Affiliate disclosure: This post contains affiliate links. If you buy through them, we may earn a commission at no extra cost to you. We only recommend tools we’d run our own business on.
Why the pricing model matters before you build anything
n8n Cloud bills per workflow execution — one complete run, no matter how many nodes it contains (checked October 2026). The workflows below are deliberately node-heavy: an enrichment pipeline might have 15 nodes. On a per-step tool, that’s expensive; on n8n it’s one execution.
| Plan (checked Oct 2026) | Price | Executions/month |
|---|---|---|
| Community (self-hosted) | Free | Unlimited (server-bound) |
| Cloud Starter | €20/mo annual (€24 monthly) | 2,500 |
| Cloud Pro | €50/mo annual (€60 monthly) | 10,000 |
| Cloud Business | ~€667/mo annual | 40,000 |
New to n8n? [AFFILIATE LINK: n8n] Cloud Starter’s 2,500 monthly executions cover every blueprint below while you validate — see our three-way cost math and self-hosting TCO breakdown.
AWL verdict: Build step-heavy on n8n; never build a 15-node workflow on a platform that bills per step. The execution model is the single biggest factor in whether these blueprints are profitable.
Blueprint 1: The lead enrichment pipeline
The problem. Raw leads arrive as names and emails. Sales ignores them — half are students, competitors, or dead addresses — and nobody enriches them, so they rot.
Node-by-node logic.
- Webhook trigger — receives the lead from your form or ad platform. Decision: accept a minimal payload (email, name, company) and validate the email format here, not later.
- Dedupe check — query your CRM for the contact email and the company domain. Decision: the same company submitting twice with different emails is one account, not two leads.
- Firecrawl scrape — pull the company’s homepage and about page into clean markdown. [AFFILIATE LINK: firecrawl] — our enrichment source of choice because it returns LLM-ready text from JS-heavy sites without us maintaining headless browsers.
- LLM scoring node — feed the scraped markdown plus the lead’s title into an LLM with a strict rubric: company size signals, tech stack mentions, buying-intent pages (pricing, careers). Decision: force a numeric score 0–100 and a one-line reason. Free-text verdicts are unactionable; scores route.
- IF/Switch branch — score ≥ 70 → create/update CRM record, assign owner, post to the sales Slack channel. 40–69 → nurture sequence. < 40 → archive with the reason attached.
- Stale-lead loop — any lead scored but untouched for 14 days re-enters for re-scoring. Decision: enrichment decays; a “hot” startup from March may be a dead domain by October.
What breaks in production. Firecrawl returns empty markdown on bot-protected sites — the scoring node scores a blank page as “0” and a potentially great lead gets archived. Fix: check character count first; under ~500 characters, retry once, then flag for manual review instead of auto-archiving. Also: pin the LLM model — scoring drifts between versions — and spot-check 20 scores a month against human judgment.
Blueprint 2: The content repurposing engine
The problem. One good post a week dies there, because turning it into the 8–10 derivatives it deserves takes a full day of human time.
Node-by-node logic.
- Schedule trigger — weekly, or a webhook fired when a post is marked “published” in your CMS.
- Fetch + chunk — pull the full article text. Decision: split into sections before sending to the LLM; one giant prompt produces generic mush, section-level prompts produce specific derivatives.
- Parallel generation branches — LinkedIn post, X thread, newsletter blurb, 5 hook variations, image brief for your designer. Decision: parallel branches with per-branch voice instructions, not one sequential mega-prompt.
- Quality gate (LLM-as-judge) — a cheaper model scores each derivative on two criteria: “contains a specific claim from the source?” and “sounds like the brand?” Decision: failures get one regeneration attempt, then go to a human review queue — never auto-published.
- Draft, don’t publish — write approved derivatives to your scheduler’s draft queue. Decision: the human presses “post.” Full auto-publishing is where brand damage happens.
What breaks in production. A vague judge passes everything — give it 3 pass and 3 fail examples in the prompt. Also: thin source content (a 400-word update) produces 10 garbage derivatives, so gate the engine on a minimum source word count.
Blueprint 3: Invoice chasing
The problem. Unpaid invoices are the most profitable automation target in any solo business: every hour of building it returns cash, not convenience.
Node-by-node logic.
- Schedule trigger — daily at 9am.
- Fetch overdue invoices — from your accounting tool. Decision: define “overdue” as due-date + 3-day grace, not the due date itself — chasing on day one trains clients to ignore you.
- Escalation ladder (Switch node) — 1–7 days: friendly reminder with the invoice link. 8–21 days: firmer tone, CC the accounts contact. 22+ days: pause service + final notice. Decision: tone lives in templates, routing in the workflow. Never let tone be decided by a node condition you can’t read at a glance.
- Payment detection loop — after each reminder, wait 72 hours and re-check status before escalating. Decision: this prevents the classic embarrassment of chasing a client who paid yesterday.
