When Fin handles email conversations, it can pause on replies it's not confident about and ask a teammate for input, rather than guessing or escalating immediately. This is called Human in the Loop (HITL). Once a teammate weighs in, Fin writes and sends the final reply. You control how often Fin asks for help.
Human in the Loop for email is separate from the Loop in teammate / agent step available inside Fin Procedures. This article covers the email-specific capability.
Why use Human in the Loop for email?
Email is different from chat in ways that make unsupervised automation riskier:
Email replies carry a lot of prior context that's hard to get right in one go.
A wrong email reply is a durable paper trail — it's costly to walk back.
Email conversations often involve action-heavy requests Fin can't always complete end-to-end.
Human in the Loop lets Fin take on email volume and query types that teams weren't comfortable fully automating, without affecting conversations that are already resolving well.
How Human in the Loop works
When Fin pauses
Fin predicts a confidence score for each email reply — the likelihood it would resolve the conversation without human help. If that score falls below the threshold you've configured, Fin pauses instead of sending.
When Fin pauses, a highlighted Ask a teammate note appears in the conversation inbox. The note shows:
Fin's draft reply
The reason Fin paused
At the same time, the customer receives a short holding reply so they know their message is being looked at.
What can a teammate do when Fin pauses?
When Fin pauses on an email reply and sends an Ask a teammate note to the inbox, the reviewing teammate can:
Provide guidance — reply to Fin's question or approve a direction. Fin uses this input to write and send the final reply to the customer.
Approve the original draft — accept Fin's draft as written and send it.
Take over the conversation — step in directly. Fin stops responding and the conversation is assigned to the teammate.
How does a conversation end?
A conversation that goes through Human in the Loop can end in any of Fin's normal states, plus two outcomes specific to this flow:
HITL resolution — Fin resolves the conversation after incorporating the teammate's input.
Takeover — the teammate steps in and takes over the conversation directly.
You will only be billed for HITL resolutions.
How to configure Human in the Loop
Turn it on
Human in the Loop for email is in open beta and available now. To enable it, go to Fin AI Agent > Train > Escalation and click + New to create a rule. Set the channel condition to Email — you can add further conditions such as audience or message content. Under Fin will, select Ask for input. Then configure the three settings below.
Note: The Ask for input option only appears when Email is selected as the channel condition. If you select a different channel, such as Chat, Human in the Loop won't be available as an action for that rule.
Set the threshold
The threshold slider controls when Fin asks for input. It ranges from Low to High — the higher the threshold, the more often Fin will pause and ask:
Low — Fin only asks when it's least confident. Teammates receive fewer requests.
Moderate — a balanced setting. This is the default.
High — Fin asks more frequently. Teammates receive more requests.
Tip: Start with Moderate and adjust based on how many teammate requests feel manageable. Move toward Low if teammates are getting too many requests; move toward High if you want more oversight.
Who to ask
The Ask setting controls who receives the request when Fin pauses. You can route it to all teammates, assign it to a specific teammate, or direct it to a specific team.
Set a fallback time
The Proceed after setting controls what happens if no teammate responds. After the time you configure (in hours and minutes), Fin automatically sends its original draft reply to the customer. Set this to a time that reflects how quickly your team can typically review requests.
Use it inside Workflows
Human in the Loop can also be configured inside the Fin step in Workflows, so it applies within custom bot flows — not just simple email deployments.
What does the customer experience?
When Fin pauses for input, the customer receives a short holding reply — they're not left waiting with no response, and they don't see an immediate escalation.
Once the teammate provides input, the customer receives Fin's final reply. Because Fin uses teammate guidance to craft the response, the reply is higher quality on exactly the hard, context-heavy, or sensitive questions where an unsupervised reply would have missed the mark.
FAQs
Does Human in the Loop affect conversations Fin is already resolving well?
Does Human in the Loop affect conversations Fin is already resolving well?
No, Fin only pauses when its confidence score falls below the threshold you've set. Conversations Fin is confident about continue to resolve as normal. The goal is to expand what Fin can handle — not to add friction to what's already working.
What happens if a teammate doesn't respond to Fin's request?
What happens if a teammate doesn't respond to Fin's request?
Fin will wait for teammate input until the Proceed after time you configured in the Escalation rule. Once that time expires, Fin automatically sends its original draft to the customer. Set a reasonable fallback time to avoid customers waiting too long.
Can I use Human in the Loop in Workflows?
Can I use Human in the Loop in Workflows?
Yes. Human in the Loop is configurable inside the Fin step in Workflows, so it can be applied within custom bot flows as well as simple email deployments.
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts