The Funnel is the process an AI Agent follows to meet its goal (it used to be called "workflow"). It's built with states: each state is a step of the process. Within each state you define what information it has, what data it collects from the user, who resolves it, and, optionally, the instructions the agent follows when conversing. This way you replace the rigid trees of earlier versions with a process the agent interprets and executes.

You can:

• Add and chain the states that make up the agent's process.
• Define the name and a short description (label) for each state.
• Specify what data the agent must collect in each state.
• Choose who resolves each step: the AI Agent, Human in the Loop, or a Human agent.
• Write the instructions the agent follows, and add knowledge, tools (skills), and logics.
• View the Funnel as a diagram or as a Kanban, and test it.

Key concepts

Funnel: the agent's complete process, made up of states.

State: each step of the Funnel. It starts at "Start" and chains to the following ones.

Required data: fields the agent obtains from the user in a state; they're saved and available for the following steps.

Who resolves: defines whether the step is resolved by the AI Agent alone, with human supervision (Human in the Loop), or by a person from start to finish (Human agent).

Instructions for the agent: what the agent actually follows when conversing in that step (the "prompt"). It's different from the short description, which is just a label for your team.

How to add a state to the Funnel

Step 1: Open the Funnel tab

Go to the AI Agent and open the Funnel tab. You'll see the Start state and the states you add. With View Kanban you switch to a board view, with Funnel settings you configure its options, and with the zoom control you adjust the zoom level.

Step 2: Add a state

Use the "+" connector that appears at the exit of a state to create the next one and chain them in the order of the process. Each state represents a step the agent must perform.

How to configure a state

Clicking a state opens its configuration window, with four numbered sections. Fill them in from top to bottom.

Section 1: Step information

Give the state a clear Name and a Short description (maximum 140 characters) of what happens there. For example: Name "Data request"; Description "Collects conversationally: name, company or venture, industry, and what problem they want to solve".

Note: this description is just a label so your team understands the Funnel. It is not a prompt for the agent: what the agent actually follows is written in section 4.

Section 2: What data is required when reaching this state?

Add the fields the agent must obtain from the user in this step (for example, Name, Company). For each piece of data, specify how it obtains it and, if applicable, mark it as Required. This data is saved and available for the following steps of the Funnel.

Section 3: Who resolves this step?

Choose one of the three options:

• AI Agent: the agent resolves this step autonomously.
• Human in the Loop: the agent resolves it under human supervision. Ideal for approvals, complaints, or sensitive cases.
• Human agent: a person resolves the step from start to finish, without the AI agent's involvement.

If you choose Human in the Loop or Human agent, in "Who attends?" you must choose who receives the case: a support team or a specific agent (person).

Section 4: Instructions for the agent (optional)

Write, in your own words, how the agent should behave in this step: how to greet, what to ask, what tone to use, and what to avoid. This is what the agent actually follows when conversing. From here you can also add, for this step, Knowledge, Tools (skills), and Logics.

Final step: Save with Done

When you finish configuring the state, click Done. To delete the state, use Delete within the same window.

Note: changes are saved automatically, and the Funnel supports real-time collaborative editing. To test the complete process, use the orchestrator's Test button.

Remember to visit our Help Center for further information.

Botmaker Team