n8n workflow templates for knowledge work: ten patterns, what each needs and what a run costs

Before the templates: how n8n counts, and where it runs

n8n bills by execution. The pricing page defines it: "An execution is a single run of your entire workflow. It doesn't matter how many steps are in the workflow or how much data it processes."^[1] That single fact shapes every pattern below. A workflow that loops over 200 calendar events inside one run costs one execution. A workflow that fires once per incoming email costs one execution per email. So the design question for cost is never "how many nodes" but "how often does the trigger fire".

n8n Cloud plans read on 2026-11-09, all billed annually (the page offers a monthly toggle that says annual saves 17%; monthly prices were not shown in the text we could read): Starter EUR 20 a month for 2,500 executions and 5 concurrent executions; Pro EUR 50 for 10,000 executions and 20 to 50 concurrent; Business EUR 667 for 40,000 executions, self-hosted only at the moment; Enterprise on request. All plans include unlimited users, workflows and steps. Overage is listed as "capped" for Starter and Pro with no per-unit price shown. Starter and Pro offer a free trial with 1,000 executions and no credit card. Hosted data is stored in the EU on servers in Frankfurt.^[1] A

Self-hosting: "Without a license key, n8n runs as the free Community edition", installed with Docker Compose or a one-line setup on your own server; a Business or Enterprise key unlocks those editions on the same core.^[2] The Community edition is published under n8n's Sustainable Use License, which permits internal business use and restricts reselling it as a hosted service.^[3] A small cloud server from a European provider costs under EUR 7 a month excluding VAT.^[4] The trade is your hours for updates, backups and the credentials store. Our comparison of Zapier, Make and n8n prices the same workflow at three volumes across all three platforms; this page assumes you have chosen n8n and want to know what to build.

A note on polling. Several patterns below use a trigger that checks a source on a schedule. We assume for the cost estimates that each scheduled check counts as an execution whether or not it finds new data; n8n's public pages we read do not state how empty polls are counted, so confirm it in the pricing FAQ before you pick a five-minute schedule. Where the source application offers a webhook, use it: it fires only when something happens.

Ten template patterns

Each row gives the trigger, the node chain, the output, the credentials you must create, where it belongs, and the monthly execution count under a stated assumption. Execution counts are arithmetic from the trigger frequency, not measurements. We have not yet published run logs from our own instance; when we do, this table gains a column.

#	Pattern	Trigger → nodes → output	Credentials	Cloud or self-host	Executions per month (assumption)
1	Inbox triage to drafts	Gmail or IMAP trigger → Filter (skip newsletters) → model node classifies and drafts → Gmail "create draft" → Slack notice	Mail OAuth or app password; model API key; Slack	Either; self-host if mail is confidential	Hourly poll: about 720; 15-minute poll: about 2,900
2	Meeting notes to tasks	Webhook from the note-taker or calendar → model node extracts action items → Notion, Asana or ClickUp "create task" → Slack summary to the organiser	Note-taker webhook secret; task tool OAuth; model key	Either	One per meeting: a team with 60 meetings a month pays 60
3	Weekly research digest	Schedule (weekly) → several RSS Read nodes → Merge → Remove Duplicates → model node summarises → Email or Notion page	Model key; mail or Notion	Cloud is fine	4 to 5
4	Form to CRM to chat	Webhook (form) → IF (drop test entries) → CRM "create or update" → Slack post → Google Sheets append	Form webhook; CRM OAuth; Slack; Sheets	Cloud	One per submission
5	Daily report	Schedule (weekdays 08:00) → HTTP Request or Postgres → Aggregate → model node writes three lines → Slack or email	Database or API credentials; Slack	Self-host if the database is internal	About 22
6	Document intake	Google Drive trigger (new file in folder) → Extract from File (PDF text) → model node returns structured fields → Sheets row → Drive "move file"	Drive OAuth; model key; Sheets	Either	One per document; polling interval sets a floor
7	Ticket triage with a gate	Webhook (helpdesk) → model node proposes category and priority → Slack message with approve/change → Wait (resume on webhook) → helpdesk update	Helpdesk API; Slack app with interactivity; model key	Either	One per ticket; the wait does not add executions
8	Knowledge-base answers in chat	Slack trigger (mention) → AI Agent node with a vector-store tool and a memory node → Slack reply	Slack; model key; vector store (for example Qdrant, Pinecone or Supabase); embeddings key	Self-host if the knowledge base is confidential	One per question
9	Morning meeting brief	Schedule (07:00) → Google Calendar "get events" → Loop over items → CRM or Notion lookup → model node drafts a brief per meeting → single email	Calendar OAuth; CRM or Notion; model key	Either	About 22; the loop runs inside one execution
10	Error handler	Error Trigger → Set (format message) → Slack or email → optional Sheets row for the log	Slack or mail; Sheets	Same instance as the workflows it watches	One per failure

What the table says about plans. A team running patterns 2, 3, 4, 5, 9 and 10 on modest volumes stays under the Starter plan's 2,500 executions. Pattern 1 at a 15-minute poll alone consumes most of Starter, and pattern 7 or 8 at a busy helpdesk moves you to Pro. At Pro, 10,000 executions cost EUR 50, which is EUR 0.005 per execution; the model provider's token bill for the AI steps is separate and, for patterns 1, 6, 7 and 8, usually larger than the n8n bill.

Notes on each pattern

1. Inbox triage. The classifier should output a label from a fixed list, not free text; feed that into a Switch node. Create drafts, never send: the section on adding AI steps safely below explains why. Watch for loops: a draft reply that the trigger sees as a new message. Filter on labels or sender.

2. Meeting notes to tasks. The weakest step is the extraction. Ask the model for a JSON array with owner, task and due date, validate it with a Code node, and route anything malformed to the error handler. The organiser confirms the tasks before they appear in the team's board; a task created by a model and never seen by a person is a task nobody owns.

3. Research digest. Cheapest pattern on the list and the best first build. Key the Remove Duplicates node on the article URL, and keep the source list in a Set node so a colleague can change it without touching the rest.

4. Form to CRM. The reference workflow from our platform comparison. Add an idempotency check: look up the email address before creating the contact, or the second submission creates a duplicate.

5. Daily report. A self-hosted instance in the same network as the database avoids exposing it. If you use Cloud, give it a read-only database user.

6. Document intake. Extract from File handles PDFs with a text layer; scanned images need an OCR service first. Test with the ugliest invoice you have, not the cleanest.

7. Ticket triage with a gate. The Wait node pauses the execution until the Slack button is pressed; the gate is a real decision only if the reviewer sometimes changes the proposal. Log both the proposal and the decision so you can measure agreement rate.

8. Knowledge-base answers. The AI Agent node calls a vector store as a tool; the documents must be embedded first by a separate ingestion workflow (Drive trigger → text splitter → embeddings → vector store insert). Quote the source document in every answer so the reader can check.

9. Morning brief. Everything runs inside one execution, so it is cheap. The CRM lookup is where it breaks: name matching between a calendar invite and a CRM contact is approximate; show the match and let the reader judge.

10. Error handler. Build this one first. n8n's documentation: "For each workflow, you can set an error workflow in Workflow Settings. It runs if an execution fails", and the error workflow "must start with the Error Trigger" node; the same error workflow can serve several workflows. The trigger receives the execution id, the error message and stack, the last node executed, and the workflow id and name.^[5]

Adding AI steps safely

A model node turns a deterministic workflow into a probabilistic one. Three rules keep that manageable.

Constrain the output. Ask for a label from a list, a JSON object with named fields, or a draft. Validate before the next node. Free text into a Switch node is the most common cause of a workflow that works in testing and fails in week two.

Keep the send, pay and delete actions behind a person. The human-factors evidence on automation is older than language models and consistent: in a controlled study, people with an automated aid made more errors when the aid was wrong than people without one, and a systematic review of 74 studies found the pattern across aviation, medicine and process control, with training, display design and accountability as mitigators.^{[6] [7]} A A workflow that drafts and a person who reads is the pattern the evidence supports; a workflow that sends is a bet on the model's error rate.

Measure agreement. For patterns 2 and 7, log what the model proposed and what the person decided. If the person changes less than one proposal in fifty for three months, you have data to argue for loosening the gate. If they change one in five, the model is doing less work than it appears.

Error handling patterns

• One error workflow for everything (pattern 10), set in each workflow's settings. Include the workflow name and the last node in the message.
• Retry on the node. Most nodes have a "retry on fail" option with a wait between attempts; use it for rate limits and network errors, not for logic errors.
• Continue on fail with an error branch. Nodes can be set to output errors to a separate branch instead of stopping; route that branch to a Slack message and carry on with the other items.
• Idempotency keys. Store the source record's id in the destination and look it up before writing. Triggers fire twice more often than you expect.
• Timeouts. Cloud trials enforce a 180-second execution timeout;^[1] long loops over external APIs should batch with a Loop Over Items node (formerly Split in Batches) and a short Wait.
• Dead-letter sheet. Append failed items to a sheet with the raw payload. It is the fastest way to replay them after a fix.

Importing and adapting templates

The official library at n8n.io/workflows lists 14,013 templates at the time of reading, filterable by category (AI, Sales, IT Ops, Marketing, Document Ops) and by the apps in your stack.^[8] Templates import as JSON; each node that needs a credential warns until you select one of yours. Four habits make imports safe:

1. Read every node before you activate. A community template may post to a channel or call an API you did not expect.
2. Replace credentials one node at a time and test with pinned data (n8n lets you pin a sample output on a node so downstream nodes run without hitting the live source).
3. Swap the model node for the provider you have a contract with, and check what data the prompt sends.
4. Rename the workflow with an owner and a date, and point it at your error workflow before it runs on a schedule.

Our own template library, with the ten patterns above as importable JSON plus a run-cost sheet, is in preparation as part of our automation product. It is not downloadable yet; the Productivity Signal briefing announces it when it is, and subscribers get it first. Until then, the official library plus the table above is the honest route. For the broader context of what to automate and how to count the return, start with our guide to business process automation for small teams and the overview of workflow automation tools.

Demand and where it goes

US searches for "n8n workflow templates" average about 320 a month with a keyword difficulty of 4; "n8n workflow examples" adds 210 and "n8n ai agents" 1,600. The template term fell 77% year on year, while the n8n brand itself draws about 165,000 US searches a month and its share of the workflow-automation group on Google Trends rose 3.6 percentage points over twelve months (DataForSEO, retrieved 2026-11-09). People search for the product more and for templates less, which fits a library that grew to 14,000 entries. n8n raised USD 180 million in October 2025 at a USD 2.5 billion valuation.^[9] B In our Future Demand Radar, "Workflow automation" scores 56.9 in the baseline scenario (band 52.6–61.2, "build now") and "AI agents and delegated work" 59.7 (band 55.2–64.1); forecasts are model outputs with stated uncertainty, not facts.

What to do next

Build pattern 10 first, then pattern 3. Count your trigger events for a month before you choose between Starter, Pro and self-hosting; the table's assumptions make that a ten-minute exercise. If you have not settled on n8n, the Tool Finder asks about your volumes, data location and whether you need AI steps, and returns a shortlist with dated prices. The Zapier, Make and n8n comparison shows the same workflow priced on all three.

Frequently asked questions

Are n8n workflow templates free?

The official library at n8n.io/workflows lists more than 14,000 community templates that import into the editor as JSON; the page we read does not state a price for importing them, and in practice the cost is the plan your executions require plus any model or API fees the template calls. Our own ten-pattern pack is in preparation and will be announced in the briefing.

What counts as an execution in n8n?

n8n's pricing page: an execution is a single run of your entire workflow, regardless of how many steps it has or how much data it processes.^[1] A loop over 200 items inside one run is one execution; a trigger that fires per email is one execution per email.

Should I use n8n Cloud or self-host it?

Cloud Starter (EUR 20 a month billed annually for 2,500 executions, checked 2026-11-09) is the right start for a team without a technical owner, with data hosted in Frankfurt. Self-host the free Community edition when your workflows touch confidential mail, internal databases or a knowledge base you do not want to upload, and when someone owns updates and backups. A small server costs under EUR 7 a month; the hours are the real cost.

How do I handle errors in an n8n workflow?

Set an error workflow in each workflow's settings. It must start with the Error Trigger node and receives the execution id, error message, last node and workflow name;^[5] route that to Slack or email and a log sheet. Add retry-on-fail to nodes that hit rate limits, an error branch for nodes that should not stop the run, and an idempotency check before every write.

Can n8n run AI agents?

Yes. The AI Agent node chains a model with tools (HTTP requests, vector stores, other workflows) and a memory node inside a workflow, and a run with many model calls still counts as one execution; the model provider bills the tokens separately. Keep send, pay and delete actions behind a human step: the evidence on automation bias shows reviewers accept wrong automated output more often than they expect.^{[6] [7]} A

Sources and evidence

1. n8n pricing — n8n GmbH (2026), vendor page A n8n.io/pricing/ (retrieved 2026-11-09)
2. Host n8n (self-hosting documentation) — n8n GmbH (2026), vendor page A docs.n8n.io/hosting/ (retrieved 2026-11-09)
3. n8n Sustainable Use License (LICENSE.md) — n8n GmbH (2026), vendor page A github.com/n8n-io/n8n/blob/master/LICENSE.md (retrieved 2026-11-09)
4. Hetzner Cloud: server prices and locations — Hetzner Online GmbH (2026), vendor page A www.hetzner.com/cloud/ (retrieved 2026-11-09)
5. Handle errors gracefully (Error Trigger and error workflows) — n8n GmbH (2026), vendor page A docs.n8n.io/flow-logic/error-handling/ (retrieved 2026-11-09)
6. Does automation bias decision-making? — Skitka, Linda J.; Mosier, Kathleen L.; Burdick, Mark (1999), RCT B doi.org/10.1006/ijhc.1999.0252 (retrieved 2026-11-09)
7. Automation bias: a systematic review of frequency, effect mediators, and mitigators — Goddard, Kate; Roudsari, Abdul; Wyatt, Jeremy C. (2012), systematic review A doi.org/10.1136/amiajnl-2011-000089 (retrieved 2026-11-09)
8. Workflow automation templates (official library, 14,013 templates at reading) — n8n GmbH and community (2026), vendor page A n8n.io/workflows/ (retrieved 2026-11-09)
9. Hyped AI business automation startup n8n raises $180m, valuing it at $2.5bn — Tech.eu (2025), report B tech.eu/2025/10/09/hyped-ai-business-automation-startup-n8n-raises-180 (retrieved 2026-11-09)