Step-by-step setup

To successfully deflect customer support tickets from your support channels, you will need to set up a project and make some choices on how you're going to implement your Deflekt.ai. 

 What you will need to do:

1. Create a project
2. Localize the support handover page
3. Ingest content
4. Select and set up channels
5. Test your channels

1. Create a project

Once you sign up you will be prompted to create a project. A project will be a container for your content and therefor your usecase. It's important to think about how you will use Deflekt.ai. Things you typically seperate in different projects:

• If you want to use Deflekt.ai for customers and employees you probably want to use separate projects. Your internal support for employees will probably use content that is not intended for customers (SOPs, personel handbook, etc.). To be able to separate these, you would need multiple projects.
• If your product has an internationa audience and you have content in different languages, you have to think about how your want to do this. If the content is all literally translated your might be able to lean on the ability of the AI model to understand the content. If you for example have different policies for different countries / languages (ie. a shipping policy) you might be better off with using a project per language. In the case of using one language; it's always best to use English as our model has some special abilities for that language like reranking the search results.
• If you have an umbrella company with different products or stores you might want to use one project per subproduct. That way you can have different channels per subproduct and ingest specific content for those subproducts.
• Same goes for companies that serve both B2C and B2B for example. Separating them on a project level, makes it much easier for the model to understand what content to serve.

2. Localize the support handover page

Go to your proect settings to translate all text on the human support handover page. This page is shown to users that click the 'I need more help button' (and yes, you can configure what that button displays). This option is available on all channels and is there for users that run into an issue or question that can't be resolved by the AI.

3. Ingest content

Ingest all content you have or want to use under Content. There is a separate little content guide explaining all the options.

4. Select and set up channels

In the sidebar menu you will find the currently available deflection channels.

Web

The web channel lets you embed a chat bubble or chat iframe on your website or intranet. It's the ideal starting channel for both internal and external usecases.

Email

The original channel Deflekt.ai supported. This gives you an email alias to which you forward emails from customers or employees. The AI will reply to all messages, just like in a chat. 

Slack

Perfect for internal usecases, this will allow you to set up a Slack bot that can answer questions for employees. Very powerful for HR support, ICT support, etc.

Teams

Same, but for Microsoft Teams. A bot will be added that employees can send questions to. 

All channels will allow you to fully translate all text that is used. It also allows you to set up a rate limit. What this does is make sure that users don't 'abuse' your account. Since you have a certain amount of deflections on your paid account, you don't want a user to hit your bot continuously and burn through credits. This rate limit sets how often a single user can ask questions each minute. For email a user is determined by email address, for Slack and Teams by the internal ID in those apps, for web chat it's a session ID.

5. Test your channels

Before you release your channels in the wild it is strongly suggested to test them yourself first. Look at you existing ticket inbox and grab some of most asked questions and try them on all channels you activated. If you like the results, expose them (by embedding the chat bubble, forwarding mails or letting employees know about the new Slack or Teams bot).

Once users start asking for human support, you will be able to locate these conversations and filter them from the Conversations tab. By evaluating these conversations you might be able to pinpoint some black holes in your content. Go ahead and add new content to better deflect these questions. If you can install a process to do this periodically, you will get into a continious circle of improvement. Better content is less questions, but also more questions automatically answered. Both mean less support tickets.