Your team is the integration layer. That’s the problem.
We design, build and operate the systems that move work between your systems without a person in the middle — then we keep them working. Nine businesses run on ours.
Start with a measurement, not a proposal
Everyone starts here. You get a document you can act on without us — including handing it to another supplier. That is the point: a paid audit is a decision artefact, an unpaid one is a sales deck. If you go ahead with a build within 90 days, the fee comes off the invoice in full.
Automation Audit
- Process map of how the work actually moves today
- 8–12 automation opportunities, ranked by hours and euros recovered per year
- Build-versus-buy call on each one — including the ones you should not build
- Monthly running-cost model at your real volumes
- 90-day roadmap with owners and sequence
- A fixed quote, if there is something worth quoting
Operations Blueprint
- Everything in the Audit, across departments
- Cross-department dependency map — where the same data is re-typed three times
- Phased 12-month roadmap with a sequencing rationale
- Board-ready business case with the numbers behind it
- 60-minute presentation to your leadership team
Before you buy either one: tell us how many times a month the process runs and how long it takes. If the arithmetic does not clear roughly €30,000 a year of recoverable cost, we will say so on the first call and not sell you an audit. That conversation is free and it takes twenty minutes.
Build
Four bands. Fixed price against a fixed scope, agreed in the audit. 50% on signature, 50% on acceptance — three payments above €25,000. Scope changes are quoted; they are never absorbed silently and then billed as a surprise.
One process, end to end. The thing one person currently does by hand every day.
- 2–5 integrations
- Error handling & retries
- Failure alerting
- Handover documentation
One department, end to end. Intake through to the record in your system of truth.
- 4–10 integrations
- One AI reasoning layer
- Human review checkpoints
- Operations dashboard
A system that decides, not just routes. Reads your documents, uses your tools, escalates when unsure.
- Tool use & retrieval over your documents
- Explicit escalation rules
- Evaluation suite you can re-run
- Full audit log of every decision
The whole floor. Several departments, a fleet of agents, and one screen that tells you what is running and what has stopped.
- Multi-department orchestration
- Supervision dashboard
- Unified reporting
- Operating cadence and review rhythm
Run & Improve
The most common way automation engagements fail has a name: build-and-leave. The system ships, it works, and about ninety days later a credential expires or an API changes and nobody notices for a fortnight. These tiers exist so that does not happen to you. Priced on how many automations are live and how fast we answer — not as a percentage of what you paid to build them.
| Tier | Monthly | Live automations | Included work | We respond within |
|---|---|---|---|---|
| Watch | €450 | up to 3 | Monitoring, alerting, error triage, monthly health report | 3 business days |
| Run | €1,200 | up to 8 | Above, plus 4h / month of improvement work and a quarterly review | Next business day |
| Improve | €2,800 | up to 20 | Above, plus 12h / month and one new workflow each quarter | 4 business hours |
| Control Tower | €6,000+ | unlimited | Above, plus 32h / month, a named lead, and ownership of the roadmap | 2 business hours |
What you are actually buying
Not maintenance on an asset. Twelve months of a system running your operations while nobody watches it — plus the improvements, plus the migrations when the ground moves underneath it.
Model providers retire models on roughly twelve-month cycles with about sixty days’ notice. That means at least one forced migration per system per year, each one needing prompt work and re-testing. If you have someone in-house who can own that, take the system and go — we document every build for exactly that reason.
Twelve-month initial term
Then 60 days’ notice, any time. Not “cancel anytime” — continuity is the entire product, and a system nobody is watching is precisely the failure we are being paid to prevent.
Model and infrastructure costs are yours
Always on your own accounts, always visible to you in real time. We hold delegated access and manage them. We do not mark them up, and we do not bundle them into a fixed fee — a fee that includes token cost is a bet on somebody else’s pricing page, and you would end up paying for that bet either way.
What it costs to run — before you ask us
Almost nobody publishes this, which is why “isn’t AI expensive?” is still a question. Move the slider to your real monthly volume. These are pass-through costs billed to your accounts, not our fee.
The thing worth knowing
Below roughly 20,000–25,000 tasks a month, the fixed platform costs more than the model calls do. A monitored, orchestrated, SLA-bearing stack runs around €660 a month before a single token is spent. Ten thousand customer replies cost about €296. Anyone telling you the tokens are the expensive part has not run one of these in production.
And what we do about it
- Prompt caching — around 39% off a retrieval workload with a stable prefix; profitable from the first cache hit.
- Batch processing — 50% off anything that does not need an answer this second.
- Model routing — the same job spans roughly 46× between a small model and a frontier one. Route by task, not by habit.
- Combined, on document extraction at volume: €1,400 / month down to about €420.
What is in every build, at every price
These are line items on your quote, not goodwill. They are also the entire difference between a workflow that runs and a workflow that runs unattended — and the reason a €300 version of the same automation is not the same automation.
- Error handling and retries on every external call, with backoff.
- Failure alerting to a channel a human actually reads.
- Credential rotation handled, and monitored before it expires.
- Rate-limit and malformed-input handling — the two things that break unattended systems first.
- Daily backups and a tested restore path.
- A monitoring dashboard showing what ran, what failed, and what it cost.
- Handover documentation written during the build, not on the way out.
- AI disclosure and content marking by default — EU AI Act Article 50 has applied since 2 August 2026 and it covers exactly this kind of system.
- Your infrastructure, your accounts. Self-hosted by default.
- Code in your repository from the first commit.
- Every workflow exports as JSON, one click, standard APIs underneath.
- A written exit procedure, agreed before we start.
Straight answers
Why €1,000 just to look at our processes?
Because the output is a document you can act on without us — including handing it to another supplier. If we did it for free we would be writing a sales proposal, and you would be reading one. It comes off the build invoice in full if you go ahead within 90 days, so if we work together it costs you nothing.
Someone on Fiverr will build this for €300.
They will, and it will run, and on day one it will look identical to ours. The difference arrives in month three when a credential expires, an API starts rate-limiting, or someone sends a malformed input — and there is no retry, no alert, and nobody to call. Compare the line items in section 04, not the totals.
That is more than the software costs.
It is. The workflow engine is around €50 a month, and free if you self-host. The software is not what you are buying. You are buying the process design that decides what to automate, the reliability engineering that keeps it running unattended, and someone who answers when it stops. If the software were the hard part, you would already be finished.
Your retainer costs more per year than the build did.
Over twelve months, yes. The build was six weeks of work. The year is twelve months of a system running your operations while nobody watches it, plus the improvements, plus the migrations when models change underneath it. You are not paying maintenance on a finished asset — you are paying for it to keep operating.
Can you guarantee the return?
We guarantee the deliverable and the date. The financial outcome depends on volumes, which is exactly why we measure them in the audit before quoting anything, and why we will tell you when the numbers do not support the work. What we will do is put the recovered hours and euros in writing in the audit, and report against them every month.
Isn’t most AI failing anyway?
The famous 95% figure came from a study built on 52 interviews whose own authors called it only “directionally accurate”. The credible numbers are still sobering: BCG surveyed 1,250 executives and found 5% getting outsized value and around 60% getting very little; S&P found 42% of firms scrapped most of their AI initiatives last year. So the scepticism is earned. The same research also found that bought-in implementations succeed roughly twice as often as internal builds, and that the best returns sit in back-office work while most budgets go to sales and marketing. That gap is where we work.
What happens if we stop paying you?
Nothing stops. The system is on your infrastructure under your accounts, the code is in your repository, every workflow exports as JSON, the integrations are standard APIs that behave the same in any other tool, and the handover documentation was written during the build. The exit procedure is agreed in writing before we start. We are not SOC 2 certified, incidentally — we build inside your environment so your controls apply, and if you need a certified vendor we will tell you that on the first call rather than in a security review.
Twenty minutes, and the volume numbers
Bring how many times a month the process runs and roughly how long it takes. That is enough for us to tell you whether there is an engagement here — and to tell you if there isn’t.
All prices in EUR and exclusive of VAT. Build prices are fixed against a scope agreed in the audit. Model and infrastructure costs are billed to your own provider accounts and are not included in any figure on this page. Running-cost estimates are modelled from published API rates as at August 2026 and from typical token volumes per task type; your own figures will vary with document length, retry rate and prompt design, and we will measure them properly during the audit.