AI Workflow Node
Overview
The AI Workflow Node lets your workflows prepare a data card before anyone has opened it. It can find related information for an issue, for example by using an agentic Action that reads additional data, and it can classify the issue, translate it, summarize it or draft the first text for it. When a support agent opens the data card, the groundwork is already done and the agent can concentrate on solving the issue.
The node can run any Neo Action that you have configured: a text Action, a classification Action or an agentic Action. You stay in control of what the AI changes. The node writes only to the attribute that you choose in the node, or, for a classification Action, to the attribute that the Action defines. An AI badge on that attribute shows the support agent which values the AI filled in, and the badge disappears when someone changes the value. This differs from Neo in the chat, where the AI can choose what to write and where, and where a person approves each change.
In version 26.2, an agentic Action in a workflow can only read additional data. Actions that would create or change data cards, which need a person to approve them, are not supported in workflows. Do not ask for such changes in the prompt of an Action that you use in a workflow.
We recommend that you try the node in a test environment first.
When the AI response fails, the node intelligently redirects the process, ensuring continuity. You can define paths for successful completions and exceptions, enabling robust, failsafe automation.
Key Capabilities
- Trigger configurable AI actions from workflows.
- Find possible duplicates and related information with agentic search, and automate tasks like content generation, translations and summarization.
- Works across all customer UIs.
- Routes failed responses to alternative paths.
- Completely configurable to fit customer needs.
- Seamlessly integrates with existing AI Actions.
Benefits
- Reduces manual effort in both content creation and support operations.
- Minimizes low effort support tickets that overload services.
- Speeds up handling in multilingual support scenarios with automatic translation.
- Enables complex automation chains for any text based value creation.
- Improves customer experience with faster, smarter communication.
Text and Classification Actions
The node can write generated text into an attribute (a text Action) or assign a value to an attribute based on AI classification (a classification Action).
- A text Action writes the AI response to the target attribute chosen in the node. If the response is empty, the run fails and follows the exception path (the branch for failed runs, set in the node).
- A classification Action sets the attribute that the Action itself defines.
- For a category hierarchy, the node fills in the parent categories of the classified value and clears the child categories below it.
- For a reference target, the node restores the previous values if setting the classified value or the parent categories fails. The run then takes the exception path.
- Administrators set the minimum AI confidence in ai.classification.confidence.threshold (0.5 by default). Below that threshold, the run fails and takes the exception path.
- Administrators can cap the number of classification options with ai.classification.maxOptions (100 by default). Above that limit, the run fails with the message "Too many options for classification." and takes the exception path.
Setting names and their availability can change between versions. Check the settings for the installed version before changing them.
Example Use Cases
Finding Duplicates and Related Information
Use an agentic Action to search for possible duplicates of a new data card, and write a short result into the target attribute. Support agents see the result when they open the data card, and they decide what to do with it.
Gathering Related Information
Use an agentic Action to look up related data cards, such as earlier tickets about the same issue, and write a short summary into the target attribute. The groundwork is ready before anyone opens the data card.
Translating Support Tickets
Use an attribute to define the target language, and instruct the AI in the prompt to translate accordingly. This enhances communication and speeds up resolution in multilingual teams.
Drafting Customer Communication
Automate resolution suggestions or basic troubleshooting steps. This reduces support load by handling simple tickets automatically.
Major Incident Announcements
Use AI to dynamically compose and send messages. Chain multiple AI nodes for greater accuracy. Cuts down on manual work for predefined templates.
Audit Checkpoint Summaries
Trigger reports at lifecycle milestones. This ensures compliance and real-time insight with minimal manual effort.
Automated Summaries
Generate summaries at any stage of the data card lifecycle. This saves time and improves visibility into key actions.
Setup
- The node is available with a Neo license or an active trial.
- The node uses the AI provider selected for AI Actions. In version 26.2, that provider is set up in AI Core.
Getting Started
To begin using the AI Workflow Node:
- Configure AI Actions.
- Set up the AI provider in AI Core, and confirm that a Neo license or an active trial is in place.
- Define your prompt and input context attributes. In a workflow, write the prompt so the AI only reads the data. Tool approvals are not shown in workflows in the current version.
- Drag the AI node into your workflow.
- Choose the target attribute that receives the AI response. For a text Action, the response is written to this attribute.
- Optionally, choose an error attribute to record failure messages.
- Set paths for both completed and exception scenarios.
Note:
You must have a Neo license or active trial to use this feature.
Without a Neo license, the node writes the message to the error attribute, if one is set, and the workflow continues on the completion path.
Best Practices
- Start with a generic AI Analysis Action and store outputs in a dedicated attribute.
- Chain multiple AI nodes for complex scenarios.
- Use Wait nodes to ensure all required data is present before triggering AI.
- Define exception paths to manage errors gracefully.
- Always mention that AI-generated content is machine-generated when shared externally.
Troubleshooting Tips
- Test the connection to the AI provider. For M42 AI Core, use the M42 AI Core connection test.
- Verify that context attributes have values.
- Ensure the feature is enabled in platform settings.
- Check the M42 Intelligence log (
m42_intelligence) for error details. At the default INFO level it also contains the request text, including ticket and personal data. For how to avoid that, see Logging and sensitive data in the Neo Actions admin manual. - Check the error attribute, if one is set, for failure messages. Each message has a timestamp.
- Use exception handling to prevent workflow interruptions.
Limitations
- It is not possible to hide Actions used in workflow from Agent UI (publish/hiding capability planned).
- Errors from a failed Action run in a workflow are written to the error attribute, if one is set.
- Text Actions do not support reference attributes as target attributes.
- Does not allow writing additional values to multi-value attributes.
- Does not support fetching data behind references.
- Tool approvals are not shown in workflows in the current version, so write prompts that ask the AI to read the data only; for text Actions, the AI response is still written to the target attribute.
- Workflows wait for the complete AI response and do not stream partial text into the attribute.
- Workflow runs do not include user context, such as the user's name or language.
Frequently Asked Questions
Is this feature separately priced? No. It's included in the transaction-based pricing of Neo.
Which languages are supported? Depends on the AI model. M42 GenAI currently supports English, Finnish, German, and Swedish (Q2/2025).
Are images supported? Depends on the provider. OpenAI supports image recognition; since version 2025.2 you can use FileUpload handler attributes.
Related Articles
Table of Contents