Procedures enable Fin to resolve complex queries like damaged order claims or account troubleshooting from start to finish. Natural language instructions can be combined with deterministic controls to keep Fin adaptable as conversations shift, while enforcing your rules and policies and taking secure actions across your systems to deliver accurate, reliable answers.

Resolve complex queries end-to-end such as account troubleshooting, damaged order claims, or identity verification.
Build in natural language so anyone can write or update Procedures without technical expertise.
Adapt dynamically to conversation changes when a customer interrupts, adds context, or switches topics.
Stay in control with branching logic, rules, code snippets, and secure data access.
Test safely at scale with Simulations to validate behavior before updates reach customers.
Move non-linearly across steps to match the natural flow of real customer conversations, revisiting or skipping steps as needed.
Balance flexibility with control by combining natural language instructions with structured steps like Conditions and code.

Note: Fin Procedures are available through managed availability. Please contact your account manager to confirm eligibility and learn more about getting access.

How Fin Procedures work

Procedures follow a single, shared instruction flow from defining when they run, to resolving the issue, to handing off to the team or ending. The procedure instructions give Fin full context across the entire interaction.

Create Procedures as easily as writing a document

Write your processes in natural language, just like you would when training a teammate, or let AI draft them for you. Fin follows every step and adapts naturally as conversations shift.

Use natural language to guide Fin - Build Procedures like documents—easy for anyone to create, update, and understand.
Let AI help you write Procedures - Share an outline of your process, and Fin combines this with your content and customer conversations to draft a Procedure you can refine and improve.
Adapt naturally as conversations shift - If a customer interrupts, changes their mind, or adds new requests, Fin reasons about what to do next and adjusts seamlessly, moving to the right step or Procedure without rigid scripts.

Note: To create Procedures you need to have the permission "can manage workspace data".

Add rules, logic, and system access to stay in control

Combine natural language with deterministic controls to define how Fin follows each step and works with your data and systems. Fin can resolve issues from start to finish or hand off to a human when needed.

Add branching logic for reliability - Ensure Fin only follows the right steps for each scenario by adding if/else conditions—making answers more predictable and consistent.
Use code for precision - Add code when you need Fin to follow strict rules when executing certain steps—such as ensuring accuracy for eligibility checks, date calculations, or record updates.
Connect Fin to your systems - Integrate with tools such as Stripe, Shopify, or Linear as well as your internal systems using data connectors or MCPs. Control what data Fin can access and how it uses it, from real-time order updates to account details.
Handoff seamlessly when needed - Control when Fin should escalate to a teammate—so customers get the right support in sensitive or critical situations that need a human touch. This follows the escalation path defined in the workflow following the Fin block that is defined in the Deploy section.
Reuse logic with sub-procedures - Extract common steps into reusable sub-procedures. Reference them from any Procedure to keep logic consistent and reduce duplication. Sub-procedures improve readability and make complex flows easier to maintain.
Use Conditions for major forks in logic, such as eligibility checks or plan-based routing, using natural language or Python for predictable outcomes.

Run Simulations for reliability at scale

Running Simulations before setting a Procedure live is the recommended way to ensure reliable behavior at scale. Trust Fin to handle complex queries reliably and safely with Simulations. Test Fin’s performance across real-world scenarios, from simple flows to tricky edge cases. Confirm it behaves as expected and catch issues before they reach customers.

Simulate full conversations from start to finish - AI acts as a simulated customer, using the context you provide, to have a conversation with Fin. This creates a complete, end-to-end interaction that replicates a real customer scenario. AI also judges the conversation based on the success criteria you set, and assigns a pass/fail results. You get full visibility into why a test failed to debug and fix issues, making testing scalable and reliable. You can rerun saved Simulations whenever you update a Procedure to catch regressions early and iterate with confidence.
Set changes live with confidence - Access all tests in one place and rerun saved Simulations whenever you update a Procedure. Catch regressions early and ensure updates go live confidently.
Create and improve tests with AI - Skip manual test writing. Use AI to generate new tests, edit failing ones, and iterate based on feedback. Expand test coverage faster and refine Fin’s behavior continuously.

Native capabilities

These capabilities come out of the box with Fin Procedures:

Autonomously escalate - Fin will escalate the conversation if it detects the conversation is not making progress, is in a loop, or if the customer explicitly asks for a human agent. Escalation ends the Procedure and routes the conversation to the appropriate team, this follows the escalation path defined in the workflow following the Fin block that is defined in the Deploy section.
Search knowledge - Fin can autonomously search the knowledge hub for context. For example, if you instruct Fin to ask the customer for their “Transaction ID” and the customer responds, “I don’t know where to find it,” Fin will perform a knowledge look-up to see if it's specified in your support content. Fin performs these searches only when needed, keeping conversations focused and efficient.

Limitations of Procedures

Procedures are designed for sequential workflows.

This means:

Actions happen one-by-one: Each step or action (like a Data Connector call) must fully complete before Fin moves to the next part of the Procedure.
No parallel processing: You cannot currently trigger multiple system queries simultaneously to merge their data into a single step.

Note: If your workflow requires real-time parallel actions across different systems, you may need to break these into separate Procedures or handle the consolidation via your own middleware before passing the data to Intercom.

FAQs

When will I get access to Procedures?

Procedures will be under managed availability. We’ve learned that successful implementation requires hands-on setup support. Right now, we’re at capacity to provide this support more broadly. We’re working towards offering this to more customers with the support they will need to be successful, and will reach out once we’re ready. If you’re already using Fin Tasks, please discuss Procedures access with your Account Manager. Learn more about transitioning from Fin Tasks to Procedures. In general, use Procedures for deterministic, multi-step interactions with branching logic or confirmations, and continue using Tasks for simpler, less structured automation.

Are Simulations only available in Procedures?

Yes. Simulations are designed to work within Procedures. To test and preview your AI agent's logic in a full conversation, you should use Procedures.

Why avoid Procedures for simple FAQs?

Building a Procedure for a single-question FAQ (like "What are your business hours?") creates a rigid experience. By using Knowledge Hub articles or snippets:

Fin can answer more flexibly using natural language.
You reduce the time spent maintaining complex logic for simple information.
Customers get a direct answer faster.

When would I use a standalone data connector rather than a full procedure?

Use a standalone Data Connector for direct, single-step, atomic operations (e.g., “cancel my subscription”) that don’t require an ordered, multi-step interaction. You can use prompts within the data connector description itself. Use a Procedure when you need a deterministic, multi-step flow: collecting multiple inputs, explicit confirmations, branching logic, etc.

Fin Procedures explained