How to Analyze Intercom Fin Conversations for Unresolved Questions
Fin's resolution rate includes assumed resolutions: customers who left without asking for more help. Here is how to export Fin conversations, label why each one failed and rank the questions to fix.
To find the questions Intercom's Fin fails to resolve, pull every conversation Fin was involved in with its resolution state and full transcript, label each one with what the customer needed and why Fin did not settle it, and then check the conversations marked "assumed resolved", because Intercom counts a customer who leaves without asking for more help as resolved. Rank what is left by how many unresolved conversations each question causes, not by Fin's resolution rate. For a team that runs Fin next to email, phone, surveys and reviews and needs that analysis on every conversation, joined to account data and shared as reports the support, product and content teams all use, Dimension Labs is the strongest choice.
Below: what Fin's metrics count, which Intercom tools show the gaps and where they stop, and the method step by step on sample data. For the general version that works on any bot, see how to find the questions your chatbot can't answer.
What does Fin's resolution rate count?
Intercom defines Fin's resolution rate as the share of Fin-involved conversations that Fin resolved, and "resolved" includes both confirmed and assumed resolutions. Two other rates sit next to it. Automation rate divides Fin's resolutions by all new conversations, and outcome rate divides billable outcomes (resolutions and Procedure handoffs) by all conversations in the period. Check which one a dashboard shows before comparing two of them.
From Intercom's help center
Fin records five resolution states, and two of them count as resolved
The values of the Fin AI Agent resolution state filter, as Intercom defines them, and whether each counts toward Fin's resolution rate.
| State | Intercom's definition | Counted as resolved? |
|---|---|---|
| Confirmed resolved | The customer replied with positive feedback after Fin's last answer | Yes |
| Assumed resolved | The customer left without asking to speak to a teammate or giving negative feedback | Yes |
| Escalated | The conversation was handed off to a teammate | No |
| Abandoned | The customer left before Fin could provide an answer | No |
| Pending | Waiting for the customer, or waiting for next steps after negative feedback | Not yet |
The assumed state is where unresolved questions hide. Intercom's billing FAQ says that a customer who is frustrated with Fin's answer and leaves without asking for more help "is counted as an Assumed Resolution and will be charged", at $0.99 per outcome. It suggests escalation rules that hand off when Fin detects frustration. The same page says a resolution is refunded if the customer comes back to the same conversation. A customer who gives up on chat and emails instead has opened a new conversation, so that one stays a resolution.
Some conversations never reach Fin's answer at all. Intercom calls a conversation "Fin constrained" when Fin is active but does not get to answer, for example because escalation guidance sends the topic to a person, a workflow fires first or a usage limit is reached. Those questions belong in the analysis too: some were routed away on purpose, and some only because nobody wrote the content.
Where does Intercom show the questions Fin fails on?
Intercom has five places to look, three of them in the Pro add-on.
From Intercom's help center
Intercom's own tools for finding what Fin misses, and their stated limits
Where each one lives and what Intercom says about it.
| Tool | What it shows | Limits Intercom states |
|---|---|---|
| Performance funnel Fin AI Agent > Analyze > Performance | Fin involved, constrained or not active; resolved, Procedure handoff, escalated (user requested vs rules and guidance), pending | Opens the conversations behind each path |
| Content performance Fin AI Agent report | Involved and resolved conversations per article or snippet | Totals only; the conversations cannot be opened from the table |
| Topics Explorer Pro add-on | AI topics and subtopics with CX Score, resolution rate and handling time | At least 15 questions to form a subtopic; reads teammate replies too; does not analyze sentiment |
| Recommendations Pro add-on | Content, customer data and action gaps, ranked by impact | Inbound conversations with at least two responses from Fin or a teammate |
| Fin Operator Pro add-on | Answers questions about metrics and analyzes conversation samples | 250 conversations per query; scoped to the Intercom workspace |
Intercom's pricing page lists Pro at "From $100 per workspace/mo" and says it includes analysis of 500 conversations a month and 6,000 credits. The Pro add-on article in its help center says "$99 per month for up to 1,000 conversations", with tiered billing above that. The two pages disagree; check the current figure with Intercom before you budget. Pro includes CX Score, Topics Explorer, weekly Trends, Recommendations, Monitors and Custom Scorecards.
For a team whose only question is which help articles Fin needs next, these tools are the simpler choice: Recommendations groups failures into content, customer data and action gaps and links each one to the conversations behind it. For a wider question, two details matter: Topics Explorer reads teammate replies and macros along with what the customer wrote, and assumed resolutions are counted as resolved throughout.
How do you analyze Fin conversations beyond the resolution rate?
Step 1: export Fin conversations with their resolution state
Intercom puts Fin's details on the Conversation model from API version 2.11: the source (such as the workflow name), the resolution state, the customer rating and the content Fin used. The Conversations Search endpoint can filter on them. Resolution state exists only for conversations after August 2, 2023. Which export carries the message text is covered in how to export Intercom conversations and find top contact reasons. Keep a customer or company ID that matches your other systems.
Step 2: split Fin's conversations into the groups that matter
Five groups need different treatment: escalations the customer asked for, escalations your rules or guidance triggered, abandoned conversations, assumed resolutions and constrained conversations. Intercom's funnel already splits escalations into user requested and rules and guidance. Escalations from your own rules are routing decisions; the customer-requested ones and the abandoned ones are where Fin's answer fell short.
