n8n Intercom Automation: Score Tickets and Catch Churn

A customer opens their third support conversation in two weeks. Tone's getting sharper each time. Meanwhile their last invoice failed and their seat logins dropped by half. Support sees the tickets. Billing sees the failed charge. Nobody sees all three at once — so the account cancels, and everyone's surprised.

That blind spot is what n8n Intercom automation can close. Most Intercom build guides stop at "AI replies to the chat." Useful, but it answers the wrong question. The harder, more valuable job is reading the conversation as one signal among several, scoring the account's health, and escalating the ones quietly heading for the exit.

n8n suits this because it can hold signals from three systems in one workflow. An Intercom webhook, a billing event, a usage number — they all land in the same execution, get scored together, and route to Slack or an email before the renewal date sneaks up.

What you can automate

• Score every Intercom conversation for sentiment and frustration level
• Combine support tone with billing health and usage signals into one account score
• Classify accounts into health tiers (healthy, watch, at-risk, critical)
• Fire an instant Slack alert for critical accounts with the MRR at risk
• Draft a personalized retention email for at-risk accounts, ready to send
• Tag the Intercom user and post an internal note summarizing the risk
• Log every event to Google Sheets for churn attribution and trend analysis

The conversation-health pipeline

Intercom Webhook (new/updated conversation)
  → OpenAI (score sentiment + frustration, return JSON)
  → Code (merge support score + billing + usage → 0-100 health)
  → Switch (route by tier)
      ├─ critical → Slack alert (MRR at risk + factors)
      ├─ at-risk  → AI retention email draft → send
      └─ healthy  → tag + log only
  → Google Sheets (append score, tier, factors)

The interesting node is the merge. One conversation's sentiment is noise. Sentiment plus a failed payment plus a usage drop is a signal you can act on. The n8n-nodes-base.code v2 node is where those three numbers become one tier.

1. Catch the conversation

Intercom doesn't have a polling-friendly model the way a database does, so use a n8n-nodes-base.webhook node and point an Intercom webhook subscription at it. The webhook delivers the conversation body, the user ID, and the event type. That user ID is the join key. It's how you'll pull the matching billing and usage records.

2. Score the support signal

Feed the conversation text to the @n8n/n8n-nodes-langchain.openAi v2.1 node. Ask for a strict JSON object: sentiment, a frustration score, and a one-line reason. The node returns its output at $json.output[0].content[0].text. Worth repeating because it's the most common copy-paste failure: tutorials written against the old n8n-nodes-base.openai node read $json.choices[0].message.content, and that path returns undefined on the current node.

3. Merge into a health score

This is the step the AI-support-agent tutorials skip. Add a n8n-nodes-base.code v2 node with jsCode. Pull the billing status and usage trend for that user (a Postgres lookup, an HTTP call to your billing API, or a Google Sheets read), then compute a 0-100 health score from weighted categories: billing health, support volume, login frequency, sentiment. Classify into tiers: critical under 25, at-risk 25-49, watch 50-79, healthy 80 and up.

Do the parse and the merge in the same Code node every time, even when one signal is missing. A missing billing record should default to a known value, not crash the score. The execution log makes a thrown error obvious; a silently skipped signal does not.

4. Route and act

A Switch node reads the tier. Critical accounts trigger an immediate Slack alert showing the customer, the MRR at risk, the event, and every contributing factor, so a CSM acts before the renewal. At-risk accounts get a personalized retention email drafted by a second LLM call, tuned to the specific risk profile rather than a generic "we miss you."

A strong opinion that working CS teams will recognize: tier the alerts, don't alert on everything. If every neutral conversation pings Slack, the channel becomes wallpaper and people mute it. Reserve the loud alert for critical, let at-risk handle itself with email, and log the rest silently. Alert fatigue kills more retention plays than slow tooling does.

5. Tag and log

Post an internal note back to the Intercom conversation with the score and the reason, so the next agent has context. Tag the user with the tier. Then append the whole event to Google Sheets. That log is your churn-attribution dataset later.

n8n Intercom automation patterns

Three patterns cover the work once the health-scoring spine is in place.

Pattern 1: Cross-signal escalation

The core pattern. Three inputs, one score, one routing decision.

Webhook (Intercom) → OpenAI (sentiment)
  → Code (fetch billing + usage by user_id, compute 0-100)
  → Switch (tier) → Slack (critical) / SendGrid (at-risk email) / log (rest)

Pattern 2: Spend-spike billing alert

A separate workflow listens for billing webhooks (failed payment, downgrade, refund). When one fires for an account that also has open Intercom conversations, it jumps the tier straight to critical. Billing and support rarely talk; this makes them.

Pattern 3: Weekly health digest

A scheduleTrigger v1.2 runs every Monday, reads the week's logged scores from Google Sheets, and posts a digest of accounts that slid down a tier. The real-time path catches fires; the digest catches slow drifts the alerts didn't trip.

n8n nodes you'll use most

Getting started with n8n Intercom automation

1. Create an Intercom access token in your developer settings and add the credential in n8n.
2. Add a n8n-nodes-base.webhook node and subscribe an Intercom webhook to its URL.
3. Wire the @n8n/n8n-nodes-langchain.openAi v2.1 node with a JSON-only sentiment prompt.
4. Build the n8n-nodes-base.code v2 merge node — start with billing plus sentiment, add usage later.
5. Add a Switch on the tier and build the critical Slack alert first.
6. Add the at-risk retention email and the Google Sheets log.
7. Replay a week of real conversations through it before trusting the tiers.

If you'd rather not hand-build the scoring algorithm and the tier thresholds, the Customer Health Score & Retention Engine ships exactly this cross-signal brain.

Scoring conversations is one queue; routing them across teams is another. The walkthroughs in n8n support ticket triage and n8n support escalation routing cover the routing side in depth, and the Freemium User Retention Campaign picks up the at-risk accounts this workflow flags and runs the win-back sequence.