All Collections
Fin AI
Fin AI Agent
Train Custom Answers to provide more resolutions
Train Custom Answers to provide more resolutions

Increase rate of resolution via Custom Answers by broadening and deepening coverage.

Ivan avatar
Written by Ivan
Updated over a week ago

Here are some methods you should follow to make your Custom Answers as effective as possible:

It’s crucial that you give time to each of these methods to train Custom Answers. By doing so, you’ll increase their coverage, provide more customers with instant resolutions, and save hours of your teammates’ time each week.

1. Train Fin AI Agent to recognise questions and answer using Custom Answers

Training Fin AI Agent for a single answer is easy. If you’ve got lots of past Intercom conversations, Fin will surface common questions for you to create a Custom Answer for:

Tip: If you use Zendesk, you can use the Zendesk Chat or Tickets importer to quickly set up Custom Answers.

Once you've selected a common question to create an answer for, you'll need to train Fin AI Agent to recognize the various ways customers ask that question that this Custom Answer can help with.

Search for additional questions and phrase the question in different ways:

Tick some additional relevant example questions that Fin AI Agent surfaces for you, aiming for a total of at least 10:

Note: You should also click ‘X’ on any irrelevant examples.

If you’re creating an answer for a question customers haven’t yet asked, you can manually add example questions to train it:

One you set your answer live, you can check that it always answers the right questions. Just open the answer and switch to the ‘Sent’ tab:


  • The ‘Routed to team’ tab is a good place to find specific conversations that weren’t resolved, so you can identify new questions that need answers, or existing questions that need more training.

  • Once you’ve added 10 examples, your training progress will appear as complete. However, you should add more than 10 examples if customers phrase the question in more ways.

  • You can always find more opportunities to train Fin AI Agent under Unanswered questions.

Unanswered questions

From Unanswered questions, you’ll find “Grouped questions” and “Dismissed”:

On the “Grouped questions” tab you can see all questions that don’t have any answers and were not dismissed. Here, you can review unanswered questions, create the answer for that question, add that question to the existing answer or dismiss it.

Note: If you create the answer for an unanswered question, that question won’t disappear from this table immediately, it updates weekly.

When you dismiss an unanswered question it will move to the “Dismissed” tab. Here, you can see all dismissed unanswered questions, and you can choose to restore them and move them back to the “Grouped Questions” tab.

When we run the question group "clustering" operation, we use all conversations over the past year and recompute without taking into account which groups are dismissed. That means a dismissed group could possibly have more or less questions between successive clustering operations but should still appear under the Dismissed tab.

You can also do the same from the side drawer, which is opened by clicking on the question:

Good to know 👇

  • New suggestions of unanswered questions are generated weekly.

  • At least 250 conversations over the last year are required to produce unanswered questions, and only if these questions (or similar variations) are repeated to form a cluster. If you don't have any unanswered questions here, you can still find all unanswered questions in All Reports > Automation > Fin AI Agent where they won't be clustered but you can still view them.

  • When a time period filter is applied, the list will only suggest unanswered questions that have at least one conversation created in that time period.

  • Only one example can be added to a single answer at any one time.

  • A cluster will not be removed even if it is used to create or train an answer.

  • A teammate cannot dismiss a cluster, even if they determine a cluster to be irrelevant.

  • A maximum of 1000 clusters are shown. These are the top 1000 clusters over the last year based on the number of examples within a cluster. Each cluster consists of at least 5 conversations over the last year.

  • Suggestion will not be removed after a new Answer is created, or if the suggestion is trained to an existing Answers.

What is a cluster?

A cluster here refers to a grouping of example questions that are considered similar based on our natural language processing algorithm. To put it simply, a grouped question is a cluster.

​Clustering is the task of grouping a set of questions in such a way that questions in the same group (called a cluster) are more similar (in some sense) to each other than to those in other groups (clusters).

Why didn’t a Suggestion show up, even though I have lots of conversations on this topic?

There might be a few reasons:

  • Conversations are replies to outbound messages, custom bots, or looping bot.

  • Not enough example conversations to form a cluster.

  • The conversations might be considered too different to form a cluster.

  • We filter out greeting words, etc (e.g. Hi, hello, hey)

  • Clusters are generated weekly, so the most recent conversations may not yet exist in a cluster.

2. Add suggested questions to an existing answer

Click "+" / "Train existing" to jump to answer training side-drawer which can be achieved from any of the following new user interfaces:

  1. from the Answer page cluster preview cards

  2. from the Unanswered Questions table

  3. from the Cluster details side-drawer (click any "Common question" in the Unanswered Questions table)

  4. from the Example Questions table (from Cluster details side-drawer, scroll down to find all conversations over the specified date range inside the cluster)

  • Note that the Text is pre-populated with the question

  • Hover over an Answer and click "Train answer" in order to add the example to the selected answer.

