The inbox is the central spot for all incoming and outgoing messages. Whatever happens between your bot and your service teams and your users will happen here.
The Inbox has - depending on user permissions - a separation into up to three tabs:
- The Inbox includes active conversations between users and service teams, such as dialogs that have been passed from the chatbot to a human.
- In All conversations you find all conversations that have been held with both human agents and the chatbot.
- In Annotation you find only those conversations that failed the interpretation of the chatbot and need re-training.
For teams that can log in and out through this (adjustable in the menu under Team > Agent Teams), it offers the ability to register in the system as present or absent. This allows you to control exactly when you are available and when you are not.
Handover sessions always appear in the Inbox tab. As a live agent with the appropriate permissions you can see the handover sessions assigned to you or your team.
All conversations are listed below each other. This list shows the most important information you need to classify the conversation.
- The avatar is the profile picture of the user. If there is none, it is formed from the initials.
- The name is taken from the bot - if there is none, a random name is generated.
- The emoji - either a bot (🤖), a service agent (🧑💻) or a user (👩/👨) - lets you know who sent the last message.
- Next to the emoji you will see a snippet of the last message.
- This is followed by the date and time of this message
- Another avatar shows who the conversation is assigned to. If it is green, you are responsible for the answer. If it is gray, it is assigned to another person.
- At the end, a timer shows how long the user has been waiting for a response. If it is blue, it means that everything is good and you answer fast enough. Yellow means that you should respond quickly before the user is no longer active. Red means that the user is already waiting too long. If you have too many conversations at the same time, and you can't answer them quickly enough, our system will notice and won't assign any more chats to you.
First of all, it looks exactly the same as in the Inbox, but with the difference that in All Conversations not only handover conversations are displayed, but also conversations with the chatbot.
Sometimes it can be a lot of conversations. Therefore you can segment the view by applying the filter parameters. For example, go to Add filter and set it to show only conversations with at least 5 user messages (minimum number of messages). In this case, only the conversations that match the definition will be displayed. Also note that only conversations within the specified retention time period are visible.
This view shows the entire history of the selected conversation as well as information about the user(s) and useful parameters from the dialog. In addition, previously created answer templates are also available.
The User Information shows various information about the user and the corresponding conversation. This includes user-specific parameters such as the name or email address, information about the history of the conversation (available by clicking Show all) and bot-specific parameters that can be set individually for each bot in the Creator. The values of the users' information can be edited by agents in the inbox, so that if the bot has entered a wrong value or has not yet entered it automatically, information can be corrected and completed. Simply hover one of the entries, click on the pencil and enter the correct value.
Below the user information is also the option to make notes. This allows you to easily keep a log of information that is important for other agents.
In the complete conversation you can see the chat between bot and user or agent and user. You can see all the specific information and use the features for agents like "Reply" or "Routing".
In the chat history on the left you can see the answers of the bot or the agent. Bot replies are in grey, agent replies are in green. Above the agent bubble you can also see the name of the agent who wrote the message.
Messages from users can be found on the right side of the history. These are blue and and have more info below the bubble on how the bot interpreted the input. The confidence tells how "confident" the bot was in the interpretation and which intent was chosen.
The button to the left of this information makes it possible to subsequently annotate messages that the bot thinks it has recognized. This allows that actually recognized intents can also be assigned differently later on.
If a conversation is not in handover, as an editor you can put it in handover mode at any time by going to Take over conversation. This will take you to the agent view and allows you to write with the user.
In handover mode, a text field will appear in which you can enter the answer for the users. In the simplest case, you just have to enter the text and click the Send button.
Other functions are B, that writes the text between the two asterisks in bold, the I for text in italic (one asterisk each before and after the text section) as well as adding links. To insert a link, type it into the popup window and the link will be inserted into the text field accordingly. If the displayed text should be customized, you just have to replace the text between the square brackets.
If you want to use your message for more than just one user, you can save it as a template. Simply select the option Add to answer templates. When you click Send now, this message will also be saved when it is sent.
Finally, you can use the End session option to release the bot for the conversation again.
The answer template view is only shown when the handover mode is enabled. Each answer template is a single entry on the right side next to the conversation view.
Adding a template to the chat can be done by clicking on it. This is then copied into the text field and can be customized before sending it.
By the way, templates can be found easily by keyword search in the search field.
The annotation view shows a different interface as it is designed for a specific scenario within bot conversations: unrecognized user input.
By opening the annotation view, you will first see all conversations with a message that need annotation. Click on the conversation to view the message. As soon as you click on it, you will see the last annotated message from the conversation. If you want to view the other messages to be annotated, you can go through all the messages using previous annotation and next annotation.
When you found an entry you want to annotate, you have the option to select either the whole message or just a part of it. Here is a comparison:
I need a robot
Hi Botty, I have a question I've wanted to ask you all along. Do you have robots, too?
In the first example, it is clear that the core meaning is the whole sentence. In the second example, marking the last sentence is sufficient to express the user's intention. Accordingly, only this part is marked by selection and confirmed with Enter. A multi-stage process then begins.
- Choosing the right game
- Select the appropriate intent
- Select entities (if applicable)
There are two things to keep in mind when doing so. First, if there is no suitable intent available, you can simply add an intent in Generic Games by clicking on "Add Intent". Then you need to give a name to the intent and pre-write a short answer and an intent with the title will be created in the Creator.
In step 3, note that entities only need to be defined in games that use content. Here is an example of a sentence with entities:
I need a robot with wifi for less than 2000 €
This sentence includes two entities: a feature (
wifi) and a price (
less than 2000 €). In the annotation, the game could be
RobotSearch with the intent
Find, plus two entities needed to be defined. Entity 1 could fill the slot
features+ (the attribute is a list, so it can have multiple entries) and entity 2 could fill the slot
price<. The bot is thus trained for these types of intents with a price and a feature.
While creating the annotation in the training data, as an annotator you can undo the annotation: just click undo in the small window. Wrong training data can always be edited under Automation > Training Data.
When you open the inbox, it is set for the live instance. However, if you want to see the conversations of another stage, you can set this using the stage filters.This ensures that you can control the quality of your bot on the right instance.
In addition to defining the stage, you can also filter the conversations. This can be done on the basis of user parameters (name, language, etc.), channel or assigned service agent.
For a user parameter you simply have to go to add filter and then select the corresponding parameter. Depending on the type of filter you can still decide how the filter definition should turn out. For example, you can choose whether you want to display a user with a name or all users except those with the defined name.
The channels filter is especially useful if you run your bot on multiple channels or even multiple channel types (Web, Facebook Messenger, Whatsapp, etc.). If the channel filter is applied, you will only see the conversations that take place or have taken place in this channel. Of course, you can easily remove the filter definition.
With the show filter functionality you can filter conversations based on the handover assignment. So you can see the conversations that have been assigned to you, your team or all users.