Tweak global instructions
Last updated
Last updated
Instruction is the simplest way to tune the behavior of Aissistant.
From the left-side panel of the Aissist console, click Workspace to edit the basic settings and instructions for your workspace. These settings and instructions are global, they will be used to guide the Aissistant's responses to every user message.
Under Basic Settings, you can define the identity and style of your Aissistant, set up greetings, specify escalation and closing instructions, and determine the language Aissistant should use.
Business Name and Description: the name and brief description of your business. This information will be used to answer relevant question and provide context to fine-tune the behavior.
Identity: the identity for Aissistant, such as a consultant, real estate agent, customer service representative, or sales agent. We recommend giving your Aissistant a friendly name (e.g., Emma), allowing it to introduce itself as Emma rather than as an AI bot.
Purpose: the overall goal for Aissistant. Some common ones are “answer user’s questions”, “collect information for sales team”, etc.
Style: The language description dictates the tone, with multiple styles possible, separated by commas. Commonly used styles include professional, casual, persuasive, humorous, and charming, among others. For instance, if you select the informal style, when responding in Dutch, Aissistant will use "je" or "jouw" instead of "u" or "uw" for the word "you".
Aissistant can automatically detect a user's language and respond accordingly if Auto Language is enabled. If your customers primarily use a single language, you can also set Aissistant to always respond in that specific language. In some cases, a message might contain multiple languages—for example, user surveys might include both English and another language used in the feedback. To handle such scenarios, you can provide language detection instructions under the Language tab in Basic Settings. For example:
"If the user message contains 'language:', use the language code or name that follows as the response language."
Simply set the greeting you would like Aissistant to use. Aissistant will automatically rephrase it each time to ensure users don’t receive the exact same greeting repeatedly.
When Aissistant is unable to fully address a user's inquiry, it can automatically generate system tags such as sys_human_help or sys_human_follow_up. These tags signal that the conversation needs to be escalated to a human agent for further assistance.
The Escalation tab provides you with the flexibility to define clear escalation instructions. Here, you can guide Aissistant on how to gracefully transition the conversation to a human agent when needed. For instance, you might specify an instruction such as:
"If the troubleshooting steps provided do not resolve the customer's issue, inform the customer that their case will be escalated to a specialized support team for further assistance."
Beyond simply escalating, you also have the option to configure an Escalation Summary. This allows Aissistant to gather and summarize key information that will be passed to the human agent, ensuring a smooth handoff and saving valuable time. For example, you can define a summary template like this:
{
"symptoms": "user's reported issue",
"troubleshooting_steps": "troubleshooting actions already taken"
}This structured summary ensures that the human agent immediately understands the context of the problem without needing to reread the entire conversation, leading to faster and more accurate resolutions.
Additionally, Aissistant allows you to implement a conversation-based fallback mechanism. You can configure it to trigger an escalation automatically after a specified number of back-and-forth interactions with the user. For example, you might set it so that if the conversation exceeds ten exchanges without resolving the issue, Aissistant will inform the user that a specialist will take over.
By default, Aissistant does not close any tickets or conversations. However, you can set specific instructions to allow Aissistant to close them when appropriate. For example, you might define:
"Close the ticket once the user has confirmed their purchase and indicated no further assistance is needed."
When a ticket is closed, Aissistant will automatically add an ai_closed tag to the ticket within your agent platform.
Under Instruction, you can define global instructions, provide business context, specify the tasks you want Aissistant to accomplish, and add temporary announcements.
Instructions for Aissistant are very flexible and can work on anything. Usually instructions have two categories:
Generic: generic instructions work for any scenario. An example of generic instruction is “be precise on your response”, “keep your response to the limit of 75 words”, “Use "je" or "jouw" instead of "u" or "uw" for the word "you" when respond in Dutch”, etc.
Scenario: scenario-based instructions usually has a condition. An example is “if user asks for meeting, send them the link …”, “if user ask for human, tell them …”. We recommend create sub agents for scenarios and put contextual instruction in the sub agents.
Contrary to the asset-based content and contextual instructions, the workspace instructions are global and affect every interaction with the AI. As such, instructions should be concise and succinct, typically not exceeding 1000 English words, to optimize Aissistant's efficiency. We recommend create Sub Agents for top scenarios and fill those Sub Agents with contextual instructions to improve accuracy and efficiency.
when giving instructions, it is useful to include reasons which helps Aissistant better understand and follow the instructions. For example, instead of say “limit the response to 75 words”, say “limit the response to 75 words because people don’t like long response in SMS.”.
You can provide additional information that you’d like Aissistant to know in the Contexts. Some useful information like “the support email is support@…”, “the sales email is sales@…”, “the office address is …”, etc.
You can set specific tasks for Aissistant to complete. For example, if you run an auto repair business, you can define the task as diagnosing vehicle issues based on user input and recommending the appropriate repair services.
Announcements are typically used to deliver temporary or urgent messages to users. For example, if there is a service outage in a specific area, you may receive a surge of inquiries related to the issue. In this case, an announcement could inform users that technical difficulties are currently affecting that region and apologize for the inconvenience caused.
