A Compelling Guide On Chatbot Testing

Having a little chat with a bot is routine stuff these days. Chatbots learned to speak a human language. They can answer your questions in a verbal or written form, as fast as your friend can. Are you used to talk to chatbots? We, consumers, know that this AI-based consultant will give us a quick response to almost any request. What else do you know about these smart guys? 

Do you know what benefits chatbots provide to businesses? Have you heard of what way a chatbot passes while it will turn out from an idea to a modern digital solution? In this post, we would like to share our knowledge about chatbot testing. It is one of the most critical phases of chatbot development, ensuring the viability and high performance of the final product. 

5 Hottest Chatbot Applications Market Trends

Chatbot applications market is growing rapidly due to a great number of advantages these helpers offer to business. Virtual assistants are available 24x7. Unlike consultants-humans, bots respond to customer inquiries around the clock. Surely, this feature increases customer satisfaction with the service. Another advantage is the cost savings. You implement a chatbot, and there is no need to handle many customer support representatives. 

What’s new on the chatbot market? Below are the hottest chatbot applications market trends that might interest you.

In 2020, Google introduced a new chatbot Meena trained to use 40 billion words.

What is interesting about Meena? It is an open-domain platform allowing users to type and ask questions from any platform. Meena uses a neural conversational model. It can learn to answer according to a context. Meena chatbot is able to predict the next word that will likely be said by a user. Google says that Meena has a good grasp of conversation and ensures that chats are more natural. 

“Hi! I’m Quincy”. A new healthcare chatbot just came on the market to improve patient engagement.

Qliqsoft introduced its Quincy aimed to enhance patient engagement. Hospitals can now automate their interaction with patients by sending notifications and messages at a pre-scheduled time to patients via Quincy. The bot also chats with patients about their medication plans and answers any questions related to treatment. 

COVID-19 has caused a significant growth of new chatbot applications in 2020, and the numbers are expected to increase.

With the coronavirus outbreak, chatbots become popular within the healthcare industry by providing remote support to patients and across all other sectors. Due to global lockdown for a few months in a row, many organizations implemented chatbots. While employees stayed at home, chatbots provided 24x7 customer support to millions of brands, financial organizations, retail stores, etc. 

Voice chatbots are trending.

Voice-activated chatbots are becoming more popular than text-based chatbots. However, not all businesses could benefit from this type of chatbot. A voice assistant is a solution suitable for big companies with no budget restrictions and the target audience having easy access to speakers. 

The highest chatbots growth is seen in the Asia-Pacific region, some chatbots are available in 120 languages.

According to Mordor Intelligence, India and China are countries that experience high growth of chatbots. For example, Yellow Messenger, an Indian conversational AI platform for sales automation and customer engagement has already raised 20 million dollars from investors and is available in 120 languages. 

As you can see, the market is developing and bringing new values to businesses. Business Insider says that it will gain over 9 billion dollars by 2024. Thus, we may say that this vertical is expected to become highly competitive. Delivering excellent quality and performance of chatbots will help chatbot developers and providers reach a competitive edge. Further are some useful tips on how to do that. 

Why Should You Test Chatbot Applications?

As mentioned above, a first-class chatbot with a perfect set of capabilities will make its developers stand out from the crowd. How to develop an ideal software solution? In our experience, quality assurance and testing are the most critical stages to consider within the development process. 

The biggest pitfall of using a chatbot for your business is its possible failures. Among the most common problems that can frustrate your customers and affect your business efficiency and reputation are the following:

  • Broken scripts and crashes (see the picture below the list)
  • Some chatbots are not taught to recognize the confusion
  • Impersonal communication
  • Chatbots build without previous research on targeted users
  • Chatbots that know too much personal information like your date of birth

Image source: Chatbots Magazine

As a business or product owner, you surely want to avoid all those errors as they influence company reputation. You want to provide the highest user experience level, but that’s impossible without testing your software. As a user, you also want to have all your requests answered in a human-like polite manner with a good understanding of a conversational context. 

To avoid failures, chatbots should be tested before going live. In the next sections, you will learn how to test chatbots and what metrics to use for chatbot testing. 

A Simple Guide on How to Test a Chatbot

A chatbot is usually nothing more than a conversational flow. Its answers are being built according to pre-defined if/then patterns. It would seem that there’s nothing special in chatbot testing. And that is true as chatbot testing is much alike with another software testing. However, there are some peculiarities every product owner, developer, and test engineer should know. 

Unlike other software products, chatbots tend to fail more often. The reasons for frequent failures lay in the conversational nature of chatbots. As a rule, we expect a human-like talk with a bot forgetting that our language and manner of speaking/writing are very nuanced. Considering this feature of chatbots, testers use a quite similar chatbot testing approach. Below we share this approach, including the features of the chatbot to be tested.

