How to Improve Intercom Fin Resolution Rate

Last updated: October 2026

Summary: To improve Intercom Fin's resolution rate, fix the knowledge base Fin reads from. Start with the conversations Fin couldn't resolve in Recommendations, rewrite the articles behind them so each answers one question, fill the gaps, and update articles with every product release. Fin reports a 76% average resolution rate across its customers, and an Intercom support engineer puts a good starting point at 30 to 50%, so the gap between those two numbers is mostly content.

In April 2025, Cursor's AI support bot emailed users that their accounts were being restricted to one device per subscription. Users began cancelling. The story spread across Hacker News and Reddit within hours, and Cursor's cofounder acknowledged the "incorrect response" from the bot and confirmed the policy had never existed. A session bug was causing the logouts, and the bot invented an explanation because no article documented Cursor's actual session behavior.

One missing article turned into cancelled subscriptions and a public correction from the company's cofounder. The bot hallucinated because it was asked a question and had nothing to retrieve.

Fin pulls from the same kind of knowledge base. If the article it retrieves is six months out of date or missing entirely, your resolution rate absorbs the cost.

A quick note on names: Intercom the company renamed itself Fin in May 2026, and Salesforce completed its acquisition of Fin in September 2026. The help desk is still called Intercom and the AI agent is still Fin, so everything below works the same.

What follows is the practical checklist for finding and fixing those articles, starting with what Intercom gives you natively and ending with what lives outside it.

Start with Fin's Recommendations

Go to Fin AI Agent > Analyze > Recommendations. This screen used to be called Optimize. Fin reviews the conversations it couldn't answer and sent to your team, then gives you weekly recommendations, each with an impact score. Content gap recommendations also fire on high volume or sudden spikes. Recommendations needs the Pro add-on.

Recommendations groups what it finds into three gap types:

Content gaps are where Fin could not answer because help content was missing, unclear, duplicated, or contradictory. For each recommendation, you can see the exact conversations that triggered it, review an AI-generated draft for new or edited content, and update articles or snippets directly. Intercom's own team says a weekly review yields 10 to 15 suggestions, and most are small updates that can be implemented in an hour. Start here.

Customer data gaps appear when Fin needed information from an external system that was not available, like order status or account details. Each recommendation includes a guide outlining the API and data connector needed, an implementation effort estimate, and the customer queries it would resolve. These require engineering work.

Action gaps show where Fin needed to take an action in another system, like updating a record or cancelling an order. These point you toward building Fin Procedures or Fin Tasks.

Data and action gaps require engineering resources and longer timelines. Content gaps you can fix this week.

Check your content performance report

Go to Fin AI Agent > Analyze > Performance and open content performance. The table shows every article Fin uses, with two columns that matter: involved (how many times Fin referenced the article) and resolved (how many times that reference led to a resolved conversation).

Sort by involved conversations. The articles at the top are the ones Fin relies on most. If the resolved number is significantly lower than the involved number, that article is dragging your resolution rate down. Intercom's knowledge management guide recommends focusing on the top 20% of content by involvement and flagging anything that resolves below 90%.

Open the underperforming articles. The usual culprits: outdated screenshots, instructions referencing a renamed feature, steps that no longer match the product, or contradictory information spread across multiple articles on the same topic.

Review negative CX scores

Go to Fin AI Agent > Analyze > Performance and look at the CX Score breakdown. Negative CX scores on resolved conversations are the most useful signal. They mean Fin answered, the conversation closed, but the customer left unhappy.

Open those conversations. You will often find that Fin gave a technically correct answer to the wrong question, or pulled from an article that was accurate six months ago. Intercom's guidance is to drill into negative Fin CX conversations to find the root cause, then fix the content behind it.

Set up a feedback loop from your support team

Your agents see Fin fail before anyone else does. Intercom recommends creating a back-office ticket type that agents can submit directly from the inbox when they spot an unhelpful response. At Fin, support teammates surface 15 to 20 content improvements per week this way, each taking about 15 minutes to action.

You can also use the Improve Answer button in the inbox on any AI-generated response to review the content Fin used and enhance it, even on conversations that resolved successfully.

"I'm only one person, and we have almost a thousand articles. I know there's misinformation. There's outdated information."