Step 3: label what the customer needed and why Fin did not resolve it
Give every conversation two labels from fixed lists. The first is the customer's need, from your own topic list. The second is the reason: no content for it, content that exists but did not match, missing account data, an action Fin could not take, a person required by policy, or an answer that was wrong. Label only what the customer wrote, not the teammate's reply. Fixed lists keep next month's counts comparable with this month's; topic modeling vs LLM classification compares that with discovered topics.
Step 4: check the assumed resolutions
Score every assumed resolution against a short rubric (truly resolved, partly resolved, left after a wrong answer, came back on another channel within seven days) and hand-read a random sample to confirm the scoring. A sample of 300 gives a margin of about five points on the share that fails. How to check whether your AI agent really resolved conversations has the rubric and the sample sizes.
Sample data, fictional software company
Fin reported 63.3% resolved; after checking the assumed resolutions, about 49.8% were
18,000 Fin-involved conversations in one month, by resolution state. The assumed resolutions are split by whether the check found them resolved.
In the sample month, 2,430 of 8,300 assumed resolutions (29.3%) were not resolved on inspection, which takes the resolution rate from 63.3% to 49.8%. At $0.99 an outcome, those conversations were billed at about $2,406. These are invented figures; the shape is what to look for in your own data.
Step 5: rank questions by unresolved conversations
Add escalations, abandonments and the assumed resolutions that failed the check, per topic. That count, not the topic's resolution rate, says where the next fix pays off, because a topic with a middling rate and high volume can leave more customers without an answer than a topic with a poor rate and low volume.
Sample data, fictional software company
Exports looked like one of Fin's best topics, and it is the second-largest source of unresolved questions
Seven topics ranked two ways. Left: Fin's reported resolution rate, lowest first. Right: unresolved conversations, counting escalations, abandonments and assumed resolutions the check did not confirm.
In the sample, exports and reports showed a 67.4% resolution rate, second best of the seven. 854 of its 1,220 assumed resolutions failed the check, mostly customers who left after a partial answer about a report Fin could not build. It moves to second on unresolved conversations. Plan changes had the lowest rate, 38.4%, but much of that was escalations the company wants: cancellations go to a person by design.
Which Fin metrics should you report beyond the resolution rate?
Checked resolution rate, with its range, next to the rate Intercom reports.
Unresolved conversations by topic and reason, so each item has an owner: content, engineering for data and actions, or policy.
Repeat contact within seven days on chat, email or phone after a Fin "resolution".
Escalations by trigger: customer request, frustration, your rules, failed Procedures.
What unresolved questions cost: handled-contact cost for escalations, and, with account data joined, renewal and expansion for the accounts behind them.
The last one needs data from outside Intercom. Joining conversation labels to CRM and revenue data is on the Dimension Labs Prove plan ($2,999 a month) and up. For a monthly view of all of this for leadership, see what a monthly customer insight report should contain.
What does Dimension Labs add to Intercom's Fin reporting?
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.
Fully omni-channel: Fin chats, teammate conversations, email, calls, surveys and reviews are labelled with the same Dimensions, so a question Fin dropped is visible where the customer went next.
Easy integration: Intercom connects in one click and brings in new conversations, with their messages, timestamps and customer attributes, enriched every day. Earlier history, and Fin fields such as the resolution state, come in as a file from your Intercom export, and account data joins on a shared ID.
Effortless enrichment: every conversation gets the need, the failure reason and the checked outcome as Dimensions built for your product, usable by support, product and content teams. Custom Dimensions start on the Discover plan.
A new visual analytics layer: a gap report for the content team, a deck for the head of support and a dashboard of unresolved questions by topic, built by the agent and shared as a link you can revoke.
Intuit QuickBooks used Dimension Labs on more than one million chatbot sessions a month across 50+ bots and cut human escalations by 57% and its fallback rate by 35.3%. Dropbox reported a 43% reduction in unhandled bot conversations.
FAQ
How do I see which questions Intercom Fin fails to resolve?
Inside Intercom, start with the performance funnel and open the escalated and abandoned paths; with Pro, Recommendations and Topics Explorer group them. For a complete list, export the conversations with their resolution state, label each one's need and failure reason, and check the assumed resolutions, which Intercom counts as resolved.
Does Fin count a customer who just leaves as resolved?
Yes, if Fin gave an answer and the customer left without asking for more help or giving negative feedback. Intercom calls that an assumed resolution and bills it as an outcome. If Fin's last message was a clarifying question, no resolution is recorded.
Is Intercom's Pro add-on enough for Fin analytics?
For fixing Fin's content inside Intercom, often yes. When the question also needs other channels, account or revenue data, or a check on whether assumed resolutions held, plan to bring those conversations together with that data in one analysis.
What is the best tool for Fin analytics beyond the resolution rate?
For teams that need every Fin conversation labelled with a need and a failure reason, the assumed resolutions checked, and the results joined to other channels and account data in reports several teams use, Dimension Labs is the strongest choice. For content fixes alone, Intercom's own Recommendations is the simpler option.
See the Intercom integration, or start with one export of Fin conversations. Every self-serve plan starts with a 14-day free trial.