# Creating AI Automations > AI automations are powered by AI agents that handle customer interaction through deterministic actions freeing up your human agents while AI handles many tasks Source: https://help.kustomer.com/en_us/creating-ai-automations-SyuRCs3tZl Last updated: 2026-06-01T14:27:45.816Z AI Automations handle customer interactions through deterministic actions. Simple AI Automations use rules and guardrails, follow procedures, and access tools to process user requests. Configured with strong guidance, these automations solve the majority of AI agent use cases.  With the [multi-agent builder](https://kustomer.kustomer.help/en_us/building-ai-automations-with-the-advanced-builder-Symva33KWl) you can configure multi-agent, guided, and deterministic workflows with full control over rules, routing, and procedures. You can also call custom coding blocks and OpenAPI blocks with tools for customizable functionality.  ### In this article * [Before you start](#understanding) * [Understanding AI Automations and conversation engagement](#understanding) * [Creating AI Automations](#creating) * [Testing AI Automations](#testing) * [Testing AI Automations in the Test Console](#test-console) * [Testing AI agents with evaluations](#evaluations) * [Deploying AI Automations](#deploying) * [Smart routing for AI Automations](#smart-routing) * [Reporting on AI Automations](#reporting) ### Before you start 1. Add any knowledge your agents will reference to your Kustomer knowledge base. 2. Consider the agent's role, and the tools they should have access to.  ### Understanding AI Automations and conversation engagement When Kustomer AI receives an inbound message, it first checks to see whether the message matches its deployment conditions. Next, Smart Routing reviews the conversation to detect customer intent.  The conversation is marked _engaged_ only if Smart Routing determines that the conversation's intent should be handled by your AI Automation. An [engaged conversation is a billable conversation that involves an AI Automation](https://kustomer.kustomer.help/en_us/engaged-conversations-r1CGisY21g). An engagement is generally considered true when: * An AI Automation forms a response and sends it to the customer * An AI Automation attempts to generate a response, regardless of whether it was ultimately sent. * An AI Automation routed the conversation to a human agent. * A human agent assigned the conversation back to the AI Automation. See [Defining AI engagements](https://kustomer.kustomer.help/en_us/engaged-conversations-r1CGisY21g#ai-engagements) for details on when a new conversation is started.  During the conversation, AI Automations can: 1. Route conversations based on explicit customer requests, including to speak to a human agent. 2. Escalate to a human agent when the customer's request falls outside of the AI Automation's guidance.  The AI Automation will act according to its guidelines for customer resolution, including involving humans to escalate as needed, and human agents can also handoff back to AI Automations. ### Creating AI Automations Automation creation is done in simple mode. Here, you provide a mix of conversational and ordered list instruction, provide access to tools, and define the procedures an individual agent team will use to accomplish that team's role. Consider what your agent team to do for you.  A few examples include: Team Capabilities Sample Procedure **Customer Service Team** * Handle shipping issues and “Where is my order” questions. * Process refund requests and check refund eligibility. * Provide low-level troubleshooting advice and escalate complex issues when needed. You are a customer service automation working for Acme co. When a user asks a question about shipping, follow these instructions: 1. Ask the user for their order number and verify that this number matches the number on file.  2. Ask the user for their address and verify the address matches the shipping address on file.  3. Confirm shipping date with the end user. If the user wishes to cancel, send the user to the cancellation team. etc. **Technical Support Team** * Answer questions about API usage and data connections. * Accept technical feedback and moderate community discussions. * Support new users during onboarding. You are a technical support automation working for Acme co. When a user asks a question about their phone's operation, follow these instructions: 1. Ask the user for their account phone number and verify that this number matches the number on file.  2. Ask the user for their phone model and verify the phone model matches the customer phone model on file.  3. Determine customer's problem and route to the battery issues, account issues, or hardware issues troubleshooting procedures   **Travel Team** * Manage bookings, reservations, cancellations, and rescheduling.  * Handle travel disruptions, including cancellations, weather delays, and rebookings. * Coordinate airport travel, including rideshare or taxi services, car rentals, and hotel bookings. You are a booking automation working for Acme co. When a user asks a question about their travel accommodations, follow these instructions: 1. Ask the user for their booking number and verify that this number matches the number on file.  2. Confirm the user's intent and route to the proper procedure.  