— Knowledge Content manager, consumer app company

Keep content in sync with product releases

Your product ships a change, nobody updates the article, Fin keeps answering from the old version. This is where resolution rates decay.

A support specialist at a remote SaaS team we spoke to in August sees about 1,000 conversations a week, and Fin handles a large share of them. Her help center work gets roughly eight hours a week, around her chat shifts. When a big feature ships before its article, Fin can't find the answer and escalates questions the human team already knows how to answer.

Intercom recommends partnering with your product team so that every release includes a content step, including updated screenshots and navigation paths. Fin's own team ships 10 to 15 releases a week, which means creating and updating around 90 articles, and a significant release can take two to three hours of content work across internal and external sources.

That is a lot of articles to trace back to releases by hand. Pageloop picks up which articles a release affected from the signals your team already produces, and the next two sections cover where that fits.

"Senior managers are now realizing that the help center is more important because it's feeding Fin."

— Enablement lead, customer success platform

Is the Fin resolution rate Intercom reports accurate?

The resolution rate Intercom reports may be higher than your real resolve rate. In a community thread, a support manager at a hardware company described a real resolve rate of about 12%. Fin was marking conversations as "assumed resolved" when a human agent stepped in before the customer clicked "speak to human." The agent was stepping in because Fin had given the customer incorrect steps and they were mid-crisis.

Billing follows the same definition. Fin now charges per outcome, which is a resolution or a Procedure handoff, and an assumed resolution still counts as one.

One support lead we spoke to saw the same pattern. Her dashboard showed 84.5% resolution. She put the real number at roughly half that.

An Intercom support engineer in a separate thread said a good resolution rate starts at around 30 to 50%, depending on how well help center content is written. If your number is much higher than that and you have not invested heavily in your knowledge base, audit a sample of those "resolved" conversations and check whether the customers actually got the right answer.

Where Intercom's tools stop

Everything above lives inside Intercom. For teams in their first few months with Fin, it is enough to get from 30% to 50% or beyond.

Recommendations works from conversations that already failed. Connecting a failing article to the product change that broke it is a separate job. The Jira ticket that closed last sprint, the pull request that renamed a setting, or the Slack thread where support discussed the workaround three weeks ago all live outside Intercom.

A spreadsheet can bridge part of this. List your last five product releases alongside the features that changed, then search your knowledge base for articles that reference those features and check whether they are current. Three to four hours per release, and it catches the most obvious drift.

Pageloop is a docs-as-code knowledge base that hosts and publishes your help center, with every article versioned in Git through GitHub or GitLab. Pageloop's agent watches signals from code, project management, ticketing and support, then proposes a change set across every page a product change affected, with the evidence attached. Your team approves each change before it publishes. Automated screenshots identify outdated interface images in your articles, capture the new screens, and suggest the replacements.

Pageloop connects to Intercom, so Fin conversations reach the agent as a support signal. Book a demo to see how Pageloop fits your documentation.

Frequently asked questions

Why is my Intercom Fin resolution rate low?

The most common cause is content quality, not Fin configuration. Fin resolves questions by reading your help center articles. If those articles are outdated, incomplete, or structured in a way Fin cannot parse well, it either gives wrong answers or escalates to your team unnecessarily. Before adjusting Fin settings, check the articles it reads from in Recommendations and the content performance report. Look for articles that have not been updated since the feature they describe last changed.

What type of content does Fin need to resolve more questions?

Fin works best with articles that answer one question per article, start with the direct answer, use clear headings, and avoid relying on screenshots to convey instructions. Articles that cover multiple topics or use vague titles ("Getting Started," "Overview") are harder for Fin to match to specific customer questions. For guidance on structuring articles for AI chatbots, see How to Make Your Help Center AI-Readable.

How do I keep Fin's knowledge base accurate after product releases?

Fin reads from your Intercom articles and from any content you sync under Fin AI Agent > Train > Content, such as Confluence, Guru or Notion pages and public websites. When your product ships a change, Fin keeps pulling from the old version until someone updates it. Pageloop is a docs-as-code knowledge base whose agent reads product signals from GitHub or GitLab, Jira, Linear, Slack and support conversations, proposes the article updates a release requires, and refreshes outdated screenshots with automated screenshots, so the content Fin reads stays current after every release.

Image courtesy : Birmingham Museums Trust and Unsplash
Loch Awe at Sunset 1899, by Myles Birket Foster