This post was originally posted in ChatCreate’s blog.
If you’re running a support operation with thousands of chats and tickets, chances are you’ve probably looked into optimising things with a chatbot. Even with well maintained knowledge bases and self-service infrastructure, your agents still spend a significant amount of their time answering questions like “how can I get a refund?” or “I can’t log into my account”.
If you’re looking to setup a chatbot for your Zendesk support system, you’ll quickly realise that there’s not many options out there to choose from. You’ll want a solution that is able to understand what the customers are asking, either respond with an appropriate answer or quickly hand over the conversation to a human agent. Currently solution like that are mostly available to big companies with budgets to allocate tens or hundreds of thousands for a custom-developed bot setup and maintenance.
ChatCreate is here to offer the same level of service to support teams without having to spend a lot of time and money on setting up a bot and seeing whether this solution makes sense for your team.
With ChatCreate, any support manager using Zendesk Chat can setup a smart bot within minutes with 0 upfront costs. In this post we’ll cover step-by-step how to setup a Zendesk Chat (Formerly Zopim) Artificial Intelligence(AI) customer support bot for free.
The whole mission of ChatCreate is to provide a support conversation automation tool for customer support teams to help them automate repetitive requests and help human teams focus on more complex tickets.
- Your Zendesk Chat plan needs to be on Enterprise (Premium if you’re on one of the legacy plans). The reason for this is that Zendesk only makes their Chat Conversations API available to customers that are on the Enterprise plan. ChatCreate works through the Zendesk Chat Conversation API. This doesn’t mean that your whole Zendesk suite or other products need to be on the enterprise plan, just the Chat product.
- You need to have an agent setup in Zendesk Chat that will act purely as the chatbot. This agent needs to have “administrator” rights in order to be able to read and respond to chats. You can either setup a new agent or use one of the existing agents that is not actively used for responding to customers’ chats. This agent cannot be used by human agents for responding to chats. You can read how to setup a new agent here.
- A chatcreate account. It’s free to signup and start using. You can signup here
In order to setup the bot, there’s 2 things we’ll need to do:
- Connect your Zendesk Chat account to ChatCreate
- Setting up the automations and activating them.
Connecting your Zendesk Chat agent to ChatCreate takes about 2–3 minutes to do. You can read detailed instructions on how to do that here.
Once you’ve connected your Zendesk Chat account to ChatCreate, the fun begins. You can start setting up your own chat automations.
The “automate” tab is where you set up and manage all of the chat automations that ChatCreate does for you.
Once you’ve connected your Zendesk account, ChatCreate automatically analyses your Chat history and tries to detect automatically the recurring support issues that it can help you automate. They’re displayed in a row under the “Automate” tab.
The topics in the red box in the screenshot above are the ones that ChatCreate has detected automatically from your chat history. Now it’s up to you to decide which ones you want ChatCreate to take over. You can toggle through different topics with the arrows on the right and left. Please note that the topics are displayed as ChatCreate finds them.
When you click on a specific topic, a new window pops up. This is where you can automate a certain topic.
Let’s take a look at a simple example of “how are you” topic and automate it so the bot can respond correctly to it.
When you click on it, this is what open up.
If you check the red box, this is where ChatCreate displays all of the examples that it has found from your chat history that it thinks mean a similar thing. These need to be reviewed and either approved or disapproved. If you approve for example “how are you?” then the next time someone asks something that looks very similar or means the same thing, then ChatCreate knows that it should give an answer that you will put it under “Answer”.
So we’ll approve everything that means essentially “how are you?”. So for example “how you doing?” means the same thing so we’ll approve that too. However “you there?” or “can you help me?” doesn’t mean that we shouldn’t approve that. I can either just leave it or disapprove it — the end result will be the same and that will not be added to the chatbot knowledge base.
Now you need to add an answer to that question. In order to do that, just insert the answer in the “type your reply here” and hit enter.
In the grey box, you can add an answer to that topic. It can be something as simple as “Good” or you can let your creativity fly. Try not to go too crazy though.
One of the main things to notice here is “Always hand this topic off to humans?” switch. When you toggle that switch, then the bot initiates the handoff to human actions that you have specified under “Connect”. This is useful in situations where you want to make sure that if a customer says a certain thing, then the bot definitely will hand over the conversation to a human. Dependent on your use case that might vary. For example, if someone says “My order status has been unchanged for 3 weeks already!” then you might want them to give you the order ID and then a human agent can check what exactly is going on with that order. So you would have the bot answer something like “Please give me your order ID and we’ll check right away.’’
Once you’ve added your answer, you’re good to go. You can set the status of the automation according to your preferences. “enabled” means that upon saving, the automation will go live and the bot will start responding with that automation.
You can use “tag for filtering” to make the later management of all the automations and searching easier by simply adding tags to your automations.
If you press “Save changes” then your automation and everything you just setup is saved.
- You tell the bot what are the different ways a customer might ask a specific question. For example if one of the questions you want to automate is “what if your refund policy” then you should train the bot to understand what are the different ways a customer might ask a question about refunds. You’d need to for example insert training phrases like “what is your refund policy?”, “Can I get a refund?”, “how can I get a refund?”, “can I get my money back?”, “I’d like to get my money back for my order” etc. The more examples of how a customer might ask one question you put in, the better.
- Once that’s done you insert the answer, for example “We only offer refunds on orders placed within the last 30 days, you can read more about our refunds here”.
- You hit the “Save changes” button and that automation is good to go.
Using the same process, you can add as many automations as you think is necessary. It’s okay to just start with a couple of automations and then continue adding more as you go along. It’s a continuous process. The more you train the bot, the better it becomes at understanding the customers’ questions and responding to them.
It’s also a good idea to add automation for common “small talk” items like “how are you?”, “how’s it going?, “hello” etc.
ChatCreate uses in the background a technology called Natural Language Processing which is based on Machine Learning. By using a complex set of rules to analyse and process text inputs, it’s able to recognise what the customer is writing and how to respond to it.
One you’ve setup your first automations you’re good to go and launch the bot. Don’t worry, you can always pause it instantly if you want. In order to make sure that your customers have a great experience, it’s important to make sure that your customers always have a chance to talk to a human agent when possible.
Whenever ChatCreate sees that the customer has asked a question that it hasn’t been trained to respond to, it triggers a handoff protocol. You can customise how the handoff happens under the “Zendesk Chat” window in the “connect” tab. You can read in detail how to do that here.
Congratulations, you’re ready to get up and running with your very own artificial intelligence chatbot. If you should run into any issues along the way, you’re always welcome to get in touch with our friendly team at firstname.lastname@example.org and we’ll get you sorted and setup.