Quick Start
Creating and configuring an Aissistant for your work is a straightforward process that can be completed in as little as ten minutes.
Last updated
Creating and configuring an Aissistant for your work is a straightforward process that can be completed in as little as ten minutes.
Last updated
To begin, create an account here. By clicking "sign up," you can easily provide your name, email address, and password to generate your account. Following email verification, you should be able to access your account by logging in.
Upon your initial login to your account, it is necessary to initiate a project for your business. Simply complete the required fields, including a project name, your business name, optional business description and contact email to establish your project.
A workspace is encapsulation of an Aissistant, including definition, assets, instructions, smart taggers, smart actions and associated gateways (agent platforms).
For instance, in the case of an e-commerce business with multiple e-stores, you can establish a distinct workspace for each store. Similarly, if your company comprises various departments, you have the option to create a dedicated workspace for each department.
To create a workspace, simply click on "CREATE NEW" in the workspace dropdown menu and proceed with the following steps.
Kindly provide the essential information for your Aissistant. Additionally, you have the option to customize the style of your Aissistant, tailoring it to either a professional or humorous tone to suit your preferences.
Assets play a vital role in constructing a domain-specific knowledge base for your Aissistant, enabling it to comprehend and respond to domain-specific queries effectively. The most common asset is a document with the frequently asked questions and answers for your business. Accepted forms of assets include websites and cloud-based articles. You can click the following for more comprehensive information on using assets from websites, Google Docs, and Google Sheets.
You can fine tune your Aissistant's behavior by providing it with precise instructions and contextual information. For instance, you can instruct the Aissistant that if it lacks sufficient information to address a user's query, it should acknowledge the user's inquiry and convey that the request has been relayed to our specialist, assuring the user of an upcoming response.
It's important to note that these instructions and contexts are global, meaning they will be fed into the Aissistant in every AI interaction. Therefore, it is advisable to store instructions for specific inquiries in assets rather than prompts to make your Aissistant more efficient. For more details, please click here.
The Smart Tagger is an essential tool for maintaining organized engagement with your users. It enables the classification of customer inquiries and the personalization of responses for each category.
For instance, by selecting the 'No Respond' option for a tag, certain events can be disregarded. Additionally, the default system tag 'sys_human_follow_up' can be utilized to alert the human team to take over specific requests. Within your agent platform, it's possible to create a blank inbox that isn't linked to any channels and establish a rule that reassigns messages labeled with 'sys_human_follow_up' to this inbox. Consequently, human agents can receive notifications from this inbox and address these unique cases.
Further information about the Smart Tagger is available here.
After a final review, the workspace can then be created.
After creating the workspace, you can link it to your agent platform via gateways. To do this, select "Gateways" from the left-hand panel and then choose "ADD GATEWAY". For more comprehensive information about gateways, please refer here.
Currently, the platforms supported are FrontApp, Gorgias, Intercom, Zendesk, and Hubspot, with plans to incorporate Chatwoot, Kustomer, Edesk, and Salesforce shortly.
During the gateway setup, you can choose between Co-Pilot, Auto-Draft, and Auto-Pilot modes for Aissistant.
In Co-Pilot mode, on-demand functions such as drafting, translating, and summarizing are available. These can be activated by clicking the corresponding buttons on the side panel of your agent platform's interface.
Auto-Draft mode enables the monitoring of your inboxes, with automatic tagging of messages and the generation of draft responses for the human agent to review.
Auto-Pilot mode offers complete automation by autonomously responding to incoming messages.
We advise activating Auto-Draft in the initial configuration. This setup ensures that Aissistant autonomously drafts responses rather than sending them directly. Prior to enabling Auto-Pilot, it's essential to perform several rounds of behavioral fine-tuning to ensure Aissistant's performance aligns with the standards of a human agent.
Once the gateways are set up, you have the option to integrate smart actions with your information system, provided it has a publicly accessible API.
