Automation agency: n8n, Make or custom build

The question asked is almost always "which tool", when the expensive part is how many edge cases your process contains. This page gives the grid in the right order.

What we are talking about

An automation agency wires your tools together so repetitive work happens without intervention: an order that triggers an invoice, a form that creates a client record, a document that goes for review at the right moment. For two years now part of these chains has called a language model to sort, summarise or draft, which moves the problem without simplifying it.

Tool choice comes last, which is the opposite of how the conversation usually starts. What determines the cost and the lifespan of an automation is how many exceptions the real process contains, the quality of the input data, and whether anyone will be able to repair it in six months.

Kinds of provider

  • Self-hosted n8n

    Keeping data on your side, connecting internal tools, and escaping per-task billing as volume grows.
    Having no operator. Self-hosted means updates, backups and monitoring are on you.
    Server cost, plus the build and a maintenance line.
  • n8n or Make, vendor-hosted

    Starting in days without touching infrastructure, on connectors already built for common tools.
    Scaling. Per-operation billing becomes the main line, and the bill surprises just as the automation finally works.
    Per-operation or per-run subscription, plus the build.
  • Custom development

    Processes with many exceptions, high volumes, and anything that must be tested, versioned and monitored like real software.
    Simple needs. Writing code for what a connector already does costs more for an identical result.
    Fixed project price, low running cost afterwards.
  • Supervised AI agent

    Steps that used to need human judgement: triaging a request, drafting a first version, extracting data from a badly structured document.
    What a deterministic rule already does well. Putting a model on a VAT calculation adds risk and gains nothing.
    Fixed-price build, then tokens and review time every month.

What actually decides

  1. 01

    The number of exceptions

    Have the person doing the task list every case where they leave the normal path. That count, not the tool choice, determines whether the automation takes three days or three weeks.

  2. 02

    Cost at real volume

    Ask for the monthly cost at the volume you will have in a year, not today's. Per-operation billing that looks painless at a hundred runs becomes the main line at ten thousand.

  3. 03

    Who repairs it in six months

    An automation breaks when a tool changes its interface, and that happens. If nobody on your side can open the workflow, you have bought an open-ended dependency.

  4. 04

    Where the data goes

    A vendor-hosted flow routes your data through the vendor. Depending on what travels, that is irrelevant or it is the criterion that decides everything. The question comes before the tool choice, not after.

Warning signs

  • A tool choice announced before analysing the process
  • No costing of the monthly bill at target volume
  • A flow delivered with no documentation and no access for your team
  • An automation laid on a process nobody has measured
  • A language model placed where a simple rule would do

Where we sit

We build in n8n when the connectors exist and the process is stable, and custom when exceptions run into the dozens. The choice is made after measuring the process, never before, and it is explained to you with what it costs either way.

What lands on your side in every case: the flows, the documentation and the access. An automation your team cannot open is not delivered, it is rented.

When we are not the right choice

  • If your need fits a standard Zapier connector, install it yourself in an hour. Paying for an engagement makes no sense.
  • If the process changes every month because it is not settled, automate later. You would pay twice, once to build and once to rebuild.
  • If nobody on the team can describe the task precisely, start with a workshop rather than a flow.

Our ranges

  • Scoping workshopfrom €900Half a day to measure the process and list the exceptions before choosing a tool.
  • Automationfrom €4,000Project, typically 2 to 6 weeks, flows and documentation delivered to you.
  • Audit and roadmapfrom €1,500When several processes are candidates and none is costed.

Excl. VAT, France. Monthly running cost is always costed separately from the build.

Frequently asked

n8n or Make, which one?
n8n if you want to host it yourself, connect internal tools or avoid per-operation billing at volume. Make if you want to start with no infrastructure on common tools. The question is only usefully settled after measuring the process volume and exceptions.
What does an automation cost?
At kengdev an automation project starts at €4,000, with the monthly running cost costed separately. The gap between two quotes comes from the number of process exceptions, which is also what most costings forget to count.
AI agent or classic automation?
Classic automation whenever the rules are deterministic, which remains the most common case. An agent only where the step needed judgement: triaging, summarising, extracting from a badly structured document. Putting a model on a calculation adds risk with no gain.
What if the flow breaks?
It will break one day, because a tool will have changed its interface. What matters is who reopens it. We deliver the flows, the documentation and the access to you, so repair does not depend on a single provider.

Scope your first AI agent

20 minutes to review your tools, data and the first useful case. No jargon, no commitment.