Get started – Preview

Get started – onboarding process

Getting started – Preview

Preview and test your assistant

Get an introduction to the Preview tab, best practices for testing your assistant prior to deployment, and how to share a very realistic mockup website

Much like proofreading an assignment before submitting it for grading, it's important to preview the conversational flows that your assistant is ready to handle before deploying it on your website, customer service call center, or corporate messaging application. While it's possible to troubleshoot your actions as you build them, it's highly recommended that conversation builders test, preview — and even try to "break" — newly created assistants as a cohesive deployment within the dedicated Preview tab.

Why would you want to "break" an assistant that you are testing? Because a user's first impression of your specific assistant determines how likely they are to return to it (or try to bypass it). Finding faults in your assistant's conversational flow before deployment lets you refine it before that first impression is made.

This tutorial provides an easy-to-follow introduction to the Preview tab, best practices for testing your assistant before deployment, and how to share a realistic mockup website with your colleagues so that they can test, preview, and potentially break your newly created assistant, too.

To see the process, you can watch the following short video.

Prerequisites

To follow this tutorial, you need:

• An IBM Cloud account
• Access to any watsonx Assistant instance on a Lite, Plus Trial, Plus, or Enterprise plan
• Access to any assistant
• An understanding of the watsonx Assistant user interface

Estimated time

It should take you approximately 10 minutes to complete the tutorial.

Steps

1. For the sake of brevity, this tutorial uses an assistant that contains two simple actions. Let's see whether we can find a way to improve them before deployment by using the Preview page. Navigate to the Preview page.

   

2. Preview always tests the changes that you've made to your assistant in the draft environment, so that when you eventually deploy your assistant, your team has a place to test updates before publishing them.

   The Preview page renders your assistant in a web chat window to emulate how it would behave after it's deployed. By default, the web chat's home screen feature provides a greeting with some example options that can be selected before any of your actions are triggered, a stark difference in comparison to simply previewing actions within the Actions page. Should you plan on deploying your assistant by using web chat, there is a way to improve it to make a better first impression. I cover how to update the web chat's customization's later in this learning path.

   For now, I want to preview the actions for improvement. Begin interacting with your assistant to see what phrases trigger your action as expected while taking note of any that do not trigger the expected action.

   

3. While previewing the conversational flows, note the areas for improvement. When you reach the end of a conversation or are ready to test again, click the Restart icon at the upper right of the web chat window to start over. Following are some examples:

   a. Test out natural conversational flows that your customers might have with the assistant.

   

   Saying "Thank you" is triggering an action that it shouldn't, so you need an action for responding to different ways of thanking the assistant. An even better practice is creating a subaction for asking if the user needs any further assistance when the end of the original action is completed.

   b. Test out similar, yet incorrect, prompts to the user examples that your actions are trained on to ensure that they are not accidentally triggered.

   

   The action "Where can I find watsonx Assistant tutorials?" is trained on a single intent matching its name, so it's great to see that asking about a similarly named product, Watson Discovery, properly routes to the fallback action.

   c. Intentionally provide wrong answers when provided a list of options to ensure graceful fallback.

   

   d. Intentionally try to change topics in the middle of an action repeatedly.

   

   In doing so, you learn that the assistant is still set up to transfer to live agents by default. In a scenario where there are no actual live agents, this would not be an ideal experience for a customer, so it's great that it was caught during previewing and testing.

4. Now that you've done some testing and previewing of the assistant on your own, let's prepare a link to share with your colleagues to see whether they can find other areas to improve. But first, you want to make the preview look more realistic by putting a sample website image behind the assistant. To do that, click Change background.

   

5. Choose the method for setting the sample website (I use the default Enter URL method), and click Continue.

   

6. Enter any public website URL, and click Continue.

   

7. After it loads, your sample website appears behind the Preview web chat render. Now that your Preview is ready to share, click Copy link to share, and paste the link into a new browser window to test it.

