The fastest way to improve your Fin CX Score is to identify which signal is driving it down, then fix the root cause not just the symptom. Start by drilling into the Negative Fin CX section under Fin AI Agent > Analyze > Performance to find your biggest driver:
Fin answer quality
User effort
User emotion
Policy feedback
Product feedback
Each section below explains what each reason means, how to confirm the pattern, and the specific steps to fix it.
How to fix low Fin answer quality
Fin's responses are unclear, inaccurate, or fail to resolve the issue — customers aren't getting the answer they need. The root cause is almost always missing, outdated, or poorly structured content.
Use a three-step content audit to find and fix the gaps driving poor answer quality:
Identify the gaps: Click into the Low Fin answer quality reason and sample 20–50 conversations flagged under this reason to identify whether the issue is missing content, unclear content, or conflicting/out-of-date content — keep a tally as you go.
Share your findings with your content owner and prioritize fixes based on conversation volume — the topics with the highest Fin involvement and lowest resolution rate will have the biggest impact on your CX Score.
Fix and fill: Use AI-powered recommendations to action quick improvements: add missing information, resolve conflicting instructions, and remove duplicates to keep Fin's answers helpful and consistent.
Tip: Set up a monitor that flags conversations where Fin's answer quality may have fallen short. For example, where the customer asked the same question multiple times, expressed that Fin's answer didn't help, or where the conversation ended without a resolution. Pair it with a Fin Quality Scorecard that evaluates accuracy, completeness, and relevance. Reviewing flagged conversations weekly helps you catch recurring content gaps before they accumulate.
Learn more:
How to reduce user effort
Customers have to repeat themselves, ask multiple follow-up questions, or go through excessive back-and-forth with Fin even if the conversation eventually ends in a resolution. This drives down your Fin CX Score even when Fin is technically answering correctly.
First, distinguish real friction from normal conversation behavior. Click into the User Effort reason and sample 20–50 conversations. Ask: is the customer repeating themselves? Are they asking the same question multiple ways? Are there 5+ messages where 2–3 would have sufficed? If yes, that's real friction worth fixing.
Tip: Set up two monitors to track user effort systematically. The first should flag conversations where Fin provides substantially similar responses across multiple turns without making progress (a reliable sign of looping). The second should use a Customer Effort scorecard to score how much work the customer had to do to get their answer. Using them together helps you separate true looping behavior from conversations that were long but linear. Prebuilt scorecard templates include a Customer Effort Score you can start from.
Once you've confirmed the pattern, here's how to reduce effort:
Add guidance for looping conversations: If customers ask the same question in multiple ways, add guidance telling Fin to proactively offer a human handover after 2 attempts — don't make the customer explicitly ask for one.
Simplify your escalation path: Review your escalation rules and make sure Fin knows exactly when and how to hand off. For example, if you haven't trained Fin on how to handle a certain topic yet (with content or data connectors), you can ask it to always escalate these conversations for a human to handle.
Build a Procedure for high-effort query types: For complex queries that require collecting information (e.g. troubleshooting, account lookups, refund checks), build a Fin Procedure to handle them end-to-end. Procedures let Fin access external data and collect the right context upfront — turning a 10-message conversation into a 3-message one.
Learn more:
How to address user emotion
Customers express frustration, anger, or dissatisfaction during the conversation — even if their issue is eventually resolved. This is a signal that how Fin is responding is landing badly, not necessarily what it's responding with. The root cause is usually missing or poorly tuned guidance.
Click into the User Emotion reason to review 20–30 conversations and look for patterns: are customers getting frustrated on specific topics (e.g. billing, cancellations, account limits)? Are they receiving technically correct but unsympathetic responses?
Tip: Set up a monitor that flags conversations where customers express frustration, anger, or dissatisfaction. For example, using flag criteria that detect emotionally charged language or repeated expressions of disappointment. Pair it with a scorecard that evaluates empathy, tone, and whether Fin acknowledged the customer's frustration before responding. This turns a subjective signal into a measurable, reviewable pattern you can act on.