3. For booking changes, route to the booking changes procedure.  4. For cancellations, route to the cancellations procedure. 5. For additional bookings, use tools such as book\_train, rent\_car, or book\_hotel to resolve user's travel concerns. To create AI Agent teams:  1. From the left side navigation, click **Kustomer AI**. The **AI Automations** screen opens.  2. Click **Add Automation**, and then: 1. Enter a **Name** for the AI Automation.  2. Enter a **Description** for the AI Automation.  3. Next, give your AI Automation guidance on its behaviors. Options for guidance include: 1. Specifying a knowledge source your automation can reference for responses.  2. Providing step-by-step instruction through written procedures your agent must follow. 3. Defining the tone and stylistic methods of communication your agent team will use. 4. Writing free-form instruction to handle edge cases or special rules.  5. Adding guardrails that determine how your automation will discuss competitors or handle secrets (such as sensitive password or identifiable information).  6. For more information, see [Managing AI Automations](https://kustomer.kustomer.help/en_us/managing-ai-automations-SkaCf6nFbg). 4. Click **Save Changes** to save your AI agent and move onto testing, or click **Discard Changes**.  ### Testing AI automations There are two methods of testing your automations. The chat test console is used in real-time as you build your automation, useful for rapid, iterative testing. Evaluations test your automation's consistency at scale, testing for ideal outcomes against specific test cases.  #### Testing automations in the Test Console Find the test console in the right-side pane. This test console is a functional chat, allowing you to interact with your agent as you build it.  To use the test console: 1. On the right-side of the **Build** screen, select a **Test Channel**. Options include: 1. **Chat** 2. **Email** 2. To use the **Chat** console: 1. At the lower right side, click **New Conversation**.  2. Type a chat prompt into the input field.  > **Note:** The chat input field accepts file attachments of any type, allows your customers and the AI to use emoji, and has no maximum input limit.  #### Testing automations with evaluations The second method of testing evaluates your agent against a specific test case. Some example evaluations include: * Testing the refund capabilities of a customer service agent to ensure that the outcome is always positive.  * Testing the booking capabilities of an airline agent to ensure it is capable of consistently changing flight times and providing accurate information to the customer about those changes.  * A rental car agent that can process customer requests for a specific vehicle, and correctly add or subtract charges according to changes in the reservation.  Evaluations help uncover flaws in your team's guidance that affect response at scale so that you can make adjustments before deploying that team.  To use evaluations, you must first build your AI agent team and give it guidance.  To evaluate automations: 1. From the **Build** screen of any AI agent team, click **Test**. The **Evaluations** page opens.  2. Create your [evaluation](https://kustomer.kustomer.help/en_us/use-evaluations-to-measure-ai-performance-BJy49yooxe).  #### Deploying AI Automations Once you have tested and run evaluations to check that your team operates well at scale, you're ready to deploy. Deployment will determine when your agent team will respond to conversations by defining triggers for conditions you want to match.  To deploy AI Automations, you will need to: 1. Add notes about your deployment. Notes are functionally a changelog for each deployment, and are used to identify deployments when attempting to roll back to a previous version. 2. Define the conditions that will trigger your AI Automation.  1. Triggers can match ALL or ANY of the defined criteria. 2. Learn more about [Business rules conditions](https://help.kustomer.com/en_us/conditions-and-actions-in-business-rules-B1PFiShj8#Conditions).  3. Define [smart routing conditions](#smart-routing). #### Smart routing for AI Automations Smart routing rules are brief statements of intent for your AI Automations to follow, for example: * Product returns * Shipping issues * Changes to bookings, etc. When smart routing rules match, your AI Automation will handle the conversation.  Smart routing will send tickets that do not match this criteria to a human agent. Conversations  are not considered engaged conversations until smart routing determines that customer intent matches its guidance. Smart routing can cause the agent to ask up to three follow up questions to clarify a customer's intent before routing will occur.  #### Reporting on AI Automations The **Reporting** screen contains all reporting for AI Automations.  To review AI Automation reporting: 1. From the left navigation panel, click **Reporting**. The **Reporting** page opens.  2. Click **AI Agents for Customers 2.0** to open the report. See [**Reporting on AI Automations**](https://kustomer.kustomer.help/en_us/reporting-on-ai-automations-S12ruA3Ybe) for metric definitions.   For metric definitions, see [Defining AI Automation metrics](https://kustomer.kustomer.help/en_us/reporting-on-ai-automations-S12ruA3Ybe#defining).