   

8. If the preview looks ready to share, then it is. Pass this link to trusted colleagues to collect feedback. Some things to consider are:

   • Do the colors of the web chat match the website's design?
   • Does the tone of the assistant's greeting match the tone of your branding?

   • Are there other elements on the page that might conflict with the web chat window?

     

9. Encourage your colleagues to interact with the assistant as much as possible. All of the conversation logs are sent to the Analytics tab, where you can quickly see which topics the assistant missed during Preview testing, as well as which topics had successful actions completed.

   

Summary

Now that you understand how your assistant appears after it's deployed, and how to share that preview with your colleagues, you might be wondering about a few topics that were brought up in this tutorial, including:

• Understanding draft versus live environments
• Customizing your web chat look and feel
• Using analytics to improve the assistant

All of these topics will be covered next in this learning path. However, before moving on, take some time to build out your assistant with your own actions, either by using templates or creating them from scratch. Preview and test them out with your colleagues repeatedly. This way, your assistant can be ready to go soon after you learn how to customize, deploy, and maintain it.

Any questions? Reach out to Alexander Dfouni.

IBM watsonx Assistant is an AI-powered virtual agent that provides customers with fast, consistent, and accurate answers across any messaging platform, application, device, or channel. Using AI and natural language processing, watsonx Assistant learns from customer conversations, improving its ability to resolve issues the first time while removing the frustration of long wait times, tedious searches, and unhelpful chatbots.

With watsonx Assistant, you can build conversational interfaces into any application, device, or channel. Most virtual assistants try to mimic human interactions, but watsonx Assistant knows when to search for an answer from a knowledge base, when to ask for clarity, and when to direct someone to a human. Like a human personal assistant, the assistant you build helps your customers perform tasks and answer questions. To accomplish this, you define actions for the assistant.

This tutorial provides an easy-to-follow guide on how to sign up for an IBM Cloud Account, provision a watsonx Assistant instance, launch watsonx Assistant, and configure the settings for your first Assistant.

To see the process, you can watch the following short video.

Prerequisites

To follow this tutorial, you must have an IBM Cloud account.

Estimated time

It should take you approximately 10 minutes to complete the tutorial.

Steps

1. Go to the IBM Cloud website, and click Create IBM Cloud account in the upper-right corner of the page.

   

2. Follow the prompts to create a new IBM Cloud account. This requires that you provide your email address, create a password, select a country and region, and enter your credit card information.

   

3. If you've already purchased a subscription or have been provided a feature code, register using the code provided by selecting Register with a code.

   

4. After your account is created and verified, log in to the IBM Cloud dashboard.

   

5. From the dashboard, click Create resource in the upper-right corner of the page.

   

6. In the IBM Cloud Catalog section, type watsonx Assistant into the search bar and select watsonx Assistant from the search results.

   

7. On the Catalog page, select a location, select a pricing plan, and read and agree to the license agreements. Finally, click Create.

   

8. Wait for the instance to be created, and then click Launch watsonx Assistant to open the watsonx Assistant user interface.

   

9. You can now begin using watsonx Assistant to build and manage your chatbot. Create your first assistant by giving it an internal nickname (not visible to customers) (a), then clicking Next (b).

   

10. On the Personalize your assistant tab, answer each question to the best of your ability. Anything that you enter can be changed later, so don't worry about anything being permanent. On the final screen, click Create.

    

Congratulations! Watsonx Assistant is ready for you to build your first assistant.

Now that you have knowledge on incorporating actions into your assistant by using templates, determining the sequence of steps in an action, and by using debug mode to fix any issues, you can begin developing your own actions. Remember, actions are considered the foundation of the build process in IBM watsonx Assistant. They represent the issues or tasks that your customers would like your assistant to address, and encompass the entire interaction between a customer and your assistant for a particular inquiry or request.

This tutorial details an approachable introduction to the action editor within watsonx Assistant by building a basic action that provides a unique response depending on one of two options selected.

