How to Automate Customer Feedback Analysis with n8n

By The LightCube Labs team

What You Will Need

• A self-hosted n8n instance (Community Edition is free) or an n8n Cloud trial
• Zendesk (or Intercom) API credentials with permission to search tickets
• A Google account with a Google Sheet that has columns date, ticket_id, subject, category, sentiment
• A Slack app with the chat:write scope, invited to #product-feedback
• Estimated setup time: 45 to 60 minutes

How the n8n Workflow Fits Together

This n8n workflow has six nodes in a straight line, so it is easy to debug in the n8n editor. A Schedule Trigger starts the run, an HTTP Request node collects tickets, two Code nodes do the analysis, a Google Sheets node keeps the history, and a Slack node delivers the result. Every analysis rule lives in plain JavaScript inside n8n, which means you can read, test and change it without installing a plugin or paying for a classification service.

The design choice worth noting is that n8n never calls a text analysis API. Categorization is a keyword lookup and sentiment is a weighted word list. That keeps the n8n workflow fast, free of per-request cost, and private, because ticket text stays on your n8n server.

Step-by-Step Workflow Setup

Node: Schedule Trigger
Trigger Interval: Days
Days Between Triggers: 1
Trigger at Hour: 9
Trigger at Minute: 0
Timezone: set in Workflow Settings (for example America/Los_Angeles)

Expression: {{ $now.minus({ hours: 24 }).toUTC().toISO() }}

{
  "timestamp": "2026-11-04T09:00:00.000-08:00",
  "Readable date": "November 4th 2026, 9:00:00 am",
  "Day of week": "Wednesday"
}

Node: HTTP Request
Method: GET
URL: https://YOURDOMAIN.zendesk.com/api/v2/search.json
Authentication: Generic Credential Type > Basic Auth
Send Query Parameters: On
  query: type:ticket created>{{ $now.minus({ hours: 24 }).toFormat("yyyy-MM-dd'T'HH:mm:ss'Z'") }}
  sort_by: created_at
  sort_order: asc
Options > Pagination: Response Contains Next URL
  Next URL: {{ $response.body.next_page }}

Expression: {{ $response.body.next_page }}

{
  "results": [
    {
      "id": 48213,
      "subject": "Export to CSV keeps timing out",
      "description": "Every time I export more than 5k rows the page hangs and then crashes. Very slow.",
      "created_at": "2026-11-03T21:14:09Z",
      "requester_id": 90211,
      "tags": ["export", "enterprise"]
    }
  ],
  "next_page": null,
  "count": 1
}

Node: Code
Mode: Run Once for All Items
Language: JavaScript

const RULES = {
  bug: ['error','crash','broken','not working','fails','bug','exception'],
  'feature-request': ['wish','would be great','feature','please add','support for','missing'],
  pricing: ['price','pricing','invoice','refund','billing','expensive','discount'],
  UX: ['confusing','hard to find','unclear','layout','navigation','design'],
  performance: ['slow','timeout','timing out','lag','loading','hangs']
};
const out = [];
const rows = $input.all().flatMap(i => i.json.results ? i.json.results : [i.json]);
for (const item of rows) {
  const text = ((item.subject || '') + ' ' + (item.description || '')).toLowerCase();
  let best = 'uncategorized', bestHits = 0;
  for (const [cat, words] of Object.entries(RULES)) {
    const hits = words.filter(w => text.includes(w)).length;
    if (hits > bestHits) { best = cat; bestHits = hits; }
  }
  out.push({ json: { id: item.id, subject: item.subject, category: best, text } });
}
return out;

Expression: {{ $json.category }}

{
  "id": 48213,
  "subject": "Export to CSV keeps timing out",
  "category": "performance",
  "text": "export to csv keeps timing out every time i export more than 5k rows the page hangs and then crashes. very slow."
}

Node: Code (Sentiment)
Mode: Run Once for All Items
Language: JavaScript

const POS = { thanks: 1, great: 2, love: 2, helpful: 1, works: 1, awesome: 2 };
const NEG = { slow: -1, crash: -2, broken: -2, hate: -3, frustrating: -2, useless: -3, confusing: -1 };
const BOOST = ['very','extremely','constantly','always','totally'];
return $input.all().map(item => {
  const tokens = item.json.text.split(/[^a-z']+/);
  let score = 0;
  tokens.forEach((t, i) => {
    const w = POS[t] || NEG[t] || 0;
    const mult = BOOST.includes(tokens[i - 1]) ? 1.5 : 1;
    score += w * mult;
  });
  const sentiment = Math.max(-1, Math.min(1, score / 5));
  return { json: { ...item.json, sentiment: Number(sentiment.toFixed(2)) } };
});

