08 — Workflows
CoderConda builds workflow automation with n8n and its own code from €495 per process, and first the four paths for when a system does not answer. A process between two systems is built in a day, as long as you assume both answer. They do not always answer.
The first process runs within days, on your own server.
Run 0412 · request → CRM → quote6 + 4
6 · when all goes well4 · when it does not
6 · when all goes well, 4 · when it does not
02 — The branches
01 — at node 2
A field is missing.
A phone number without an area code, a company name in the first-name field, a date in the wrong format. The run cannot calculate with that, and it still must not lose the case: behind it is somebody who wanted something. It lands in a clarification path with the raw data beside it.
Raw data is logged before anything interprets it.
02 — at node 3
It runs twice.
A form submitted twice. A webhook the sender repeats because the answer took too long. A run somebody restarts by hand. This happens, and it must not produce two contacts, two cases and two quotes. The node recognises that it already knows this one.
Triggered twice does not mean two records.
03 — at node 4
The case fits nowhere.
Two paths are written down, and the case belongs on neither. This is where a process decides whether it can be trusted: it can take the more likely path, or it can say it does not know. The second takes longer and is the only one you can explain afterwards.
A rule decides, not a model that estimates.
04 — at node 5
The interface goes quiet.
No error message, no crash, just silence and a timeout. It is the most common case and the one finished tools give up on. The run retries three times with growing gaps, because most outages last minutes, and then reports: with a reason, the raw data and a button to start it again.
The retry is cheap. The silence after it is not.
Branch 1 of 4
Run 0412 · request → CRM → quote6 + 4
at node 2A field is missing.Clarification path
6 · when all goes well, 4 · when it does not
03 — The silent failure
After four branches it seems obvious to swallow every error so nothing breaks. That produces the one case nobody can repair: a process that looks green and has done nothing for days.
Ends in nothing
The error is caught and thrown away. The run reports success. It will surface in three weeks, when somebody looks for a quote that was never written.
Ends at a list
The error is caught and written down: with a reason, the raw data and a timestamp. Somebody sees it the same day, and the run can be restarted with one button.
The rule
No branch ends in nothing. Every one ends at a list or at a person.
That is why I build the failure paths first and the happy path after: the happy path turns out fine anyway, the other one only if you start with it.
04 — The evidence
A phone front desk is a process made of failure cases: the caller speaks freely, the request fits no form, the calendar is full. Still, no call may disappear.
Every call ends as a note, an appointment or a transfer — on the record instead of lost.
The rule on this page
No branch ends in nothing. Every one ends at a list or at a person.
What it does not prove
It is a product of my own, not a client engagement. What it proves is that the rule holds under load, not that it fits everywhere.
05 — What it costs
The unit is one process, and the first is finished within days. What it costs is settled up front. Only one thing recurs, and it is not a fee per run. All prices are final prices. No VAT is charged under § 19 UStG.
Once
Monthly
That is the whole list: upkeep, monitoring and changes. n8n can be self-hosted, and its source code is open to read.
Does not apply
06 — The server
A process only its builder can see is not a tool, it is a dependency. So everything sits on a machine you have access to, in a tool whose source is open. What is inside it is listed here.
Your machine
If you do not yet know which process it is, start one step earlier. And if the path you need is not between systems but between people, you need something other than wiring.
If it first has to be counted which steps should exist at all.
If the path is not between systems but between you and the client.
07 — The objections
This page claims that no path ends in nothing. The same goes for these four: each hangs off an answer, and it is written underneath.
“We have Zapier for that.”
For the happy path that is often enough. The difference starts at the fourth failure case, and at the per-run price, which grows with success.
“Our system has no interface.”
Then I take the way that does exist: a database, an export folder, SFTP, or a cleanly built import if it comes to that. This is everyday work with software that has run for ten years.
“And if one of the systems changes?”
Then the process breaks, and it breaks loudly. That is exactly what the error list is for: it shows the same day which node is stuck and why.
“We could build this ourselves.”
Probably true, and the happy path is quick to build. What costs time is the four branches, and those rarely get built the first time round.
Finished wiring tools are good, and for two systems with a straight path I use them myself. Two things change the picture. First the branches: past about three paths the interface gets unreadable, and what you cannot read, you do not maintain. Second the price, which is charged per run and therefore rises exactly when the process starts paying off. n8n costs a server instead, and that costs the same whether a hundred or a hundred thousand runs go through it.
A missing interface is rarely a missing possibility. Almost every system in operation can hand something out: a report into a folder, a table overnight, database access for a read-only account. That is less comfortable than a documented interface and carries just as far, as long as the path is written down and the failure case is built. What I do not do is a click robot driving a user interface: that breaks at the next update, and it breaks silently.
Systems change, and a process that does not allow for it is a bet. The question is not whether it breaks eventually, but whether anybody finds out. That is why the error list is not at the margin of this work but at its centre: every failed run sits there with a reason, the raw data and a timestamp, and can simply be restarted after the repair. A process that stops immediately and visibly on a change costs an afternoon. One that keeps running and writes wrong data costs the week you spend undoing it.
This is the objection I agree with most often. n8n is expressly built so that people without a developer background can work with it, and the straight path between two systems is wired up in an afternoon. The difference is in what you do not build the first time, because you have not lived through it yet: retries with growing gaps, a run that writes one record even when triggered twice, a clarification path instead of a bin, a list somebody actually looks at. If you want to build it anyway, I set the first process up so the others fit beside it, and you carry on.
08 — Appointment
Pick a time that suits you straight from the calendar. I go through with you where your company stands and what is worth doing. No obligation and no sales pressure.
The calendar is hosted by cal.com.
Loading it opens a connection from your browser to cal.com (USA), which transmits your IP address; cal.com may set cookies when it loads. You can also go straight there instead.
