Read on for more information on how to get CXassist to work the best for you or your business.
You will want to start with training the AI after signing up. This is where you can give it all of the information about your company, products, pricing, and more. It's done in 3 easy steps:
That's it! Depending on the size and quantity of files, this step may take a couple of minutes to complete. Simply stay on the screen until it indicates that the training is complete. See the screenshot below for the steps described:
You can also add plain text to the AI. For example, you may have a simple piece of information you want to add to it's training set from a question you are receiving often, like hours of operations. Simply enter your text, and follow steps 2-3 above.
Once your content has been uploaded and the AI has been trained, you can download or delete it at any time. For example, if you wanted to upload an updated product guide, you could delete the old one, and re-upload the latest content.
Once you have the AI trained on all of your content, you may want to test it to see how it will perform before you turn it on for your inbox. You can do that in the "Playground" section of the service, allowing you to try any example email content.
Settings is where you'll go after you've trained the AI. This is where you can determine specific rules for how it functions. Each setting has tooltip help - just hover your mouse above the question mark and you'll see some helpful descriptions for each setting. Here, we'll break down in more detail how each settings works.
The "Act as" Personas help tailor the responses from the AI in a certain direction or style. The Sales persona will act more like a typical salesperson, and guide the person towards action, such as a quote, purchase, or similar. The Support persona will act more like a typical support representative, with more empathy and will work towards a resolution for whatever the customer is experiencing. The Generalist persona is exactly that, a blend of support, sales, and administrative support in it's responses, and will take a more neutral approach.
Keywords are specific words in the email message received, which then trigger a human override. Some examples might be "purchase", "emergency", "promotion", "outage". In this case, the email will not receive a response by CXassist, and instead will be sent to the Email you have in settings below.
This is where you can designate a specific email address to receive emails that CXassist believes needs human review. Any emails with keywords will be sent to this address, as well as any emails that CXassist cannot answer from it's training set.
You can use the native Labels functionality within Gmail to create flexible rule based ignore lists for CXassist. Within Gmail, the Labels functionality supports many different email attributes, which you can use to trigger the application of a label to an email. If CXassist sees the specific label listed here, it will then ignore the email(s). To provide some examples of how this might be useful, here are some use cases you may want to use Labels:
You can have CXassist ignore entire domains, so that it will not respond to any messages from anyone within that domain. This is especially useful for ignoring emails from co-workers, automated platforms, specific partners, and more.
Set how often you want CXassist to run. It will check your inbox for unread messages at whatever frequency you set here. It's recommended to set it to 5-10 minutes for normal operations.
This is where you can turn the service on or off, at any time. Simply toggle on or off.
By default, CXassist will only check your inbox for messages created today, and respond to those messages. However, if you want CXassist to respond to messages older than the current date, you can toggle this setting. Note that doing so means that CXassist will reply to every unread message in your entire inbox. Use with caution.
The queue is where you can go to view the activity from the bot. This will show the threaded conversation for every message sent, any messages that were ignored or sent to the human review email address.
