Overview

Starting with Version 2026.1, AI is now a native, provider-agnostic capability of the ESM Platform. 

Customers can design, configure, and execute AI-powered functions using a low-code approach, embedding AI directly into existing workflows, UI actions, and the M42 Assist experience. AI is therefore available as an integrated platform capability rather than as a separate application or isolated feature.

This approach:

• Enables AI extensibility through low-code configuration. 
• Embeds AI directly into existing business processes and user interfaces. 
• Supports different AI providers through a standardized integration model. 
• Reduces dependency on a single AI vendor. 
• Simplifies the adoption, deployment, and scaling of AI use cases. 
• Preserves the existing Matrix42 security and action model.

A centrally configured AI provider framework currently supports: 

• Matrix42 Intelligence
• OpenAI
• OpenAI-compatible providers 

The provider can be changed without redesigning the AI Function or the business process that uses it. 

AI Function executions are recorded with relevant execution information, such as the selected function, provider, request, response, and execution status. This supports troubleshooting, auditability, and controlled use of AI capabilities.

Requirements

This capability requires Matrix42 Itelligence subscription using the Action Credit mode and one of the following:

• ITSM 2023 Advanced/Enterprise
• ITSM 2025 Enterprise
• or later

Additional consumption or licensing costs may occur when customers connect their own supported AI models (BYOM). Any such costs charged by the selected AI provider remain the customer's responsibility.

Predefined AI Functions

Matrix42 provides a predefined set of AI Functions that can be used immediately after a supported AI Service Provider has been configured.

AI-powered actions are displayed together with standard actions in the user interface. They are visually identified as AI actions, but follow the same access, security, and configuration principles as other platform actions. 

The predefined functions include: 

Detect Ticket Category 

Analyzes the ticket content and suggests the most appropriate service category. 

The suggestion is based on the available ticket information, including fields such as the title and description. 

Detect Ticket Impact

Analyzes the ticket context and suggests an appropriate impact value. 

The result can be reviewed by the user before it is applied to the ticket. 

Detect Ticket Service

Identifies the service that is most likely related to the reported issue or request. 

This can help reduce manual classification effort and improve routing and reporting quality. 

Detect Ticket Type 

Determines whether the ticket is more likely to represent an Incident, Service Request, or another supported ticket type. 

Detect Ticket Urgency

Analyzes the reported situation and proposes an urgency value. 

The function supports consistent ticket assessment while keeping the user in control of the final value. 

Detect Ticket Resolution

Analyzes the ticket and available context to suggest a possible resolution. 

Depending on the configured scenario, the result can be shown as information, copied into the ticket, or used by an automation process. 

Text Enhancement

 Improves the quality of user-entered text. 

Typical use cases include: 

• Correcting spelling and grammar.
• Improving readability.
• Rephrasing text in a more professional tone.
• Making descriptions clearer and more concise. 

Auto Translation 

Translates text into a selected target language. This can be used for ticket descriptions, journal entries, knowledge content, and other supported text fields.

Where predefined AI Functions can be used

Predefined AI Functions can be integrated into: 

1. UI Actions
2. M42 Assist
3. Workflow Designer
4. Automated ticket-processing scenarios
5. Custom application actions
6. Partner or customer extensions 

The same AI Function can be reused in several contexts without duplicating the underlying business logic.

Predefined AI Agents

Matrix42 can also provide predefined AI-driven automation scenarios based on AI Functions. 

Depending on the installed product version, subscription, and activated extensions, these may include scenarios such as: 

Ticket Preparation 

Analyzes incoming tickets and prepares suggested values before an agent starts processing them. 

This may include suggestions for: 

• Ticket type
• Category
• Service
• Impact
• Urgency
• Possible resolution 

Service Detection 

Identifies the most likely affected service based on the ticket description and available service data. 

Auto-Resolution 

Analyzes eligible tickets and attempts to identify a suitable resolution based on the available context and configured knowledge sources. AI Agents are controlled through their respective configuration, workflow, and security settings. Customers should validate the proposed behavior and results before enabling AI-driven processing for productive scenarios.

Read more here (coming soon).

AI Service Providers

An AI Service Provider defines which AI backend is used to execute an AI Function. 

The provider configuration is managed centrally. This allows administrators to configure the connection once and reuse it across multiple AI Functions and business processes. 