If this is your first time building an assistant in watsonx Assistant, following these steps helps build familiarity with some of the most common behaviors that conversation builders repeat in watsonx Assistant when building virtual assistants. Try following these steps for your assistant's specific use case to result in a small action that you can build upon in the future.

Note that this tutorial is a part of a learning path for beginners. More in-depth information on building actions with templates can be found on the watsonx Assistant documentation pages.

To see the process, you can watch the following short video.

Prerequisites

To follow this tutorial, you need:

• An IBM Cloud account
• Access to any watsonx Assistant instance on a Lite, Plus Trial, Plus, or Enterprise plan
• Access to any assistant
• An understanding of the watsonx Assistant user interface
• A basic understanding of how the sequence of steps is determined within an action through the condition builder (covered in Build watsonx Assistant conversation content with templates)

Estimated time

It should take you approximately 15 minutes to complete the tutorial.

Steps

1. Within watsonx Assistant, navigate to the actions (a) Created by you tab (b), and click New action + (c).

   

2. Select Start from scratch to build an action for your own use case.

   

3. The New action window appears. Enter any phrase that your customer might say to start the interaction (a). Click Save (b).

   

   Note: At first, only a single instance of a typical user input is required to initiate the action. Additional examples of user input can be included later.

4. The action editor appears with the first step selected. Enter a response for your assistant to reply with when the action starts (a). Click Save (b), then click Preview (c).

   

5. When your assistant is finished training, enter the phrase that the customer starts with into the preview chat to begin the action (a). The assistant responds with the response that you entered for the first step.

   

   This conversation logic is an example of a simple question and answer flow. This is the quickest way to train your assistant on new topic coverage. However, there are a few best practices to follow to ensure that your assistant provides a polished customer experience. The first of which is to practice adding clarification steps at the start of the action. Click New step + (b).

6. After the new step appears on the left side of the screen, drag it above the first step. The blank Step 2 becomes Step 1, while Step 1 and its responses move to Step 2.

   

7. Within the Assistant says of the newly created Step 1, ask a clarifying question that helps the user to be better qualified for the response in Step 2 (a). Then, select Define customer response (b), and select Options (c).

   

8. In the Edit response window, enter the options that your user can select from (a, b). Click Apply (c), then close the Edit response window (d).

   

9. Back in the action editor, notice how Step 1 now has the options added. Let's help watsonx Assistant understand which step to go to when one of the options is selected. Click Step 2 (a) in the Conversation steps list because Step 2 has the appropriate response for one of the options. Then, update Step 2 to be taken with conditions (b, c). Next, ensure that the condition is set to the appropriate option from Step 1 (d). Finally, update the And then section to End the action (e). Now that Step 2 is set up for one of the response options from Step 1, add a new step for the other option (f).

   

10. Drag the new step down to become Step 3 (a). Set Step 3 to be taken with conditions (b, c). Ensure that the condition is set to the other option from Step 1 (d). Then, enter an appropriate response for the other option within the Assistant says section (e). Finally, update the And then section to End the action (f).

    

11. Open and refresh (a) the Preview to test out your assistant's new action (b). Try out both options.

    

    Tip: Try providing information about one of the options in the first prompt and watch watsonx Assistant skip to the right step without asking the clarifying question.

    

Summary

This example of creating an action without a template is only scratching the surface of the types of conversations that you can build by using actions. Aside from Options, there are many other types of customer responses available to explore. Within And then there are near-endless possibilities of where your assistant can be directed, from pointing to other actions or repeating previous steps, to finding an answer through Watson Discovery and NeuralSeek — or even connecting the user to a live agent.

There are many best practices to eventually consider when it comes to building efficient conversational flows, such as adding more user examples within the Customer starts with section and using session variables. As a beginner, continue learning about the rest of watsonx Assistant's functions before returning to begin honing a proficiency in actions.

Any questions? Reach out to Alexander Dfouni.

Categories

Artificial Intelligence

watsonx Assistant

Trials

Try watsonx.ai

Try watsonx.data

watsonx.governance

Try watsonx Assistant

Interested in generative AI?

Learn generative AI skills