Builder Prompt
Define what the AI Agent should accomplish using a simple prompt. State the goal or outcome the agent need to achieve for this workflow.
ResearchAI
The agent breaks down the prompt, gathers data, analyzes information, and executes research tasks using integrated tools.
Human in the loop
At designated decision points or when complex judgement is needed, the agent alerts a user for review, validation, or escalation.
Routing AI
Based on insights, the agent uses defined conditions to determine next steps, branching workflows into different actions or resolutions depending on dynamic outcomes.
To set up your AI Agent:
1. Go to the AI Builder on the left in the menu.
2. Then go to the Create Agent button on the top right.
3. Add a name for your AI Agent and then Create AI Agent.
4. Type in the command and click "Generate agent".
5. A summary of the AI workflow will be presented in the left sidebar.
6. Besides the overview of the AI Agent steps, you will be presented additional suggested prompts that you can use.
7. For example, click on "Add human review before sending alert". It will show up at the bottom, and you can press the arrow to "send" the prompt to add it.
8. The prompt was added as mentioned in the left side menu bar, the Agent builder.
Please note: The builder prompt can not only add a step, but can also move a step or delete a step. You can add you own prompt logic to the suggestion as well. This means that you have total control on setting up the agent builder and let the prompt arrange your AI agent to your needs.
9. Select the ResearchAI node: Open the Website Industry Change node to inspect how the agent analyzes website content and flags potential industry shifts for each client.
10. The little icons show what tool available to it the agent will be used in a particular step. The list of steps constitute the ResearchAI plan of action for this part of agent's objectives..
11. Process node: open the Process settings to link the onboarding dataset and define which records the agent will scan - filtering keeps alerts relevant to conversion targets.
12. Link the Process to another process, or to a dataset.
Then schedule in the frequency. (Day / Week / Month).
13. Routing node: Here you can see what the logic is: For example:
"If the analysis concluded that industry changes were detected, route to 'Changes Detected'".
If you want to change anything in the routing logic, you can instruct the builder prompt to do so.
14. HITL Alert: A Human-in-the-loop is intended to be used for decision branching in automated processes. When an AI agent process reaches a point where human judgment is needed, HITL pauses execution and presents the user with action buttons (e.g., "Approve", "Reject", "Escalate").
The selected action then determines which path the process takes next.
The HITL flow will look like this:
Your process pauses when it reaches a HITL node.
The alert is created and appears in the case management UI with action buttons.
(1-5 configurable actions)
User reviews the case and clicks one of the action buttons.
The process resumes and the selected action is stored as humanDecision
Downstream routing uses the decision to branch the process
(e.g., "Approve" → verify customer, "Reject" → block customer)
Please note:
HITL is only available in AI Agent Builder mode - it cannot be used in manually configured processes (onboarding, loop, risk, monitoring, etc.).
15. To run the AI Agent: Click on the Deploy agent button at the top right.
Publish the version, and then click Run now. Publishing locks in your tested workflow so monitoring and alerts run reliably.
AI AGENT FAQs
Question: How do I deploy it across more than one branch?
Answer: It is not possible to connect multiple branches of a parent process to it right now.
Question: Is there a way to create an alert in the Command Centre for the output from specific agent's results?
Answer: Yes you can. You can configure your Customer Profile Dashboard overview through the "Platform Settings" and then "Client Page Configuration".
Then just drag the AI Research menu bar into the view of the Client page.
Make sure to press Save.
When you then go to the Customers part of the platform, and click on a customer, you will see the AI Research results published as well under "Customer Details".