For those with a Shopify store, you can follow this Shopify integration guide to link your store and incorporate actions to retrieve customer data or details of an order.
Moreover, if your information system has an available API, such as an API gateway leading to your internal database, Aissistant can connect to this API to gather relevant information in response to user inquiries.
You can read this Smart Action document for detailed steps.
After completing the setup, you are encouraged to test Aissistant's functionality. Utilize our simulator to assess its response to both hypothetical and actual user queries. Additionally, we offer an assets debugger to facilitate the search for pertinent data within assets, along with a smart action debugger that allows you to evaluate the execution and results of the actions.
The simulator is accessible from either the Workspace Config page or the Smart Tagger page. To begin, click "Start" within the simulator, and you're free to enter any questions. The Aissistant will deliver responses based on the context and also show the source of its responses.
The DIAGNOSE tab displays the states and contexts of the AI interaction. It also allows you to customize the context and email to mimic a real user scenario. For instance, if you operate a Shopify store, you can test with your customer's email. With Shopify integration activated, Aissistant is capable of retrieving customer details and order history through the customer's email, enabling it to address related inquiries effectively.
Additionally, we offer an assets debugger on the Assets page. Here, you can enter keywords or questions, and the debugger will generate a list of relevant information along with its sources. This allows you to confirm the accuracy of the information retrieved. It also enables you to identify and resolve any discrepancies or conflicts within the returned data.
On the Smart Action page, we've incorporated a Smart Action Debugger. This feature allows you to input a question and verify if it activates a smart action. Whenever an action is initiated, the debugger displays the API's response, and provides a summary of the executed action, detailing the specific actions taken and the parameters employed.
Now, it's time to launch our Aissistant in auto-draft mode (not the auto-pilot). You can monitor the responses generated by Aissistant either through the Sessions page in Aissist or directly on your agent platform.
Should you encounter any incorrect responses from Aissistant, there are a few possible explanations. If Aissistant is unable to provide an answer due to insufficient information, you'll need to incorporate the missing details into the assets, enabling Aissistant to access the necessary information for the query. In cases where Aissistant delivers inaccurate or inconsistent answers, this might stem from conflicting data within the assets. In such instances, you should employ the asset debugger to pinpoint and resolve these discrepancies. If Aissistant's performance doesn't meet your expectations, you have the option to refine its behavior using instructions and smart tags.
We recommend the following procedure to fine-tune your Aissistant for full automation:
Categorize your customer queries
Turn on “Respond only to tagged” options in Smart Tagger
Develop a smart tag for one of your primary queries.
Populate the assets with relevant information and provide precise instructions associated with that tag.
Monitor how Aissistant handles events associated with that tag.
Adjust the assets, tag, and instructions as necessary for optimization.
After Aissistant consistently manages events in this category without any problems, return to step 3 and select a different category. Repeat the process until all categories are addressed efficiently.
Turn on auto-pilot
For more detailed information, please refer to Fine-tune Behavior.
While we strive for full automation, there are instances where Aissistant may not be able to address every customer query. This could be due to the complexity of the queries, limited access to your internal systems, or the preference to have certain sensitive tasks, like refunds, handled by human agents. In these situations, smart tags can be used to seamlessly transition tasks from Aissistant to your human team.
Our system includes predefined tags such as sys_human_help and sys_human_follow_up for immediate notifications, but you also have the flexibility to create custom tags tailored to your specific requirements. The sys_human_help tag is used when Aissistant lacks the information needed to respond to a query, while sys_human_follow_up is employed when Aissistant recognizes a user's request but cannot execute the related task, necessitating human intervention.
Please refer to Streamline with human team for more details.
For a streamlined handoff process and clear notifications from Aissistant regarding tasks that require human attention, it's advisable to set up a dedicated inbox or view within your agent platform. This special inbox should be isolated, not linked to any communication channel, and should only be associated with sys_human_help, sys_human_follow_up and your customized tags. Consequently, any message that appears in this inbox clearly indicates the need for a human agent's response.
