Before you can turn your user experience into a dialogue, first we have to make sure that your AI is an expert in your company and product. To do this, you can head over to the Sources tab of the AI zone - the center of all things about your AI.
Note: It is also possible to organize your sources into segments to avoid conflicts. Answers will then be generated using the sources of one segment at a time, preventing mix-ups. This might be relevant if you have sources that relate to different products, internal vs. external use cases, etc. Find out more in this article about segments.
Training your AI
First, you have to train your AI. Luckily this can be done easily and in little to no time. All you have to do is click the +Add training source button and select the type of information you will be adding whether it be website pages, FAQs, PDFs, etc.
Note: The AI is only as good as the content it is trained on so it is important to have comprehensive and up-to-date information. Check out this article about good training sources.
Website pages
If you select website source, this will lead to the pop-up that can be seen in the screenshot below. Here you can simply add your website URL and define whether you’d like the entire website to be scanned or a single page. You can include or exclude certain paths if you think the content on those pages might lead to answers that are not relevant or up to date for example.
It is also possible to add a sitemap as a training source using a URL or uploading a xml file.
FAQ entries
If you select FAQ entry, you will see the pop-up below where you can add a single FAQ by filling in the question and its answer, along with a title.
Multiple entries
Optionally, you can add (or edit) multiple FAQ entries at once. You start by exporting a CSV file containing your existing FAQ entries. You will use this file as a template for adding new FAQs so make sure you have at least one FAQ to begin with.
In this file, if you would like to update existing FAQs, you can simply change the text where needed, without making changes to the ID column. This enables you to do bulk edits without editing each FAQ one at a time in the Unless dashboard.
To add new FAQs, simply insert rows with an empty ID column. You can then upload the CSV file to import the new or updated FAQ entries.
PDFs
Uploading PDFs is pretty straightforward. You just have to mark whether the PDF is public or not. If a PDF is marked as public, the AI will show a link to the source.
However, keep in mind that the formatting of a PDF is very important for ensuring optimal training of the AI. They should be well-structured and organized in a logical and coherent manner.
This can be achieved through the use of appropriate headings, subheadings, bullet points, and other formatting techniques to break up large chunks of text into more digestible pieces.
Images within PDFs will not be indexed, and tables with extensive data might not yield optimal results, as the AI thrives on written content with contextual meaning.
Integrations
In addition to the training sources mentioned above, it is also possible make use of integrations with Zendesk, Salesforce, and the like. These integrations allow you to pull knowledge base articles directly which can mean less clutter. Additionally, this gives you the option to include or exclude sections to get more specific about your training sources.
Once you’ve specified these details for whichever source type you are using, you can click Start training and the AI will scan the source you’ve defined. This usually doesn't take long. The source(s) you’ve added will then appear under the training tab with details about when they were created, when the last scan was, etc. Under Actions, you can also edit or delete a source, view the indexed items, or retrain the AI if the content has been updated since the last scan.
Configuring your AI
You can then use the Configuration tab to adjust the settings of your AI. You can add the name of your product, add your preferred fallback language for cases when the language cannot be confidently determined, and select a persona with a specific focus such as support, sales, or informative.
You can also define a response length. Generally speaking, we would recommend opting for long or max answers. Especially, if you are in an industry that can be more complicated and nuanced, longer responses might be the way to go. However it is best to test the AI with some questions to see what yields the best results for your particular case.
Additionally, you can input a customer service email address. We will then give the user the option to forward their conversation to your customer support team. This is intended for cases where the user isn’t satisfied with the answers given or prefers getting in contact with a human.
You can then add a reply for when the user gives positive or negative feedback to the answer, when they request to get in touch with the support team, and a confirmation message for when an email with their conversation gets forwarded to your support team.
Make sure you click Save before heading into the editor by clicking the AI component editor button.
AI component editor
Once in the editor, click Configure component, which will bring up many options for adjusting the AI to your exact needs.
Layout options
The first field under the layout tab is the starting message, the first message that the user will see when they start interacting with your AI. Here you can also define which language the starting message should be in.
You can choose to automatically detect the language of the starting message based on the page language and browser language. Alternatively, you can select if you want to use page language, browser language or use the original language of the starting message.
Here it's also good to add the logos you want to be displayed with the AI. You also have the option to add starting AI actions and make adjustments around colours and texts and labels that appear. By adjusting these you can ensure that the AI matches your company branding.
Contact details
You have the option to have an additional tab in your AI components for contact information. This way there's always a clear escalation path that people can use when interacting with the AI.
You can have up to three contact cards and you get to define what should be in the cards. You can enable a card and then give it a title and description, select an icon to show with it, and attach your desired action to it.
Above is a screenshot of our Contact tab where we've opted to have two cards which users can select from to email us or to give us a call when needed.
Labeling the tabs
Then we have the Appbar tab where the options are pretty straightforward. Here you can define what the text should be for each of your tabs: chat, search and contact (optional).
Make sure you save your changes and create version and publish!
