
  4. The whisperer analyses how well the chatbot is performing and develops its know-how based on end-user messages.

Sounds easy, doesn’t it? Not quite – such work can sometimes feel like looking for a needle in a haystack, especially if the chatbot is a relative newcomer. A bot whisperer may well have to wade through hundreds of messages in a chat history, and ponder which issues they should be connected to.

Groundwork takes a lot of time, but makes future work easier when done well: knowing what the chatbot’s expertise consists of makes it easier to identify possible blind spots.

Of course, a bot whisperer works in active collaboration with other experts. For example, defining and describing service processes requires the expertise and support of the parties running the services. If a chatbot is used on a customer service channel, the customer service team lead is the bot whisperer’s new best friend!

Chatbot training must also ensure that the quality of customer experiences remains high and the chatbot communicates in line with the company’s brand. In most cases, a ‘persona’ is defined for the chatbot, based on which it serves end users and that acts as a guiding light for communications generated by the bot.

However, such a persona cannot answer all questions, so the bot whisperer must consider issues from the customer’s viewpoint and place themselves in the end-user’s shoes.

The text is an English version of the last part of Sofigate’s ‘Chatbots in Business: Guide to Success’ blog series. Other parts of the series have appeared on Sofigate’s website and on the LinkedIn page.

?

This blog post has been made in collaboration with Sofigate’s partner ultimate.ai.?

?

About the authors

Jussi Vuokko is the CTO of Sofigate’s Business Automation business. He has helped organizations of all sizes leading their services and automating service production using modern technologies for more than two decades. If you want to chat with Jussi about the topic, you can reach him by email jussi.vuokko@sofigate.com, on Twitter @jussivuokko or on LinkedIn: linkedin.com/in/jussivuokko.

Pekka Stenlund is a Customer Success Manager at ultimate.ai. He helps customer service professionals understand how Artificial Intelligence solutions can be leveraged to enhance customer service and create a better customer experience. You can discuss these topics with Pekka by email – pekka.stenlund@ultimate.ai? – or on LinkedIn: linkedin.com/in/pekka-stenlund-b3703976.?

?

About ultimate.ai

ultimate.ai is one of Europe’s leading artificial intelligence companies focusing on customer service automation. With our help, companies serve their customers on the customers’ terms – regardless of time, place and language. We are a team of researchers, developers and customer service professionals with a passion for developing the work of millions of customer service representatives around the world and helping companies make customer service their competitive advantage.

Read these next

yy6029/又粗又大的机巴好爽/亚洲欧美综合在线天堂/元卿凌宇文皓免费阅读