- Human handoff — at 30+ days, create a task with full history attached and stop the automation. Decision: automation chases; humans collect.
What breaks in production. Accounting API sync lag: the invoice shows “unpaid” but the client paid by bank transfer this morning. Fix: the 72-hour re-check plus a manual “paid” override field every reminder checks first. Also: schedule in the client’s timezone — nobody wants a final notice at 3am.
Blueprint 4: Support ticket triage
The problem. Every ticket lands in one inbox, and a human reads each one to decide if it’s urgent, who owns it, and whether it’s a duplicate. That’s 30–90 seconds of human time per ticket before any actual support happens.
Node-by-node logic.
- Webhook trigger — from your helpdesk (Intercom, Zendesk, or a shared inbox via Gmail trigger).
- Classification node (LLM) — category (billing, bug, how-to, feature request), urgency (P1–P3), sentiment. Decision: classification and urgency are separate outputs — an angry how-to question is not a P1, and a calmly-worded outage report is.
- Duplicate detector — embed the ticket text and compare against open tickets from the last 7 days. Decision: start the similarity threshold around 0.85 and tune it on your own data — too low merges unrelated tickets, too high misses real dupes.
- Router — P1 → page the on-call human immediately (Slack + SMS). P2/P3 → assign to the right queue with 3 suggested help-center articles as a draft reply. Feature requests → backlog board with an auto-reply.
- Draft replies, never send — for the first 30 days, the workflow drafts and a human approves. Decision: earn the right to auto-send with a month of approved drafts as evidence.
What breaks in production. The classifier confidently mislabels novel issues — a brand-new bug type gets filed as “how-to” and sits for days. Fix: attach a confidence score; below threshold, route to a human triage queue. And weight urgency keywords (“down,” “broken,” “can’t log in”) over sentiment scores, which are unreliable on non-native-English tickets.
Blueprint 5: Competitor monitoring
The problem. Competitors change pricing and launch features while you’re heads-down building. You find out months later from a lost deal.
Node-by-node logic.
- Schedule trigger — weekly for pricing pages, daily for blogs/changelogs.
- Firecrawl scrape — pull the target pages into markdown. [AFFILIATE LINK: firecrawl] Decision: scrape the pricing, features, and changelog URLs specifically — homepage hero copy changes are noise.
- Diff against last snapshot — store each scrape in a database; compare word-level. Decision: store the raw markdown, not a summary — you can’t diff a summary next week.
- Significance filter (LLM) — “did anything change that affects our positioning?” Decision: this node exists because 90% of page changes are typos, date updates, and cookie banners. Without it, alert fatigue kills the workflow in a month.
- Digest, not alerts — batch significant changes into a Monday briefing posted to Slack. Decision: one weekly digest gets read; five real-time pings get muted.
What breaks in production. Site redesigns change the page structure and the diff explodes with false “changes.” Fix: hash the structural fingerprint separately from the text; on a dramatic structure shift, flag “redesign detected” instead of reporting 400 text diffs. Also: some competitors block scrapers on pricing pages — fall back to their sitemap or a cached version rather than hammering a blocked URL.
FAQ
Q1: Do I need to know how to code to build these?
No. Every blueprint above is buildable in n8n’s visual editor with standard nodes — webhook, HTTP request, IF/Switch, LLM nodes, and your CRM’s integration. The hard part is the decisions (scoring rubrics, thresholds, escalation rules), not the syntax.
Q2: Which of these pays for itself fastest?
Invoice chasing, by a wide margin — it converts directly into recovered cash and is the simplest build of the five.
Q3: How much does it cost to run all five on n8n Cloud?
Count executions, not nodes: a solo founder running all five typically lands between 500–2,000 executions a month — comfortably inside Cloud Starter (€20/mo annual, 2,500 executions, checked October 2026). Your LLM API bill is separate and usually exceeds the n8n bill — budget for that first.
Q4: Why Firecrawl instead of a simple HTTP node?
Modern sites are JS-rendered and bot-protected; maintaining your own headless-browser setup is a second job. Firecrawl returns clean markdown via one API call — with the honest caveat that it’s a paid dependency with its own failure modes, which is why every blueprint above checks scrape quality first.
Q5: Should I self-host n8n instead of paying for Cloud?
If you’re technical and already run a VPS, the free Community edition is cheapest at any volume — our self-hosting TCO breakdown puts realistic server costs at $5–50/month. If you’d rather not own uptime and backups, Cloud Starter at €20/month is cheap insurance.
Methodology
n8n Cloud prices (Starter €20/mo annual, Pro €50/mo annual, Community free) were verified on October 8, 2026 via 2026 pricing trackers cross-checked against n8n’s published plans. Firecrawl’s affiliate terms came from their partner-program docs as cited by trackers in October 2026 — confirm in the partner portal before joining. Thresholds quoted are starting points from practice, not constants — tune them on your own data.