Supported provider types include: 

Matrix42 Intelligence 

Matrix42 Intelligence provides the recommended Matrix42-managed AI capabilities and can offer additional services beyond basic prompt execution. The exact available functionality depends on the customer’s subscription and deployed Matrix42 services. 

OpenAI 

The platform includes an out-of-the-box integration for supported OpenAI models. Customers must provide the required credentials and configuration for their OpenAI account. 

OpenAI-compatible 

Providers that expose an OpenAI-compatible interface can also be connected. Compatibility depends on the provider’s API implementation and the supported model capabilities.

Configure an AI Service Provider

Before AI Functions can be executed, at least one AI Service Provider must be configured and activated. 

The exact configuration fields depend on the selected provider. In general, the administrator must provide: 

• Provider type 
• Service Connection
• Model or deployment name 
• Activation status 
• Capabilities
• Priority

To configure a provider: 

1. Go to Administration > Intelligence > AI Service Providers. 
2. Create a new provider or open an existing provider. 
3. Select the provider type. 
4. Select the required service connection. 
5. Configure the model configuration. 
6. Specify a Capability, e.g., AI Functions.
7. Activate the provider.
8. Save the configuration. 
9. Test the connection, where available. 

After the provider has been configured, it can be assigned to one or more AI Function implementations.

Configure the System Prompt

The system prompt defines the general instructions and behavioral rules applied when AI Functions are executed through the configured AI Service Provider. 

It can be used to establish common requirements such as: 

• Expected response style and tone
• Language behavior
• Formatting conventions
• Security and compliance instructions
• Restrictions on unsupported or unsafe responses
• General rules that should apply across multiple AI Functions 

The system prompt is combined with the function-specific prompt at runtime. The system prompt provides the general execution context, while the AI Function prompt defines the concrete task, such as detecting a ticket category, translating text, or suggesting a resolution. 

To configure the system prompt: 

1. Go to Administration.
2. Open System Settings → Intelligence.
3. Enter or update the AI Function System Prompt. 
4. Save the settings.
5. Test the affected AI Functions. 

Administrators should keep the system prompt generic and reusable. Instructions that are specific to one business use case should be defined in the corresponding AI Function implementation instead. 

Changes to the system prompt may affect all AI Functions. The updated prompt should therefore be tested before being introduced into productive use.

How AI Functions Work

An AI Function represents a reusable business capability, such as detecting a ticket category or translating text. 

Each AI Function can have one or more implementations. 

At runtime, the platform: 

1. Identifies the AI Function to execute. 
2. Selects the applicable implementation. 
3. Resolves the configured AI Service Provider. 
4. Collects the required input data. 
5. Sends the request to the selected provider. 
6. Receives and processes the response. 
7. Returns the result to the calling UI action, workflow, or automation. 

An AI Function can return: 

• Informational text 
• Formatted HTML 
• A suggested value
• Multiple structured output values
• A reference to a generated object 

Depending on the scenario, the result can be shown to the user for review or processed automatically by a workflow.

Custom AI Functions

Create AI Functions

Users can create new AI Functions or configure existing ones to extend their capabilities. 

These custom functions support two implementation types:

• Prompt-based: this implementation type uses Document/Prompt Designer with data model binding.
  • Prompts can dynamically include data model properties (e.g., ticket data).

Example: Analyze the following ticket and suggest the most appropriate category.
Ticket title: {Title} 
Ticket description: {Description} 
Current service: {Service}

The exact binding syntax is provided by the designer and the selected data context.

• Service-based: this implementation type relies on API or service execution.
  This option is suitable when: 
  • The AI capability is exposed through an existing service.
  • Additional server-side business logic is required.
  •  A custom provider or AI orchestration layer must be called. 
  • The response requires processing before it is returned to the platform. 

The service implementation can return informational or structured output, depending on the function definition.

Use AI Functions in the User Interface

AI Functions can be exposed as UI Actions for supported business objects. 

For example, an Incident can provide actions such as: 

• Detect Ticket Category 
• Detect Ticket Service  
• Detect Ticket Impact 
• Detect Ticket Urgency 
• Suggest Resolution 
• Enhance Description 
• Translate Text 

AI actions are displayed together with standard actions and are visually highlighted to make AI-assisted capabilities easier to identify. 

