n8n and AI automation
n8n consultant for B2B revenue teams
We design, build, and hand over n8n automation for revenue teams: enrichment, routing, CRM hygiene, and the reporting around them. Independent, based in Berlin, working across Europe. We run our own revenue operation on this class of tooling, which is also why we will tell you when Zapier, Make, or nothing at all is the better buy.
The short version
- n8n bills per workflow execution rather than per step. That one fact is why teams move to it, and why the saving is real only if your workflows are long or your volume is high.
- The self-hosted Community Edition has no licence fee and no execution cap. It also has no SSO, no role-based access control, no git-based environments, and no external secrets, which is usually what a security review asks for.
- n8n is fair-code, not open source. Internal business use and paid consulting around it are allowed. Hosting someone else's workflows as a service, white-labelling, or embedding it in your product needs a paid licence.
- Self-hosted instances rarely fail on cost. They fail on an executions table nobody pruned, an encryption key nobody backed up, and a version nobody dares upgrade.
What n8n demand actually looks like from our own property
We publish four n8n pages that rank, so we can report this from our own Search Console rather than repeat a keyword tool. Across ziellab.com, 9 June to 6 September 2026, those four pages drew 11,390 impressions and 42 clicks. In the query view, 16 rows contain the string n8n: 955 impressions, 7 clicks.
The useful part is what those 16 rows are. Not one of them is service-shaped. There is no n8n consultant, no n8n agency, no hire n8n developer. Every human-looking row is a learning query: n8n use cases at 751 impressions and position 13.0, n8n use case at position 9.0, what is n8n used for at position 1.2, n8n enterprise use cases, best n8n workflows for business. Two more are analyst-shaped, asking for n8n's customer count and revenue, and one is a literal search operator excluding n8n's own domain, sitting at position 100. Our money-term tracker records no impressions at all for n8n consultant against this property in the same window.
Stated plainly, because the same honesty has to apply to our own service page: people search n8n to learn how to build with it, not to hire someone to build it for them. Nobody is going to arrive here because of volume. So this page is written for the smaller group who have already decided, and who want to know what they are actually buying before a call. If that is you, the sections below are the things we would tell you on the call anyway.
One limit on the above: this is one property, ours, ranking for the topics we happen to have written about, and reading a query as human, analyst, or machine is our own judgement on the text rather than a published taxonomy. Treat the shape as evidence and the exact counts as indicative of this site, not of the market.
The four ways this decision gets billed
The feature lists overlap more than the vendors suggest. The billing model is where these tools genuinely differ, and it is what decides your bill once the workflows grow. The longer version, with worked numbers at 10,000 runs a month, is in our n8n versus Zapier versus Make comparison.
| Option | How it charges | When it is the right fit |
|---|---|---|
| Zapier | Per task, so every step of every run | Short workflows owned by a non-technical builder who needs to fix them without filing a ticket. The bill grows as workflows get longer, which is the part nobody models in year one. |
| Make | Per operation, cheaper per unit than a Zapier task | Ops people who want to see the whole flow on a canvas and branch heavily. The middle option on both price and complexity. |
| n8n Cloud | Per workflow execution, whatever the node count | Long workflows at moderate volume. A fifty-node run costs the same as a five-node one, which is the entire argument for moving. Workflows and users are unlimited on the current cloud plans, so execution volume is the only lever on the bill. |
| n8n self-hosted | No licence fee, no execution cap, you pay for the server | High volume, or data that cannot sit with a hosted vendor. You are trading a subscription for an operations job, so it only pays off if someone owns that job. |
On the independent 2026 pricing breakdowns that agree with each other, n8n Cloud runs about 24 euros a month for 2,500 executions on Starter and about 60 euros for 10,000 on Pro, with a discount for annual billing. We do not quote a Business tier price here because those same breakdowns contradict each other on what it costs and what it includes, and we would rather leave a number out than publish one we cannot stand behind.
What the licence actually lets you do
This comes up in every procurement conversation and is usually answered wrongly. n8n is source-available under its own Sustainable Use License, a model n8n named fair-code. You may use, modify, create derivative works from, and redistribute the software, with the limitation that use is for your own internal business purposes or for non-commercial and personal use.
Two practical consequences. First, n8n lifted the restriction on charging for consulting and support work around the software, so an agency building workflows for you does not need a separate agreement with n8n, and neither do we. Second, the line you cannot cross is running n8n as a service for other people: permanently hosting your customers' workflows or credentials on an instance you operate, white-labelling it, or embedding it inside your own product. Those need a paid Enterprise or Embed licence.
Why we build in your instance, not ours
The workflows and the credentials live in your n8n, under your billing, on infrastructure you control. That keeps the licence question simple, keeps your data out of an agency's tenancy, and means the day we stop working together you lose a contractor rather than a system. An automation estate only your agency can reach is a bill you keep paying, not an asset you own.
Self-hosting is a cost trade, not a free lunch
Self-hosted n8n has no licence fee and no execution cap, and the published 2026 breakdowns put the server at roughly 4 to 7 USD a month on a small VPS. Both things are true and neither is the number that decides whether this was a good idea. What decides it is the operations work that comes with owning the instance, and that is mostly settled at setup.
The four decisions that make a self-hosted instance survive
- Prune the execution data. An unbounded executions table is the most commonly cited cause of a dead self-hosted instance. Pruning is on by default and the thresholds are tunable, so the failure is always someone turning it off or never checking it.
- Choose queue mode before you need it. Production runs go to worker processes with Redis as the broker and Postgres as the store, and n8n advises against queue mode on SQLite. Retrofitting it under load means new dependencies and a redeploy on a bad day.
- Treat the encryption key and the webhook URL as first-class configuration. Losing the key means losing every stored credential, and it does not belong only on the box it protects.
- Decide who upgrades it, and when. An instance nobody dares move off its current version is the same problem as an instance nobody backs up, arriving more slowly.
There is also a governance gap worth knowing before a security review rather than during one. Community Edition does not include SSO with SAML or LDAP, role-based access control, projects, git-based environments, or external secret stores. Those sit in the Enterprise tier. If your requirements list contains any of them, the real decision is not n8n against Zapier. It is n8n Enterprise against a hosted vendor whose compliance paperwork already exists.
What we do
Revenue workflows built end to end
Enrichment, lead routing, CRM hygiene, alerting, and the reporting that says whether any of it worked. Built inside your instance, named and documented so an ops person can read the flow without opening a code node.
Migration off per-task billing
Zapier and Make bills grow with steps, not runs. We size the move honestly first, because a fifteen-workflow estate on a small plan is usually not worth migrating, and a long workflow running thousands of times a month usually is.
Self-hosting set up properly
Queue mode, Postgres, Redis, execution pruning, the encryption key held somewhere other than the server, backups, and a documented upgrade path. This is where self-hosted instances quietly die, and it is nearly all decided at setup.
Rescuing an instance nobody maintains
The common shape: workflows a contractor built, a database that has never been pruned, and no one who can safely upgrade it. We read it before rebuilding, and the fix is often smaller than the rebuild.
The layer around n8n
Attio or HubSpot as the system of record, Clay where enrichment belongs, and n8n moving records between them. n8n is plumbing. Buying it as a revenue strategy is the mistake we get called in to unwind.
The workflows we build most often are written up in our n8n workflows guide for B2B revenue teams, with a wider set of examples, and their failure modes, in seven n8n use cases revenue teams actually run. We do not publish our rates on this page. The shape: engagements are scoped and bought in blocks against a written plan, with no retainer, and nothing is built before it is scoped.
When we say you should not automate it yet
Three cases come up often enough to state in advance. When the CRM underneath is not trustworthy, automation moves bad records faster and makes the mess harder to see. When nobody has decided who follows up, a routing workflow just relocates the decision. And when the process runs a few times a week by hand, a workflow costs more in maintenance than it saves in minutes.
The order we would sell you in those cases is the data model and the routing rules first, automation second. That work sits on our Attio consulting page, and the routing side, from enriched record to the rep who works it, is covered in our speed-to-lead routing guide. If enrichment is the actual bottleneck, the Clay and n8n split is worked through in our GTM launch playbook built on Clay and n8n.
How the engagement works
Read it before building it
What runs today, in which tool, at what cost, and what breaks when it fails. Output is a build plan: which workflows, in what order, what stays manual on purpose, and what we think you should not spend on. The plan is yours either way, whether your team executes it or we do.
Build one workflow end to end
One trigger, one path, wired through to the CRM and to a report you can read, with error handling and an alert when it fails silently. Proving one workflow in production beats twelve half-finished ones, and it prices the operations load honestly before you commit to more.
Hand it over documented
Naming conventions, a runbook for the failure cases, credentials owned by you, and a walkthrough. We measure success by whether your team can change a workflow without us. If the answer is no, we did the job badly.
Questions we get asked
Do we even need a consultant for this?
Often not. n8n is designed to be built in by the people who own the process, and if you have someone technical with time, the community material is good enough to get a first workflow live. Where outside help pays for itself is the parts that are expensive to get wrong later: how the instance is hosted, how workflows are named and versioned, what happens on failure, and which processes should not be automated at all.
We have an instance a contractor built and nobody touches. Now what?
The first step is not rebuilding. It is finding out which workflows still run, what they touch, whether the database has ever been pruned, and whether anyone holds the encryption key. Frequently the workflows are fine and the break is operational, which is a much smaller fix than a rebuild, and we will say so.