Expression: {{ $json.sentiment < -0.3 ? "negative" : $json.sentiment > 0.3 ? "positive" : "neutral" }}

{
  "id": 48213,
  "category": "performance",
  "sentiment": -0.6
}

Node: Google Sheets
Resource: Sheet Within Document
Operation: Append Row
Document: By URL (paste your sheet URL)
Sheet: Tickets
Mapping Column Mode: Map Each Column Manually
  date: {{ $now.toFormat("yyyy-MM-dd") }}
  ticket_id: {{ $json.id }}
  subject: {{ $json.subject }}
  category: {{ $json.category }}
  sentiment: {{ $json.sentiment }}

Expression: {{ $now.toFormat("yyyy-MM-dd") }}

{
  "date": "2026-11-04",
  "ticket_id": 48213,
  "subject": "Export to CSV keeps timing out",
  "category": "performance",
  "sentiment": -0.6
}

Node: Code (Summary)
Mode: Run Once for All Items
Language: JavaScript

const items = $input.all().map(i => i.json);
const counts = {};
items.forEach(t => { counts[t.category] = (counts[t.category] || 0) + 1; });
const avg = items.reduce((s, t) => s + t.sentiment, 0) / (items.length || 1);
const worst = [...items].sort((a, b) => a.sentiment - b.sentiment).slice(0, 3);
const store = $getWorkflowStaticData('global');
const prev = typeof store.lastAvg === 'number' ? store.lastAvg : avg;
store.lastAvg = avg;
const delta = avg - prev;
return [{ json: { total: items.length, counts, avg: Number(avg.toFixed(2)),
  delta: Number(delta.toFixed(2)), worst: worst.map(t => t.subject) } }];

Expression: {{ $json.delta > 0 ? "improving" : "worsening" }}

{
  "total": 37,
  "counts": { "bug": 9, "feature-request": 11, "pricing": 4, "UX": 6, "performance": 7 },
  "avg": -0.12,
  "delta": -0.08,
  "worst": ["Export to CSV keeps timing out", "Billing page crashes", "Cannot find SSO settings"]
}

Node: Slack
Resource: Message
Operation: Send
Send Message To: Channel
Channel: #product-feedback
Message Type: Simple Text Message
Text:
*Feedback digest for {{ $now.toFormat("MMM d") }}*
Tickets: {{ $json.total }} | Avg sentiment: {{ $json.avg }} ({{ $json.delta > 0 ? "improving" : "worsening" }})
Bug: {{ $json.counts.bug || 0 }} | Feature: {{ $json.counts["feature-request"] || 0 }} | Pricing: {{ $json.counts.pricing || 0 }}
UX: {{ $json.counts.UX || 0 }} | Performance: {{ $json.counts.performance || 0 }}
Top issues: {{ $json.worst.join("; ") }}

Expression: {{ $json.worst.join("; ") }}

{
  "ok": true,
  "channel": "C04PRODFB",
  "ts": "1793722800.000100",
  "message": { "text": "*Feedback digest for Nov 4*..." }
}

Extending This n8n Workflow

Once the base n8n workflow runs cleanly for a week, there are three upgrades that pay off quickly. First, add an IF node after the sentiment Code node so that any ticket scoring below -0.7 triggers its own immediate Slack alert, while everything else waits for the daily digest. Second, add a weekly n8n workflow that reads the Tickets sheet with the Google Sheets Get Row(s) operation and charts category counts by week. Third, tune the keyword lists using the uncategorized rows, since those are the words your customers use that your rules do not yet know.

If you outgrow the lexicon, n8n can call a hosted LLM API through the HTTP Request node, but keep the keyword engine as a fallback so the workflow still runs when that API is unavailable.

Frequently Asked Questions

How accurate is keyword sentiment compared with a trained model?

It is less accurate on individual tickets but very stable over time, which is what matters for a trend line. The same input always gets the same score, so a sudden drop reflects a real change in wording across tickets rather than model drift.

Can I use Intercom or Freshdesk instead of Zendesk?

Yes. Only the n8n HTTP Request node changes. Intercom uses the conversations search endpoint with a created_at filter, and Freshdesk uses /api/v2/search/tickets with a query string. The Code nodes only need subject and description fields.

How do I add a sixth category such as security?

Add a new key to the RULES object, for example security: ['breach','password','sso','permission'], and add the new key to the Slack message. No other node needs a change because counts are built dynamically.

Does this send customer data to a third party?

Ticket text is processed inside your own n8n instance. Only the summary goes to Slack and the row data goes to your own Google Sheet, so no external classification service ever sees the raw messages.