Pricing for AI work is usually vague, and often vague on purpose. The honest reason is that the real cost of automation is not a single number on a sticker. It is three different numbers, paid at different times, and the one most likely to surprise you is the one nobody quotes. This is the transparent version: what the software costs, what a build costs, what keeping it alive costs, and the line items that quietly add up underneath all of it.
The three layers of cost
Every automation you run sits on three layers, and confusing them is how budgets go wrong. The first is the software itself, the models and tools that do the work. The second is the build, the engineering that turns those tools into something shaped like your business. The third is maintenance, the ongoing care that keeps the whole thing from rotting. Treat them separately and the picture gets a lot clearer.
Layer one: the software
At the bottom are the tools. Language model usage is typically billed by the volume of text processed, which sounds abstract until a busy workflow turns it into a real monthly figure. On top of that sit the platforms you connect things with, the automation runners, the databases, the email and messaging services. Individually these are modest, often tens of euros a month each. Collectively, across the median small-business stack of several tools, they add up to a recurring cost worth tracking rather than ignoring.
Layer two: the build
Software on its own does not automate anything. Someone has to design the workflow, wire the tools together, handle the edge cases, test it against real inputs, and ship it where it will actually run. That engineering is the build, and it is usually the largest single line, paid once. Where it lands depends on how much you are building.
| What you are buying | Typical cost | What it covers |
|---|---|---|
| Tactical audit | €1,999 | A prioritized plan: what to automate, how, and with which tools |
| Single build (Sprint) | €4,999 | One workflow built, tested, and handed over in five days |
| Multi-system build (Custom) | from €10K | Several systems, bespoke logic, scoped per engagement |
| Off-the-shelf tools | subscription | Pre-built software you configure and rent monthly |
Layer three: maintenance
This is the layer the cheap quotes leave out, and it is the one that decides whether your automation is an asset or a liability a year from now. Models get deprecated on a schedule you do not control. APIs change. Prompts that worked perfectly in spring drift as the underlying model updates. A business that ignores this layer is not saving money. It is deferring a failure to a worse moment. Budgeting for ongoing operations, whether you handle it in-house or through a retainer from €2,500 a month, is part of the true cost, not an optional extra.
The hidden costs nobody puts on the invoice
Beyond the three layers sit the costs that never appear in a proposal yet routinely outweigh the ones that do. There is the integration glue, the unglamorous work of making two tools that were never designed to talk actually talk. There is supervision, the staff hours spent babysitting an automation that works most of the time. There is the per-action pricing on platforms that feels trivial at launch and becomes a tax on growth once volume climbs. And there is the largest hidden cost of all, the opportunity cost of the thing you did not automate because you were unsure where to start.
That last one is why spending a little to know a lot is almost always the cheapest opening move. The most expensive mistake in this whole field is not paying too much for a build. It is building the wrong thing well, then maintaining it for a year before admitting it never mattered.
Build versus buy, in one line
The build-versus-buy decision usually comes down to how central the workflow is and how fast you are growing. Renting a tool is cheaper today and stays cheaper as long as the workflow stays peripheral and your volume stays low. Owning a custom system costs more up front and wins over time once the workflow becomes core and the platform's per-action or subscription tax starts compounding. We work through that trade-off properly, with the three-year math, in our piece on custom AI systems versus SaaS subscriptions.
What to do with these numbers
Treat the ranges here as a frame, not a quote. The figure that matters is the one attached to your specific operation, and the cheapest reliable way to get it is to have someone map your work first. That is exactly what a tactical Audit produces, and if you would rather do a first pass yourself, the method is no secret: our guide on how to choose what to automate first shows how to rank candidates by what they actually return.