Why can't you just give me a fixed price up front?

We do give a fixed price — after a short scoping conversation. We cannot quote responsibly before we understand the workflow, the data and the integrations, because those are what drive cost. Once scope is clear, you get a fixed number against a real deadline, paid 50/50.

Are the running costs included in the build price?

No. The build is a one-time cost; running costs (model usage, hosting, infrastructure) are ongoing and billed separately. We pass them through transparently at cost and never add a markup, so your running cost stays your running cost.

What does the 50/50 split actually mean for me?

You pay half to begin the work and the remaining half only once the application is delivered and working as agreed. It moves the risk onto us: we are only paid in full when you have something that works.

← All insights

·June 5, 2026

What it costs to build an AI application

By Sandeep Bali · Founder

A focused AI proof of concept typically runs €8,000–€25,000. A production AI application — a copilot, an internal tool, a customer-facing assistant — usually lands between €25,000 and €120,000 depending on integrations and data work. Running costs (models, hosting, infrastructure) add a few hundred to a few thousand euros a month. The build is one-time; the running cost is forever.

// Key takeaways

Proof of concept: roughly €8,000–€25,000 to prove the idea works.
Production application: roughly €25,000–€120,000 depending on scope.
Running costs are separate and recurring — budget for them from day one.
Most of the cost is integration and data work, not the AI model itself.
A 50/50 split means the balance is only due once it ships and works.

Why AI projects resist a single price tag

There is no honest flat rate for an AI application, because the phrase covers everything from a single-purpose document classifier to a multi-tenant platform with a fine-tuned model and live human review. What you are really paying for is engineering time, and the time depends on how much of your world the system has to understand and connect to.

The model itself — the part people fixate on — is rarely the expensive bit. The cost is in the work around it: clean data, reliable integrations, sensible guardrails and a product people actually want to use.

Proof of concept: €8,000–€25,000

A proof of concept answers one question: does this actually work with our data and our users? It is deliberately narrow — one workflow, one data source, enough polish to judge it honestly, and not a line more.

At this stage you are buying certainty, not a finished product. A good proof of concept either earns the budget for a full build or saves you from spending it on something that was never going to work.

Production application: €25,000–€120,000

A production application is the real thing: authentication, multiple integrations, error handling, monitoring, a polished interface, and the reliability to put in front of paying users or your whole company.

The range is wide because scope is wide. A focused internal copilot wired into two systems sits at the lower end. A customer-facing assistant with several integrations, role-based access, audit logging and a data pipeline sits at the higher end. The honest way to narrow it is a scoped quote, not a guess.

The running costs nobody warns you about

Beyond the build, an AI application costs money to operate every month. Model usage (API calls or self-hosted inference), hosting and infrastructure, and any third-party services together typically run from a few hundred to a few thousand euros a month depending on volume.

These costs scale with usage, so they grow as you succeed. We pass them through transparently and never mark them up — the running cost of your software should be your running cost, not a margin for us.

How the 50/50 model removes your risk

The biggest fear in commissioning software is paying in full for something that never ships or never works. Our answer is structural: you pay half to start and the balance only when the work is delivered and working. If it does not work, you do not owe the second half.

Combined with two free iterations on every feature and full ownership of the source and IP, this turns a leap of faith into a controlled, reversible decision — which is exactly what it should be.

// faq

Frequently asked questions

Why can't you just give me a fixed price up front?: We do give a fixed price — after a short scoping conversation. We cannot quote responsibly before we understand the workflow, the data and the integrations, because those are what drive cost. Once scope is clear, you get a fixed number against a real deadline, paid 50/50.
Are the running costs included in the build price?: No. The build is a one-time cost; running costs (model usage, hosting, infrastructure) are ongoing and billed separately. We pass them through transparently at cost and never add a markup, so your running cost stays your running cost.
What does the 50/50 split actually mean for me?: You pay half to begin the work and the remaining half only once the application is delivered and working as agreed. It moves the risk onto us: we are only paid in full when you have something that works.