Once you've identified the pattern, use one or both of these approaches:
Add communication guidance: Tell Fin how to handle emotionally charged query types. For example: "When a customer raises a billing dispute or account closure, acknowledge their frustration before explaining the process. Avoid restating policy without empathy." You can also instruct Fin to use softer language on specific topics.
Add escalation guidance using Fin Attributes: Create a Fin Attribute that tracks customer sentiment or query sensitivity (e.g. 'Billing', 'Cancellation', 'Complaint'). Then set an escalation rule so that when Fin detects a sensitive topic combined with strong negative emotion, it automatically acknowledges the impact, explains the situation once, and offers to route to a human — rather than deflecting.
The goal is to ensure customers never feel dismissed or stuck.
How to respond to product feedback
Customers are hitting limitations or bugs in your product — not issues with Fin's knowledge or behavior. Fin can't fix product bugs, but it can be the most reliable signal you have for surfacing them to the right team. Product teams rely on support data to prioritize fixes, and your Fin CX Score is one of the highest-signal sources available.
Make product feedback from Fin conversations impossible to miss or lose. Set up a monitor that flags conversations where customers report product bugs, limitations, or unexpected behavior. Pair it with a scorecard that evaluates whether Fin acknowledged the limitation clearly, set appropriate expectations, and offered a useful next step (such as routing to a teammate or providing a workaround). This gives you both a reliable feed for your product team and a quality check on how Fin handles these moments.
To make this feedback actionable:
Sample weekly: Review 10–15 conversations each week. Look for clusters — the same product area, friction point, or limitation appearing across multiple customers is a signal worth escalating to your product team.
Be the voice of the customer: Document a lightweight format for sharing these conversations with your product team (e.g. a Slack message or ticket with: product feedback/limitation, frequency, 3 example conversations). Consistency makes it easier for product to act on.
Track whether it's improving: Monitor the product feedback volume over time. If a product area is improved or a product gap is addressed, you should see this reason decrease. If it doesn't, re-investigate.
Learn more:
How to handle policy feedback
Negative policy feedback reflects how customers are reacting to your company's policies — such as refunds, returns, account rules, limits, or eligibility requirements. Fin's CX Score drops in these cases not because Fin is wrong, but because the policy itself creates friction.
Go to Fin AI Agent > Analyze > Topics Explorer and filter by CX Score where the reason is "Negative policy feedback". Sample around 20 conversations in the topic(s) with the biggest volume to pinpoint where policy is causing friction, then add content or guidance to improve how Fin communicates and handles it:
Make sure content clearly explains what the policy is, why it exists, and what options the customer still has (next-best alternatives, partial solutions, or workarounds).
Add guidance for Fin to use empathetic, plain language when delivering bad news (e.g. refund refusals, eligibility denials).
These changes turn "policy wall" moments into clearer, more considerate experiences, which directly improves CX Score even when the policy itself doesn't change.
You can also use escalation guidance so that when Fin detects a policy constraint + strong negative emotion, it:
Acknowledges the impact,
Explains the limit once, and
Offers escalation or review (e.g. route to billing/success for exceptions or credits where your business rules allow).
Even if the final decision is unchanged, the customer experience of how they got that answer usually improves.
You should also close the loop on policies that need to change. Set up a workflow to tag conversations where the CX Score reason is negative Policy feedback, and send them to the right ops/product owners as a queue of "policy pain" conversations.
Use that evidence to:
Simplify or relax policies that are out of step with your goals, or
Make them more predictable and transparent (e.g. clearer in pricing pages, sign-up flows, and lifecycle emails).
Learn more:
Tip: Set up a monitor that flags conversations where customers push back on a policy. For example, refund refusals, eligibility denials, or account limit explanations. Pair it with a scorecard that evaluates whether Fin explained the policy clearly, used empathetic language, and offered the customer a meaningful next step. Over time, this lets you track whether your content and guidance improvements are actually reducing policy friction.
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts