Pricing Indicative ranges Quoted per project You own the code

AI project pricing.
Rough but honest.

The numbers on this page are drawn from what past projects have actually cost, so you can tell whether we are in your range before spending an hour on a call. They are estimates, not a price list: every project is scoped and quoted on its own terms, and a large or unusual project will not fit any band below.

ENGAGEMENT MODELS

Ranges are indicative, based on past client work, in EUR and exclusive of VAT. The quote for your project comes after a scoping call, against a written scope — see how we price.

Prototype Sprint

For proving a technical approach works — investor demos, internal buy-in, or de-risking an architecture before committing to it.

TYPICALLY €8–25k / 2–4 WEEKS

◆A working system on your real data
◆Deployed somewhere your team can actually use it
◆Written assessment of what production would take
◆Code in your repositories from day one

Discovery

For deciding what to build. Assessment of data, systems, constraints and feasibility, with a costed shortlist.

TYPICALLY €8–20k / 2–4 WEEKS

◆Data and systems assessment
◆Prioritised use-case shortlist with effort ranges
◆Reference architecture and evaluation plan
◆Written to be usable by another vendor

Production Build

MOST COMMON

Building the system properly: integrations, evaluation, monitoring, error handling and handover.

TYPICALLY €30–120k / 6–16 WEEKS

◆Integrated with your real systems
◆Evaluation set and regression gates in CI
◆Monitoring, cost tracking and runbooks
◆Security documentation for your review process
◆Larger multi-system builds are quoted separately

Ongoing

Fractional capacity after launch — quality monitoring, iteration, and building the next thing alongside your team.

TYPICALLY €4–9k / MONTH

◆Agreed days per month, working in your systems
◆Quality and cost monitoring, model migrations
◆Knowledge transfer as an explicit goal
◆Month to month — no annual lock-in

CUSTOM AI AGENTS

Agents are the most common request and the hardest to price from a distance, because the word covers everything from a single automated step to a system that acts across your whole stack. These are the three shapes we build, with the range each has tended to land in. Every figure is a rough estimate from past work; your agent is quoted on its own scope.

Single-workflow agent

One job, one or two tools. Triage inbound email, draft replies from a knowledge base, enrich CRM records, summarise and route tickets.

TYPICALLY €8–25k / 2–5 WEEKS

◆One or two integrations
◆Human approval before anything is sent
◆Evaluation set for the one task it does
◆Logging of every action taken

Production agent

MOST COMMON

An agent that carries a role: support, sales research, data analysis or internal ops. Retrieval over your documents, several integrations, escalation paths.

TYPICALLY €25–75k / 6–12 WEEKS

◆Three to five integrations, permissions respected
◆Retrieval with citations and guardrails
◆Escalation to a person with full context
◆Monitoring, cost tracking, regression gates in CI

Multi-agent system

Several agents coordinating across ERP, CRM, email and your data warehouse, with approvals, audit trail and a compliance review to pass.

TYPICALLY €75k+ / 12+ WEEKS

◆Orchestration, hand-offs and shared state
◆Approval workflows and full audit trail
◆Security documentation for your review
◆Priced in phases, re-estimated at each one

Running costs are separate and are yours. A live agent pays for model inference, hosting and any third-party tools it calls. For a single-workflow agent that is often tens of euros a month; a production agent handling real volume tends to run in the low hundreds to a few thousand; a multi-agent system can be more. We model this during the build, put a budget cap on it, and treat cost per task as a metric to optimise, not a surprise on the invoice. If you want us to keep the agent healthy after launch, that is the Ongoing model above.

Agent ranges are indicative, in EUR excluding VAT, and drawn from past client work. See what the agents themselves do under Autonomous Employees.

WORKED EXAMPLES

Bands are abstract until you can see a project like yours in one. These are the shapes of work we are asked for most often, with the range each has tended to land in. Names are left off deliberately; the shapes are real.

	What it involves	Typical range	Typical timeline
Investor or board demo	One workflow on real data, deployed to a URL, written production assessment	€8–20k	2–3 weeks
Internal knowledge assistant	Retrieval over your documents with citations, permissions respected, evaluation set	€20–45k	4–8 weeks
Document processing pipeline	Extraction and classification of invoices, contracts or forms into your systems, with human review queue	€25–60k	5–10 weeks
AI product MVP for a startup	Six-week build of the core feature, one or two integrations, ready to put in front of users	€30–60k	~6 weeks
Customer-facing support agent	Ticketing and CRM integration, escalation paths, guardrails, monitoring and cost tracking	€40–90k	8–14 weeks
Multi-system automation	Agents acting across three or more systems (ERP, CRM, email), approvals, audit trail	€60–120k	10–16 weeks
Marketing site with AI-readable content	Laravel build, design system, structured data, llms.txt and lead capture	€12–35k	3–6 weeks
Large or multi-team programme	Several systems, several phases, compliance review, dedicated capacity	Quoted separately	Months

Indicative ranges drawn from past client work, in EUR excluding VAT. A specific project can land outside its row; the quote is what counts.

WHAT MOVES THE NUMBER

Integrations

The single biggest driver. One clean API is a week; three legacy systems with undocumented behaviour can be most of the budget. Each additional system tends to add more than its share.

