Test your content readiness by seeing how Fin responds to actual customer queries. Review the sources and settings that shape Fin’s answers, and receive tailored recommendations to improve answers to deliver the highest quality support for your customers.

Features available for this beta:

🛠 Bulk answer import & testing: Easily assess Fin’s accuracy by testing responses against real customer questions. Import queries directly from your support inbox or upload them in bulk to get confidence in Fin’s performance at scale.
🎯 Content targeting: Choose specific content to test with Fin to understand how Fin's answer varies based on your existing audience targeting.
🔍 Answer inspection: Understand why Fin responded in a certain way, with a snapshot view of the sources and configurations (like Fin Personality and Guidance) used to form the response.

Fin Testing is designed for all customers who use Intercom to support conversations – whether you’re an existing Fin customer or simply looking to explore its potential.

To access Fin Testing, teammates must have a full seat and their conversation access permission set to "All conversations".

How to access Fin Testing

Go to Fin AI Agent > Test.
Set the Test settings by:
1. Selecting the Fin audience(s) you've applied to content to see Fin's responses to customers who match the audience.
2. Selecting the brand on your workspace you want to test. (You may only have one brand. Brands are connected to specific Help Centers).
Then run a test on Fin's answers to customer questions by choosing to:
1. Generate from past conversations
2. Upload CSV
3. Add questions manually

Note:

Generating from past conversations requires a minimum of 10 conversations.
You can upload up to 50 questions at a time via CSV or manually, enabling you to test multiple queries simultaneously.
Using content audiences requires setting up audiences and content targeting for Fin.

Features of Fin Testing

Filtering

Easily filter test results to identify which questions were answered and which were not. This feature is particularly useful for testing large volumes of questions and prioritizing areas requiring attention.

Answered: Fin successfully provided a direct response.
Unanswered: Fin was unable to generate a direct response.

Answer inspection

Dive deeper into how Fin’s responses are generated by reviewing configurations such as Guidance and Personality settings. You’ll also see which content sources influenced the answers, making it easier to refine content for optimal performance.

Answer rating

Fin’s responses can be evaluated at the question level. You can rate a response as “Good” or “Poor” and, if poor, select a reason from a predefined list of options. You can also provide additional context for any rating, with all feedback captured in a downloadable report to help prioritize updates effectively.

Note: By rating a response "Good" or "Poor" you’re not training the AI Agent. These ratings are for your own reference and are used when downloading/sharing the report with your teammates.

Recommended suggestions

When you rate an answer as “Poor,” Fin will offer tailored suggestions based on the reason selected, making it easier to address any gaps quickly and efficiently.

Convenient test controls

Export your test results as a CSV file, including questions, answers, ratings, and content sources.

You can also reset the test, re-run the test, or re-test specific queries after updating Fin’s settings using the “Answer again” option.

Current beta limitations

There are a few features which the Fin Testing beta doesn’t support, but we’re currently working on these and hope to launch them in the near future. These include:

Simulate a user

While you can run a test using content for a specific Fin audience or brand, it's currently not possible to test Fin as a specific user/customer and only see results relevant to them. We're working on a feature which would allow you to simulate a persona and see Fin’s responses exactly as the user/audience segment you're testing.

Test Custom Answers

When a question is asked where a Custom Answer has been set up, it won't be surfaced in the test. Instead it will either show Fin trying to respond using content you've enabled in Knowledge, or it would be marked as "Unanswered".

Test Data connectors

Fin Testing does not support triggering Data connectors that would normally be activated in real-world scenarios. Instead, Fin will either attempt to provide an answer or display a message indicating that a direct answer could not be provided.

Test Fin Tasks

Fin Testing does not support triggering Fin Tasks that would normally be activated in real-world scenarios. Instead, Fin will either attempt to provide an answer or display a message indicating that a direct answer could not be provided.

Test Fin Vision