Will n8n still be here in three years?
A fair question for any dependency. The public record: n8n raised a 180 million USD Series C in October 2025 led by Accel at a 2.5 billion post-money valuation, bringing total funding to around 240 million since 2019, and reported more than 60,000 self-hosted installations and over 3,000 cloud customers in March 2026. That is a healthy vendor rather than a guarantee. The mitigation that actually holds is the one above: your instance, your credentials, your workflows exportable as JSON.
Do you host it for us or run it as a managed service?
No. We set up and maintain instances that you own and pay for directly. Hosting other companies' workflows on infrastructure we operate is exactly the use the Sustainable Use License reserves for a paid licence, and it is also the arrangement that makes leaving an agency painful. We would rather build something you can take with you.
Where these numbers come from
Every external figure on this page and where it originates, so you can check it rather than trust us.
- The Sustainable Use License, its three limitations, fair-code, the lifted restriction on paid consulting and support, and the Enterprise or Embed licence needed to host others' workflows or embed n8n: n8n's own licence documentation and repository licence (docs.n8n.io/sustainable-use-license, github.com/n8n-io/n8n LICENSE.md), with corroborating readings from Nordflux (nordflux.de/en/guides/the-n8n-sustainable-use-license-explained), Scalevise (scalevise.com/resources/n8n-automation-license-commercial-use) and Digitalcube (digitalcube.ai/en/blog/n8n-licencias-fair-code-community-enterprise).
- Cloud plan prices and execution allowances, unlimited workflows and users on cloud plans, execution as one full workflow run, and the free Community Edition with no execution cap: independent 2026 pricing breakdowns that agree with each other, from No Code MBA (nocode.mba/articles/n8n-pricing), Techjack Solutions (techjacksolutions.com/ai-tools/n8n/n8n-pricing), Goodspeed (goodspeed.studio/blog/n8n-pricing), Lindy (lindy.ai/blog/n8n-pricing) and InstaPods (instapods.com/blog/n8n-pricing). The same sources disagree on the Business tier, which is why no Business price appears above.
- Queue mode with Redis and Postgres 13 or later, the advice against queue mode on SQLite, execution pruning being on by default and tunable, the unpruned executions table as a common cause of failed instances, and the encryption key and webhook URL as first-class configuration: n8n's hosting documentation (docs.n8n.io/hosting/scaling/queue-mode) and practitioner write-ups from Royfactory (royfactory.net/posts/ai/202601/n8n-self-hosting-docker-compose-queue-mode-checklist), AiCybr (aicybr.com/blog/n8n-self-hosting-docker-postgres-guide) and Sigmalogic (sigmalogicai.com/blog/n8n-queue-mode-scaling).
- The Community Edition gaps: no SSO with SAML or LDAP, no role-based access control, no projects, no git-based environments, no external secrets: n8n's community edition feature documentation (docs.n8n.io/hosting/community-edition-features) and enterprise feature summaries from LowCode (lowcode.agency/blog/n8n-enterprise), Openhosst (openhosst.com/n8n-community-vs-enterprise) and AiUnpacker (aiunpacker.com/blog/n8n-enterprise-features-everything-you-need-to-know).
- The 180 million USD Series C in October 2025 led by Accel, the 2.5 billion post-money valuation, roughly 240 million raised since 2019, and the March 2026 figures of more than 60,000 self-hosted installations and over 3,000 cloud customers: n8n's own funding announcement (blog.n8n.io/series-c) and the trade coverage of it, from FinSMEs (finsmes.com/2025/10/n8n-raises-180m-in-series-c-funding-at-2-5-billion-post-money-valuation.html), Tech Funding News (techfundingnews.com/n8n-raises-180m-series-c-2-5-billion-valuation-automation-ai) and The AI Insider (theaiinsider.tech).
- The 11,390 impressions, 42 clicks, 16 n8n query rows, their positions, and the absence of any service-shaped n8n query: our own Google Search Console property for ziellab.com, 9 June to 6 September 2026. Reading a query as human, analyst, or machine is our own judgement on the query text rather than a published taxonomy.
One limitation worth stating plainly: n8n.io and its documentation are not reachable from the environment this page was written in. Figures attributed to n8n above were established from multiple independent 2026 sources that agree with each other and are named to the organisations publishing them. Anything that could not be corroborated that way was left off rather than softened, which is why no Business tier price, no enterprise licence cost, and no time-saved benchmark appears here.
Written by Abhishek Singla, who runs Ziel Lab's own revenue operation on this stack. More about us.
Not sure automation is the answer? Get a second opinion first.
Describe your current setup and what is stuck. Within 7 days you get our written opinion: what we think is going on, how we would tackle it, and what we would avoid. In total privacy, no pitch inside.
Ask your question