State of the data

Clean, accessible, labelled data keeps a project at the low end of its band. Scattered, inconsistent or permission-locked data pushes it up before any AI work begins.

Who sees the output

An internal tool used by ten people is cheaper than the same tool facing customers. External users bring guardrails, escalation paths, monitoring and a higher bar for evaluation.

Accuracy bar

Getting a system from good to reliably good is where the effort concentrates. The acceptance criteria we agree in discovery decide how much evaluation and iteration the build needs.

Compliance and security

Regulated data, on-premise or EU-only hosting, security questionnaires and audit trails all add real work. They are worth doing properly, and they are costed rather than absorbed.

How much you do

A client engineer who owns data access and deployment can take a meaningful share of the work in-house. We price that in; the project gets cheaper and the handover gets easier.

HOW WE PRICE

Everything on this page is an estimate. The ranges are drawn from what past clients have actually paid, and they are published so you get an honest sense of what we are likely to cost. They are not a rate card, and no two projects have ever come out the same. Where yours lands depends mostly on how many systems we have to integrate with and how much work your data needs before it is usable, and neither is knowable before a scoping call.

After that call we quote against a written scope. For well-bounded work, that quote is a fixed figure and it stays fixed unless the scope changes; if the scope changes, we re-quote at the boundary rather than absorbing it quietly. For large, multi-phase or open-ended work, a single fixed number would be a guess dressed up as a commitment, so we price it in phases or as agreed capacity per month, with a fresh estimate at each phase.

Big projects do not fit the bands above, and we do not try to make them. If you are bringing something with several systems, several teams or a compliance review attached, expect a proposal written for that project rather than a number read off this page.

We do not charge per seat, per token or per agent. Those meter usage rather than value, and they make your costs grow precisely as the system becomes more useful, which is a strange thing to want.

What we do not offer is an accuracy guarantee with a refund attached. You will see those advertised. They are unenforceable in practice, because nobody defines what counts as a failed output, who adjudicates it, or over what sample — so the guarantee cannot be invoked and functions as marketing. What we do instead is agree acceptance criteria against a held-out evaluation set during discovery, and measure against it. If a milestone misses the agreed criteria, we fix it at no charge; if we cannot, we agree a fair reduction for that milestone. The details are written into the statement of work, which is what makes it a commitment you can hold us to.

THE WIDER MARKET

It helps to know where our ranges sit relative to the market. Offshore agencies and marketplace freelancers quote materially below us; the large consultancies quote materially above, often by a multiple, for work of a similar shape. Our ranges sit in the middle of that spread.

The difference is mostly structural. A small senior team without account managers or a bench to keep busy can price the work rather than the overhead.

Positioning only — we do not publish other people's figures without a source you can check. All prices on this page are in EUR and exclude VAT; Spanish IVA is charged where it applies, and EU businesses with a valid VAT number are reverse-charged.

QUESTIONS

QUESTIONS — 9

No. They are estimates drawn from past client work, published so you get a realistic idea of cost before we talk. Every project is different and is quoted on its own scope after a scoping call. Once quoted, well-bounded work is a fixed figure against a written scope; larger or open-ended work is priced in phases.

Because the alternative wastes both our time. If your budget is €10k and the work is a €60k build, a call establishing that is a call neither of us needed. Hidden pricing often ends up being priced against perceived budget rather than against the work, and we would rather not start a relationship that way.

Then none of these bands apply and we will say so. A large programme — several systems, several phases, compliance review, dedicated capacity — gets a proposal written for it, usually priced phase by phase with a fresh estimate at each one. Trying to squeeze it into a published band would give you a number that is confidently wrong.

Say so early and we will tell you honestly whether the scope can be cut to fit. Sometimes it can — a narrower prototype, one integration instead of three. Sometimes it cannot, and then we would rather point you toward an off-the-shelf product or a smaller supplier than take an underfunded project that disappoints everyone.

Roughly €8–25k for a single-workflow agent with one or two integrations, €25–75k for a production agent that carries a whole role with retrieval, several integrations and escalation, and €75k upwards for a multi-agent system priced in phases. Those are estimates from past work; the number of systems the agent has to act in and the accuracy bar you need are what move it. Running costs — inference, hosting, tools — are separate and paid by you.

For ongoing and open-ended work, yes: the Ongoing model is a set number of days per month, and phased programmes can run on agreed capacity. For bounded projects we prefer a quoted figure, because it puts the estimating risk on us rather than on you.

No. The first call is free and usually 30 minutes. Discovery — the structured assessment with a written deliverable — is a paid engagement, because it is real work producing something valuable whether or not you go on to build with us.

Yes, and they are yours rather than ours: model inference, hosting and storage. We model these during the build so there are no surprises, and cost optimisation is part of the work rather than an upsell. If you take the system fully in-house, you owe us nothing further.

We do not do equity-for-services or revenue share. It aligns incentives in theory and in practice creates a conflict about scope on every subsequent decision. Quoted fees keep the relationship clean, and we would rather be a supplier you can fire than a shareholder you cannot.

Keep reading

Reference Architectures