Create a Chatbot that Lets Customers Track Packages

1. Go to SAP Conversational AI and log in.

2. Click New Bot and select the following:

   Field Name	Value
   1. What do you want your chatbot to do?	Perform Actions
   2. Predefined Skills	Greetings
   3. Create your bot	Name: ShippingBot Description: A bot to help shipping customers track and price packages Language: English
   4. Data Policy	Non-personal and Non-vulnerable
   5. Bot Visibility	Public

   

3. Click Create a Bot.

Our bot must be able to extract the parcel number from within the conversation. Such data extracted from a conversation is defined as entities.

1. Under the Train tab, go the subtab Entities.

   

2. Click Create an Entity.

3. Call the entity PARCEL-NUMBER, specify it as a Free entity, and click Create.

   

   Free entities: You specify the type of data it is, and you teach the bot how to identity a value for that entity. You specify the form of the data but not the actual acceptable values.

   Restricted entity: You specify a limited list of acceptable values.

4. Click on PARCEL-NUMBER.

   

5. Enter the following values for the List of values.

   The values give the bot the valid format for a tracking number.

   CAREFUL: Enter the values in the field where it says Add a new entity value. After entering the value, press Enter.

   • 1Z2220060291994175
   • 1Z2220060292002190
   • 1Z2220060292634221
   • 1Z2220060292690189
   • 1ZISDE016691676846
   • 1ZY756Y60380128446
   

   We have also supplied the numbers as a CSV file. Instead of manually entering the numbers, you can download the file, then click Import a CSV File, select the file, and click Upload.

6. Click Train (at the top).

   

   Generally, the chatbot will be trained automatically, but clicking Train will force the chatbot to better understand the different possible parcel numbers.

Built-in Entities

SAP Conversational AI comes with built-in entities (called gold entities) for common data, like location, dates, and money values. For more information, see List of Gold Entities.

Customers will have to indicate whether they accept or reject certain requests from the bot, so the bot needs to be able to recognize when a customer means “Yes” and when they mean “No”.

You do not have to create these yourself. Instead, you can fork them from others who have already created them.

Forking is simply copying, and does not create a relationship between your bot and the source of what you forked. You can fork intents, entities, skills or an entire bot. For more information, see Forking Bots, Skills, Intents, and Entities.

1. In a different browser tab, go to cai-adoption / ups-bot / intents / @yes.

   

2. Just above the list of values – NOT at the top of the page (that button is to fork the entire project) – click Fork.

   Select your project, and click Fork.

   

3. Go back to your project and click the Train tab (refresh the page if necessary), and you now see the @yes intent.

   

   Click into it, and see all its expressions, that is, all the ways someone could say “Yes”.

4. Test the intent by clicking the Test flap on the right of the screen, and entering expressions at the bottom.

   Enter Sure and I think so and OK to see that the bot understands the intent. Feel free to add additional expressions.

   

   You can see that the bot detected the intent.

5. Repeat the above steps for the @No intent, located at cai-adoption / ups-bot / intents / @no .

Customers will be indicating that they want information on a parcel, so you need to create an intent for that.

Again, you do not have to create this intent yourself, but you can fork an existing intent.

1. In a different browser tab, go to cai-adoption / ups-bot / intents / @track-parcel.

2. Just above the list of values – NOT at the top of the page (that button is to fork the entire project) – click Fork.

   

   Select your project, and click Fork.

3. Go back to your project and click the Train tab (refresh the page if necessary), and you now see the @track-parcel intent.

   Click into it, and see all its expressions, that is, all the ways someone could indicate they want to track a package.

   Notice that some of the expressions contain a tracking number and some don’t.

4. Test the intent by clicking the Test flap on the right of the screen, and entering expressions at the bottom.

   Try the following:

   • Where is my package
   • Track 1ZISDE016694068891
   

   In the second phrase, you can see that the bot not only detected the intent, but also detected the parcel number.

   Feel free to add additional expressions.

After you’ve determined that the user wants to track a package, you have to build a skill so the bot knows what to do in response to this intent.

1. On the Build tab, click Create skill.

   Call the skill track-parcel, and specify that it is a Business skill, then click Create Skill.

   

2. Open the skill for editing by clicking on its name.

3. Add a trigger so that this skill is triggered when the @track-parcel intent is detected.

   • In the Triggers tab, click in the field next to If.
   • Select @track-parcel.
   • Click Save.
   

Add the action if the user says No.

1. Go to the Actions tab.

2. Click Add New Message Group.

   

3. Click Add Condition, and then in the field next to If, select @no, and click Save.

4. Click Send Message, and then:

   • Select Text.
   • Enter No problem. Let me know anytime if you want to track a package. as the message.
   • Click Save.

5. Click Update Conversation | Edit Memory.

   

   In the Unset memory field, enter no (the value is case-sensitive), and then click Save.

   

   Why unset No ?

   This will make it so the skill no longer fulfilled the requirement of having a Yes or No answer. So now if the user changes their mind, they can enter track package and the conversation will already have the tracking number stored and will only ask for confirmation.

If the customer says Yes, then you will make an API call and retrieve the information to display. You will code this in the next tutorial.

If the Test panel is open, close it.

Click Chat With Your Bot (bottom right) and start a conversation with your bottom. Use this tracking number: 1Z12345E6205277936.

In the example below, notice the bot detects the intent to track a package, but notices the tracking number is missing, and asks for it. Finally, it asks for confirmation, using the tracking number by retrieving it from memory.

After the customer says No, this reply is unset in memory, so if the user again says track package, the bot simply asks for confirmation because the latest tracking number is still in memory.

If the customer says Yes instead, for now the bot shows No Reply because you have not defined an action in this case.

No Reply vs. Fallback Actions

When a bot has detected an intent and that intent triggers a skill, but that skill does not have an action that returns a message, then No Reply is shown.

If a bot cannot figure out the intent of the user, or no skill has been triggered by the detected intent, then the fallback skill is triggered. When you created your bot, a fallback skill was created automatically for you with a default message.