Architect: Call an Agent

Call an Agent is a journey element that lets you embed AI (artificial intelligence) reasoning directly into your Architect journeys. At runtime, it evaluates each user's context against a plain-language instruction you define and returns a structured result that can drive routing decisions, personalize message content, and enrich user profiles.

This feature is released in beta. If you want to get an early access, contact the Insider One team.

Call an Agent removes the need to build external AI enrichment pipelines or involve engineering to add intelligence to your journeys. You write instructions in plain language, reference the user data you want the AI to evaluate using dynamic content, and define what kind of answer you need back. When each user reaches the element during a journey, the AI processes that user's specific context and returns a result that downstream elements can use.

That result can do three things:

• Route users: Send different users down different journey paths based on the AI's decision, using a single Check Response condition on the agent output instead of a multi-branch rule tree.

• Personalize messages: Insert the AI's output directly into Email, SMS, Web Push, App Push, and WhatsApp messages.

• Enrich the user profile: Save the AI's output as a user attribute with the Update User Attribute element, making it reusable across future journeys, campaigns, and segments.

Call an Agent is available in all standard (non-transactional) Architect journeys.

How does Call an Agent work?

The Call an Agent inner page has one configuration area called Agent Configuration, made up of two sections: Instructions and Outputs. When a user reaches the element in the journey, the system resolves the user data referenced in your instructions, sends it to the AI along with the instructions, and writes the typed result back into the journey so downstream elements can use it.

Instructions

The Instructions field is where you describe what the AI should do. Write it as you would explain the task to a person, clearly, in plain language, with a specific expected output in mind.

You reference user data inside the instructions using dynamic content. Click the Add Dynamic Content button inside the field to insert a reference to a user attribute, event parameter, or upstream journey variable. The dynamic content you include determines exactly what data the AI receives: only the fields you reference are sent. This keeps your data governance clean and avoids unnecessary exposure of user information.

You can reference the following components:

• Default and custom user attributes

• Event parameters

• Variables from earlier journey steps, including Call an API response variables and outputs from upstream Call an Agent elements

Each dynamic content reference can include an optional fallback value to use when the attribute is missing for a given user.

The instructions field is limited to 4,000 characters. Longer instructions consume more input tokens at runtime and increase the cost per execution. Keep your instructions focused on one task. If you need multiple AI outputs in the same journey, use separate Call an Agent elements in sequence.

A warning banner appears at the top of the configuration explaining that users will proceed to the next element immediately when the token limit is reached.

Outputs

In the Outputs section, you define the structured data the AI must return.

Fields

• Output Group Name: The human-readable label for this agent's outputs (shown in the UI and used when referencing outputs in downstream elements)

• Output: The key used to reference this output downstream (e.g., follow_up_action, urgency_score)

• Data Type: One of

  • String: A text label or generated copy

  • Number: A score or count. Optionally configure From and To bounds to constrain the value range

  • Boolean: Yes/No, True/False

  • List of Values: A closed list of allowed String values you define under the Values column. The AI is constrained to return only one of these values.

Click the Add Output button to add more output fields. Each output appears as a separate row.

The AI's response is validated against the schema before it is written to the journey context. If the response does not match the declared schema (wrong type, missing field, value outside the allowed list, or a number outside the From/To range), the execution is logged as a failure, and the user proceeds without an output value.

When using the List of Values, define all acceptable values before saving. Any AI response outside the set is treated as a failure. The maximum number of allowed values is 50, with each value up to 255 characters.

Outputs downstream

Once the Call an Agent element completes, the output variables are available to the rest of the journey for the duration of that user's execution. They are journey-scoped and do not appear on the user's profile unless you explicitly add an Update User Attribute element after the Call an Agent element.

You can use the outputs in:

• Check Response element to route users based on the AI's result. Set the source to "Agent Response" and select the agent output variable.

• Message content to insert the output as dynamic content in Email, SMS, Web Push, App Push, and WhatsApp messages

• Update User Attribute to save the output to the user's profile for use in future journeys and segments

Example: Churn risk routing

Imagine a fashion retailer wants to identify users at risk of churning after a return event and send them a targeted win-back offer.

The journey starts when the user completes a return event. The user enters the journey and reaches the Call an Agent element.

The element is configured with:

• Instructions referencing the user's last three purchase dates, number of returns in the past 90 days, and email open rate in the past 30 days, asking the AI to classify churn risk as High, Medium, or Low.