The existing UI Action model remains unchanged. This means that AI actions support the standard platform mechanisms for: 

• Security roles 
• Object permissions 
• Conditions 
• Availability rules 
• Context 
• Localization 
• Action placement 

When a user executes an AI action, the result can be displayed through M42 Assist or directly in the action result. 

Depending on the function, the user may be able to: 

• Review the suggestion. 
• Accept and apply the result. 
• Ignore the result. 
• Copy the generated content. 
• Open a generated object. 
• Continue editing the generated content.

 AI-generated suggestions are not automatically applied unless the corresponding action or workflow has explicitly been configured to do so.

Show AI Actions in the Object Context

AI actions can be displayed directly next to the field or control they are intended to support. 

This gives users access to AI assistance at the point where the result is needed, without requiring them to open a separate action menu. 

Out of the box, selected ticket previews include contextual AI actions for:

• Detect Ticket Category
• Detect Ticket Impact 
• Detect Ticket Urgency 

For example, the Detect Category action can be shown next to the Category field. 

When selected, the AI analyzes the ticket context and suggests an appropriate value. The suggestion is only applied after user confirmation and can be adjusted before saving. 

Administrators and solution designers can add the same pattern to other forms and controls through the Layout Designer. 

To add a contextual AI action: 

1. Open the required layout in Layout Designer. 
2. Select the control where the AI action should be available. 
3. In the control properties, define the Action to be executed. 
4. Define the Action Context that will be passed to the action. 
5. Save and publish the layout. 

The Action determines which AI or standard UI Action is executed. 

The Action Context defines the object or data context available during execution. In a ticket preview, this is typically the current ticket object. The action can use this context to resolve properties such as the title, description, service, category, impact, or urgency. This mechanism is not limited to predefined AI actions. Any compatible UI Action can be attached to a supported layout control, provided that: 

• The action supports the selected object type.
• The required context is available. 
• The user has permission to execute the action. 
• The action’s availability conditions are fulfilled. 

Contextual actions follow the standard Matrix42 security and UI Action model. They are shown only when the configured conditions and permissions allow execution.

Configure a Confirmation Wizard for AI Results

An AI Function can use a dedicated Confirmation Wizard to present its result in a structured, guided interface. 

The wizard allows users to: 

• Review the AI-generated proposal. 
• See an explanation or supporting information. 
• Adjust the suggested value where supported. 
• Accept and apply the result. 
• Reject or close the proposal without changing the object. 

This is useful when an AI Function returns an actionable suggestion rather than only informational text. 

For example, the predefined Detect Ticket Category function uses a confirmation wizard to display the suggested category and the AI-generated explanation before the user applies the change. 

How the Confirmation Wizard Works 

The Confirmation Wizard is assigned in the AI Function configuration. 

When the function is executed: 

The AI Function returns its result. 

1. The result is added to the wizard execution context. 
2. The configured wizard opens in M42 Assist. 
3. Controls in the wizard read the AI response from the context. 
4. The user reviews, adjusts, accepts, or rejects the proposal. 

The result is not displayed automatically unless the wizard has been explicitly designed to consume and present the AI Function output. 

Assign a Confirmation Wizard. To assign a wizard to an AI Function: 

1. Go to Administration > Intelligence > AI Functions. 
2. Open the required AI Function. 
3. Select the wizard in the Confirmation Wizard field. 
4. Save the AI Function. 

The selected wizard should be designed specifically for the output returned by that AI Function. 

Create the Wizard Context 

When creating the confirmation wizard, register the AI response as part of the wizard context. 

The wizard context can include values such as: 

• AIResponse 
• Explanation 
• AIFunctionImplementationID
• Additional function-specific values 

The exact context depends on the AI Function and its returned output. 

For example, the predefined category-detection wizard uses: 

• The suggested category returned by the AI Function. 
• The explanation generated by the AI. 
• The current ticket as the action context. 

The wizard controls are then bound to these context values. 

Design the confirmation interface. In Layout Designer, add the controls required to display and confirm the AI result. 

A confirmation wizard typically includes: 

• A text control for the AI-generated explanation. 
• A field or selection control for the proposed value. 
• An option to adjust the suggestion before applying it. 
• Confirm and reject actions. 

For example, the predefined AI Detect Ticket Category wizard displays: 

• The AI explanation. 
• The suggested ticket category. 
• A selectable category field that can be changed by the user before confirmation.