  • Only one example can be added to an answer at one time.

3. Create a new answer from the suggested question

Click “New Answer” to jump to answer creation flow from any of the following new user interfaces:

  1. From the Answer page cluster preview cards

  2. From the Unanswered Questions table

  3. From the Cluster details side-drawer (click any “Common question” in the Unanswered Questions table)

  4. From the Example Questions table (from Cluster details side-drawer, scroll down to find all conversations over the specified date range inside the cluster)

  5. From the Answer Training Side Drawer (click “Train answer” from any aforementioned interfaces to reach this view)

Note: The Answer is pre-populated with the question and the cluster language.

4. Broaden Custom Answers coverage

Customers ask questions about a broad range of topics. You should create answers for each of these topics, so Fin AI Agent can provide more customers with instant resolutions using Custom Answers

Identify your common topics

Common topics are different for every business. Some common examples might include ‘subscription’, ‘password’, ‘delivery’, ‘adding users’, or ‘notifications’.

You should work with your Support/Success team to identify the common topics your customers ask about, as well as using common questions Fin AI Agent surfaces for you.

List your topics

Create a list of the common topics you identified. Here’s an example of what a list of topics might look like:

  • Installation

  • Adding users

  • Password

  • Using feature A

  • Using feature B

  • Mobile app

  • Invoices

  • Notifications

Create answers for each topic

Once you’ve created your list, start creating answers for the most common questions within each topic.

By broadening your answer coverage, you’ll increase its coverage and resolution rate. For maximum impact, you should combine broad coverage with deep coverage. 👇

5. Deepen Custom Answers coverage

Customers also ask many questions about the same topic. To deepen Custom Answer coverage, you should create answers for every common question you receive about a single topic.

Identify common questions about a single topic

Once you’ve chosen a single topic, list at least three common questions customers ask about that topic.

For example, if your topic is ‘invoices’, your list of common questions might be:

  • Where can I view my invoices?

  • How do I change my invoice address?

  • How often will I receive invoices?

  • Can I download my invoices?

As you can see, each of these questions are closely related to one topic, but require different answers.

Create an answer for each common question

You should create an answer for each common question you identify within a single topic. Once you set them live, Fin AI Agent will accurately answer more questions on this topic, provide customers with more resolutions via Custom Answers, and reduce your team’s conversation volume.

When creating answers, you should continuously apply each of these three methods - broad coverage, deep coverage and answer training. In doing so, you’ll expand Custom Answers' coverage, provide more instant resolutions, and increase the impact of automation on your business.

Tip: To easily see the answers that need improvement, apply the ‘Underperforming’ filter to your list of answers.

6. Tag Custom Answers

You can tag Custom Answers in Intercom in order to organize content so that you can filter for Answers by a tag, and see any tagged Answers in the Manage Tags page.

Tags are visible only internally, and won’t show to your customers on Answers in the Help Center or Messenger.

How to tag

Adding a tag to an Answer from the Answer View Page (navigated to from the Answers list, or any link which opens an individual Answer) can be done by selecting "Add tag" from underneath the title of the answer.

This should trigger the tags list which will let you add tags to the Answer.

From here, you can search for the title of a tag, add multiple tags, create new tags and save the additions to have the tag(s) added to the Answer.

The list of tags is populated from your Workspace Tags list that can be accessed when you click at Manage Tags.

From there, you can edit or remove your tags:


  • After removing or updating a Tag, all Answers with the Tag added should have the Tag removed.

  • After updating a Tag, all Answers with the Tag added should have the Tag name updated.

If an Answer contains a tag within the content, this will notify the user that the tag cannot be deleted without removing this reference from within the Answer. For example:

You can filter by one or more Tags from the Answer list, this will show all Answers which have the chosen Tag applied.

The Tag filter is applied once the dropdown is selected, where you can click the “Tag is ….” button again, or click anywhere on the page to close the window

Improving Custom Answers

When viewing a user's conversation in Intercom you can choose to rate if the Answer provided to the user was helpful or not for the question that was asked. While this doesn't improve your Answer's training, we use this data internally here at Intercom (along with a range of other metrics) to help us identify where we can make improvements to your Custom Answers. All these data points help us identify where they are performing well , as well as where improvement is needed!

Removing Tags from Answers

To remove a tag from an answer, you can click the cross "X" on the Tag from the Custom Answer Page:

What’s next?


Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts

Did this answer your question?