• Outputs: An Output Group Name of Churn Classification, with one List of Values output named churn_risk containing High, Medium, or Low.

The AI evaluates the user's data and returns High. The journey proceeds to a Check Response element that reads churn_risk. High-risk users receive a win-back email with a 20% off offer. Medium-risk users receive a standard re-engagement push. Low-risk users exit the journey.

Users classified as High receive a personalized win-back message, routed with a single Check Response condition rather than a multi-branch decision tree.

Limitations

Do not use Call an Agent in journeys that send OTPs, financial transaction confirmations, medical notifications, or other critical messages where an incorrect AI output could cause direct harm to users. Warnings are displayed in the UI, but it is your responsibility to review your journey design for these cases.

Call an Agent sends selected user data to an external AI model. Only reference the fields your instructions require. Review your data selection for compliance with GDPR, CCPA, and any applicable privacy obligations before activating the journey.

Analytics

After launching your journey, you can see live statistics on the Call an Agent element directly on the canvas. Click the element to open the full analytics breakdown.

• Arrived: The number of users who reached the Call an Agent element

• Success: The number of executions where the AI returned a valid, correctly typed output

• Failed: The number of executions where the AI call failed (timeout, type mismatch, missing field, value outside the allowed set, or infrastructure error)

• Skipped: The number of users passed through to the next element without an AI call because the usage limit was reached

• Dropped: Users dropped from the journey at this step

• Exited: Users who exited the journey at this step

Failed and Skipped executions both result in the user proceeding to the next element with no output value. Add a default path in the downstream Check Response element to handle these cases.

Use Cases

Intelligent journey routing

Use the AI to classify a user's intent, risk level, or current state when they reach the element, then automatically route them to the most relevant path. Instead of maintaining a rule tree with dozens of conditions, one Call an Agent step evaluates the full picture and returns a classification you can branch on directly.

Example: In a SaaS renewal journey, score each user's likelihood of expansion based on their feature usage and plan tier. Route high-likelihood users to an upsell path, mid-tier to a standard renewal reminder, and low-likelihood to a retention flow.

Churn prevention

Evaluate a user's recent behavior and engagement signals to identify churn risk before they disengage. Route high-risk users into a tailored retention flow as soon as they trigger a key event.

Example: After a return event, classify the user's churn risk as High, Medium, or Low based on return frequency, purchase recency, and email engagement. Send High-risk users a win-back offer with personalized copy; send Medium-risk users a standard re-engagement push.

Sentiment-based personalization

After a support interaction or survey response, read the user's sentiment and adapt the follow-up journey and message tone accordingly.

Example: After a CSAT survey event, classify the response as Positive, Neutral, or Negative. Route Negative responses to an apology flow with a recovery offer. Route Positive responses to a referral or upsell sequence.

AI-generated message copy

Generate personalized copy for each individual based on who they are and what they have done, rather than writing a single message for everyone.

Example: In a post-purchase email, use Call an Agent to generate a personalized product recommendation paragraph based on the user's purchase history and browsing behavior. Insert the output directly into the email body using dynamic content.

Smart summarization

When a journey pulls in complex external data via Call an API, such as booking changes, order updates, or billing details, use Call an Agent to turn that data into a clear, concise message the user actually wants to read.

Example: After fetching a disrupted itinerary via Call an API, use Call an Agent to summarize the changes into two sentences suitable for a push notification. Each affected traveler receives a readable, personalized update rather than raw data fields.

Replace complex rule trees

Consolidate dozens of branching conditions into a single AI step. The AI evaluates the full context and returns one clear, actionable output.

Example: Replace a journey with 15 condition branches based on loyalty tier, recency, channel preference, and lifecycle stage with a single Call an Agent element. Configure it to return the correct next action for each user and branch on that output. One element replaces the entire rule tree.

Build AI-enriched user profiles

Use Call an Agent to classify or score users, then persist the results in the user profile using Update User Attribute. Over time, this builds a richer, AI-enriched profile that powers smarter segmentation and more personalized journeys without repeating the AI call.

Example: Run a sentiment classification journey after each support ticket is closed. Persist the sentiment label to a custom attribute ai_last_support_sentiment. Use that attribute in your CRM segments to identify users who have had repeated negative experiences and target them with a dedicated recovery program.