Map the AI response 

The wizard creator must explicitly map the AI Function output to the wizard context and UI controls. 

For example: 

• The returned explanation is mapped to an Explanation context property. 
• The proposed category is mapped to a category selection control.
• The current ticket is mapped through the Object action context. 

The mapping must match the response structure returned by the AI Function implementation. If the expected output is missing or mapped incorrectly, the wizard may open without displaying the generated proposal.

Use AI Functions With Automation

AI Functions can be executed from the Workflow Designer and other supported automation scenarios. 

This allows AI to be used as part of a broader business process rather than only as a manually triggered action. 

Example scenarios include: 

• Classifying a newly created ticket. 
• Detecting the related service. 
• Suggesting impact and urgency. 
• Generating a ticket summary. 
• Translating incoming content. 
• Preparing a response for an agent. 
• Generating a Knowledge Base draft. 
• Evaluating whether a ticket may be resolved automatically. 
• Returning structured values for subsequent workflow decisions. 

A workflow can: 

1. Start from an event, schedule, UI Action, or another supported trigger. 
2. Pass object data to the AI Function. 
3. Receive the output. 
4. Evaluate the returned result. 
5. Update the object or continue with additional actions. 
6. Request human approval before applying a sensitive result. 
7. For productive automation, Matrix42 recommends including validation or human approval where an incorrect AI result could have a material operational impact.

Security and Governance

AI Functions use the existing Matrix42 security and object-access model. 

Access can be controlled through: 

• Roles
• Permissions
• Audience
• UI Action security
• Object-level access

Administrators should ensure that only authorized users can: 

• Configure AI Service Providers. 
• View or change credentials. 
• Create or edit AI Functions. 
• Access execution details. 
• Execute functions against sensitive data. 
• Enable automated AI processing. 

The data sent to an AI provider depends on the configured prompt, input parameters, and object context. 

Administrators must therefore review each implementation to ensure that: 

• Only the required data is sent. 
• Personal or confidential information is handled appropriately. 
• The selected provider meets organizational and regulatory requirements. 
• Data retention and processing conditions are understood. 
• The configuration complies with internal security policies.

Execution Transparency

Each AI Function execution creates an execution record. 

Depending on the provider and implementation, the execution information can include: 

• AI Function
• Implementation
• AI Service Provider
• Execution time
• Execution status
• Input parameters
• Prompt or request
• Response
• Output parameters
• Error information
• Related business object
• Initiating user or process

This information supports: 

• Troubleshooting 
• Prompt optimization
• Auditing 
• Usage analysis
• Provider validation
• Identification of failed executions 

Access to execution details should be restricted because prompts and responses may contain business or personal data

Troubleshooting

AI Function executions can be reviewed from the corresponding AI Function Executions area. 

When an AI Function does not return the expected result, verify the following. 

The AI action is not visible 

Check: 

• Whether the AI Function and UI Action are active. 
• Whether the current object type is supported. 
• Whether the user has the required role and permissions. 
• Whether the UI Action conditions are fulfilled. 
• Whether the required extension is installed. 
• Whether a supported AI Service Provider is configured. 

The action is visible but cannot be executed 

Check: 

• Whether the provider is active.
•  Whether credentials are valid. 
• Whether the configured model or deployment exists.
•  Whether all mandatory inputs are available. 
• Whether the customer has sufficient Action Credits. 
• Whether the provider endpoint is reachable. 

The execution fails 

Open the corresponding AI Function execution and review: 

• Execution status
• Provider response
• Authentication error
• Endpoint error
• Rate limit
• Timeout
• Invalid model configuration
• Invalid prompt
• Missing data binding
• Invalid structured response
• Matrix42 Dispatcher Service not running

For external providers, also review the provider’s own monitoring and consumption portal. 

The result is incomplete or inaccurate 

Review: 

• Prompt wording 
• Data included in the prompt
• Input object quality
• Selected model
• Output format
• Available knowledge or context
• Temperature or provider-specific parameters, where supported 

AI output depends on the supplied context and the selected model. Administrators should test functions with representative customer data before making them broadly available. 

Structured output is not populated 

Check: 

• Whether output parameters are defined in the AI Function. 
• Whether the implementation returns the expected output structure from the Prompt.
• Whether parameter names and types match. 
• Whether the calling UI Action or workflow maps the returned parameters correctly.