Conversation flow

You never know what your interlocutor will say next. However, provided a conversational context, you may predict the next words. So does a chatbot. However, as we already mentioned, the human language is full of nuances like slang, double meaning, non-native speakers, humor, etc. Giving clear and natural responses is a primary goal of a chatbot, that’s why a tester should check the following things while testing a conversational flow:

  • Does a chatbot clearly understand the questions?
  • Does it always give instant responses to these questions?
  • Are the answers relevant to the given questions?
  • Should a user ask a series of questions until he gets the answer?
  • Does a chatbot engage the user to continue the conversation?

Business-specific questions

Once you tested the conversational flow, proceed to test how the chatbot is answering business-specific questions. Today, chatbots are used mostly by banks, retailers, hospitals, service businesses. Each industry has its specific terminology, notions, nuances. So, a tester should have a list of domain-specific questions to check if the chatbot is able to answer those. 

Confusion handling

Confusion may arise if a user enters some expression with double meaning or an unknown word for a chatbot. The latter should be taught to answer in such a situation. The tester’s goal is to check if the chatbot can handle misunderstandings, exceptional conversational scenarios, and unusual patterns. This capability of a chatbot is showing how “emotionally intelligent” it is. 

Speed and accuracy

How quickly the chatbot gives response matters as people want to get answers to their requests instantly. Another critical thing to test is accuracy. Test engineers should calculate the number of times when the bot gave correct answers.

Chatbot Testing Framework: 5 Top Tools to Use and Key Metrics for Chatbot Testing

To streamline chatbot testing, test engineers may use a wide range of open-source and proprietary tools like below. 

Chatbottest is an open-source tool providing a list of questions for chatbot testing. This tool offers seven key metrics for chatbot testing:

  • Intelligence - Is a chatbot intelligent enough to understand the context?
  • Error management - Is a chatbot good at troubleshooting?
  • Navigation - Is the conversation flow well-structured? Is it easy for the user to find the necessary information in the discussion?
  • Answering - Are chatbot’s responses relevant to the user’s questions?
  • Understanding - What kind of language items (humor, idioms, slang) does a chatbot understand?
  • Onboarding - Is it easy to understand how to use a chatbot from the very start of the conversation?
  • Personality - Does a chatbot have the name, voice, tone?

Chatbottest offers its bot called Alma as a Chrome extension to ask 33 questions. If you want to access a broader set of questions (120), you can check the Github. 

Botium is an open-source tool that offers its BotiumScript that can be taught to interact with your bot. In fact, your bot is tested by Botium’s bot for failures, errors, accuracy. It can text, talk, listen, click — do this just like humans do. This solution is definitely an excellent way to save time and costs for testing. 

Botium tests conversational flows and has modules for voice testing and non-functional testing like load and performance testing.

Dimon is a proprietary solution for chatbot testing automation. The provider says that the Dimon tool can decrease the testing time from hours to minutes. It generates various conversational scenarios, notifies you in case of any trouble with your bot, and provides APIs to test a bot like a real user.

Qbox is an NLP training data optimization platform used by dozens of companies to improve their bots. Qbox offers two options for chatbot testing: standard test for NLP classifiers and “Bring your own test set,” which uses pre-defined utterances.

Qbox has a five-step procedure to make your chatbot better. First off, you should download your NLP provider’s training data into Qbox. Then, run the test. Next, you get problem training data areas identified and can modify the data accordingly. To validate the changed data, you should go back to testing. 

Zypnos is a cloud-based solution for chatbot regression testing. It allows testers to rerun tests, which means you could save time and effort. 

Possible Chatbot Testing Scenarios Every Tester Should Know

Below is a list of possible chatbot testing scenarios applicable and adjustable to a wide range of industries, business domains, and technologies. 

  1. A chatbot should be loaded with a website it is built for.
  2. A user should clearly see and/or hear how a chatbot is loaded (pop-up, sound, etc.).
  3. If expected, it should greet a user according to the user’s timezone.
  4. If named, it should display its name.
  5. A chatbot should ask a user his/her name to use it during the conversation.
  6. If required, a chatbot must ask any predefined user’s contact details like email or phone number.
  7. When saluting a person, it should work well, recognizing male and female users.
  8. A chatbot should be trained to see typical spelling mistakes.
  9. A chatbot should recognize numbers and currencies.
  10. It should verify the programmed formats for a contact number, date number, time number. 
  11. A chatbot should be taught to deal with confusion arising from various intricacies.
  12. If required, it should be able to redirect the user to a contact person who can provide further assistance.
  13. It should perform well, provided the user pasted some copied text or an image into his message.
  14. A chatbot should store a conversation’s history and send it to the given repository if programmed.
  15. It should work well, provided that many users are asking questions at the same moment.

These scenarios can be used as a checklist for chatbot testing.

Feel free to reach our team out for consulting on your project. We deliver a professional chatbot testing strategy and plan for projects of any complexity and scope. 

Get started today
Grzegorz Kłos
Co-Founder
office@apphawks.com
Grzegorz Kłos - Apphawks Co-founder
© 2020 Apphawks. All rights Reserved.