AI Chatbot leverages cutting-edge AI models, (GPT-4o mini) to comprehend customer inquiries, conduct targeted searches based on the knowledge you manage in your help center, and deliver precise responses centered around your help center content. This approach minimizes inaccuracies, misleading information, and off-topic answers.

The chatbot's responses are designed to be conversational, and it has the capability to ask follow-up questions, ensuring the most relevant and helpful answers are provided. In situations where the AI cannot find a suitable response, the system smoothly transitions the conversation to customisable steps in your chatflow, such as connecting you to a support team, in order to maintain a seamless customer experience.

💡AI Chatbot only returns answers based on content that a users have permission to access.

Getting Started

There are two prerequisites needed before you can setup AI Chatbots in your chatflow. First you need to ensure you have established an AI connection to OpenAI. Follow this super simple guide to get your AI Connection in Deskpro here.

Second, you’ll need to ensure you have created AI Data Sources. Think of AI Data Sources as the pool of knowledge that the AI bot then references to return answers to customers. This can be setup using this super simple guide here.

Once you have set up both an AI Connection and the AI Data Source, then you’ll be ready to setup the AI Chatbot interaction in Chatflows.

Setting up AI Chatbots in the Chatflow Canvas

First, drag and drop an AI Chatbots interaction onto the canvas. Then click on the AI Chatbot interaction on the canvas to open the settings drawer.

Once the setting drawer is open, give the interaction a name. This is for internal use and only sets the name for the AI Chatbot node when you view it on the chatflow canvas.

Next, choose the AI Data Source you want this AI Chatbot interaction to use. Remember, the AI Data Source acts as the pool of knowledge the customer’s query is run against. Therefore it is important to select the correct data source so that customers can expect the correct information.

Answer Found versus Answer Not Found

There are 3 possible outcomes for the AI Chatbot interaction in a chatflow:

1. Answer found and it resolves a customer’s query: Help Center content is found that matches to the customer’s query and the chatbot delivers a positive result for the user. The user can continue to ask follow on questions or leave the AI Chatbot interaction via an escape hatch which can be configured by the admin.

2. Answer found and it does not resolve the customer’s query: Help Center content is found that matches to the customer’s query, however, the user does not find it useful. The user can then select an option which provides an escape hatch for further support.

3. Answer not found: Help Center content is not found for the customer’s query: The chatbot can ask a user to rephrase the question, and if no answers are found, then a fallback message and follow on action can be configured.

In each case, you can customize what is displayed to the customer.

To begin, let’s setup the Answer Found section.

Help Center Content Found

First, you’ll see a text box to configure a message that only appears with a positive result, when Help Center content is found. Be mindful that if you have multiple languages setup in your messenger, then you should also add the corresponding translations.

Next, you’ll see a setting for setting the TRUE and FALSE labels. These serve as the options that are presented to a user after the Chatbot returns a response. An example option for what to use as the TRUE value can be something like "Yes, that helped". Inversely, an example option for what to use as the FALSE value can be something like "No, get more help".

You can configure these using text or even emojis. The choice is yours!

💡After the Chatbot returns a response, we recommend asking the user whether it was a helpful answer or not. For the TRUE/FALSE labels, it’s best practice to keep these short and sweet. Usually a thumbs up or down emoji, or even a simple “Yes, that helped” or “No, that did not help” will do the trick perfectly!

It is important to understand what happens when a user is presented with the TRUE/FALSE labels.

If a user selects the TRUE option, then they will be passed to the next step in the chatflow. This can be configured by connecting the GO TO component called Successfully handled in the AI Chatbot Node, to a subsequent step in your chatflow.

When a user selects the FALSE label, then they will automatically receive the Fallback Message, and will exit the AI context and be passed to the next step in your chatflow. This can also be configued by connecting the Go To component in the AI Chatbot Node to a subsequent step in your chatflow.

Now let’s set up what happens when an answer is not found.

Follow On Questions

When Help Center content is found, and the AI Bot responds to the user, they can ask follow on questions by typing out another question and submitting it. The AI Bot will use context from previous answers to find the appropriate answer.

Allowing follow on questions can be enabled or disabled by an admin.

If it is enabled, then an admin can specify how many follow on questions a user is allowed to ask. Once the question limit is exceeded, then the user will be directed to the next step in the chatflow. This can be configued by connecting the Go To component called Follow-on question limit exceeded in the AI Chatbot Node to a subsequent step in your chatflow.

If follow on questions are disabled, then a user will only be able to ask a single question before being directed to the next step in the flow.

💡When follow-on questions are enabled, selecting a TRUE/FALSE option after the AI Bot returns a response is not required. If the user asks a follow-on question, then the TRUE/FALSE options will populate at the bottom of the most recent AI Bot response.

Help Center Content Not Found

You can elect whether you want customers to be able to re-ask their question if the AI Chatbot does not find content on the first try. This can be turned on via the toggle for "Ask customer to provide more infor or rephrase."

Just as in the Answer Found section, you’ll see a text box to configure a message that only appears with a negative result, when Help Center content is not found. Be mindful that if you have multiple languages setup in your messenger, then you should also add the corresponding translations.

Additonally, you can set the max number or rephrase attempts that a user can try before they receive the fallback message and are passed to the next step in the chatflow.

💡We recommend using a sensible max rephrase attempt limit to help save costs and prevent any customer frustration.

If rephrase attempts are enabled, when a user exceeds the max rephrase attempt limit, they will automatically receive the Fallback Message and will be passed to the next step in the chatflow. If rephrase attempts are disabled, and the user selects that the answer was not helpful, they will automatically receive the fallback message and be routed to the next step in the chatflow as an escape hatch.

The Fallback Message can be configured in the last step of the AI Chatbot settings drawer.

Once you’re done configuring the settings, click Save in the settings drawer and connect the AI Chatbot interaction wherever you need it in your chatflow journey. Ensure you save the canvas, publish the chatflow, and assign the chatflow to the messenger you widget you want to use so it’s available to your customers!

The User Experience

When a user reaches the AI Chatbot interaction in a chatflow, the reply bar will unlock and they will be able to type and send a question. Once submitted, the AI Engine will search indexed knowledge sources, and if it finds a match, then it will return an answer to the user. At that point, a user will be asked a follow on question, and depending on the choice they make, they will either exit the AI context and move on to the next step in the chatflow, or be directed to the fallback message, and then to the unhappy path of the chatflow. In the event that there is no matching content that the AI Chatbot can answer, then a user will be asked to rephrase up to the max limit set by an admin. If the AI Chatbot is unable to answer, then the user will be automatically directed to the next step (unhappy path) in the chatflow.