- Hours of your week disappear into triage, filing, scheduling and retyping.
- You want AI, but not at the price of client data in an American cloud.
- You want a quick, contained first project before committing to anything bigger.
AI & AUTOMATION · WHERE MOST CLIENTS START
Get your hours back.
Keep your data.
Automations that clear the busywork away, a private assistant that knows your documents, and a voice assistant that answers your phone, built on the systems you already have. Fixed prices from €950, excl. btw. And unlike the typical AI stack, it runs on your own servers: the model comes to your data, your data does not go to the model.
I · THE PROBLEM
“Every week disappears into triage, filing and retyping, and nobody has time to fix that.”
Underneath the AI question sits a quieter one that is worth more money: the busywork. Leads that need triaging, documents that need filing, meeting notes that need to become tasks, invoices that need drafting. Some of that needs a model. A lot of it needs a plain rule, well placed. Part of my job is telling you which is which.
Every business is also being told to adopt AI, usually by companies selling it. The pitch is always the same: hand over your documents, your inbox, your client conversations, and intelligence will come back. Most of what gets built this way is disposable, expensive, or both, and almost every AI product wants your data in someone else’s cloud on top of it, a trade this company exists to end.
Meanwhile, the real AI risk in most companies is already in the building: employees pasting client data into free chatbots, under terms nobody read, into training sets nobody can audit. A ban does not work; people use what helps them. The only durable fix is a sanctioned alternative that is genuinely better, and runs on your own systems.
II · WHAT IT IS, CONCRETELY
Four shapes, one rule
Automation work at Ironstack comes in four shapes, from a plain workflow to an assistant that knows your documents. Wherever AI is genuinely the right tool, one rule holds throughout: the model comes to your data, your data does not go to the model.
The four shapes · fixed prices, excl. btw
50% on signing · 50% on go-live · managed under retainer for €75 a month per component
Workflow Automations connect the tools you already use, no AI required, and priced accordingly: leads route themselves, documents land in the right folder, invoices go out on schedule. AI Automations take on the busywork that needs judgment or language: leads triaged and answered, documents read and filed, meeting notes turned into tasks. The Private AI Assistant is your own chat assistant on a European server, connected to your company documents, answering your team in plain language with sources. The AI Voice Assistant answers the phone, takes bookings and handles intake against your calendar and systems; it runs at its first paying client today.
What you hold afterwards: the workflows and prompts documented in your runbook, the automation platform under your accounts, and a clear written record of which model runs where. The honest version, stated up front: small open models run on your own hardware, and some tasks need bigger ones. When they do, I use an EU-hosted model under a written agreement, disclosed in your documentation, and only for the data that task needs. You always know which is which.
See the full price register for how these fit with everything else.
III · A FAIR MATCH
Who this is for, and who it is not for
- You expect AI to replace a department. It replaces busywork; people keep the judgment calls.
- You want the newest model demo-of-the-week. I build boring, reliable versions of proven patterns.
- Your problem is a plain rule wearing an AI costume. Then I will build you the rule, cheaper, and say exactly that in the proposal.
IV · WHERE IT SITS
The entry for quick wins
AI & Automation is one of the two entries, alongside the Audit. It works on the systems you already have, which makes it the fastest way to feel what owned infrastructure is like without migrating anything. Smaller projects follow the same discipline as the five-phase method, just lighter: scope, build next to production, validate, hand over documented.
Automations run best when they can reach every system, which is why they get stronger on the Stack or after a migration, where everything lives on the same server under your keys. And like everything else, automations end up under Managed Sovereignty: monitored, updated, and reported on in plain language every month.
Fair questions
Can a small company use AI without sending data to American clouds?
Yes, if you are deliberate about it. Small open models run on your own hardware, and when a task needs a bigger model, I use an EU-hosted one under a written agreement, disclosed in your documentation, and only for the data that task needs. You always know which is which.
What does an AI assistant for a business actually do?
It works where you already talk, in Telegram, WhatsApp or your team chat, and on the phone if you want it to have a voice. It searches your files, books appointments, creates tasks and drafts invoices, all against your own systems rather than a third-party copy of them.
Do we need to replace our systems before adding AI?
No. AI and automation work is the entry for quick wins, built on the systems you already have. It is also, in practice, how many clients discover which of their systems are worth owning, but that step is yours to take or skip.
What does AI and automation work cost?
Workflow Automations, plain integrations with no AI involved, are €950 for up to three, extra workflow automation €250 each. AI Automations, where judgment or language is involved, are €1,500 for up to three, extra AI automation €400 each. A Private AI Assistant is €2,000. An AI Voice Assistant is €3,000. All prices are fixed, excl. btw, billed 50% on signing and 50% on go-live, and managed AI components add €75 per month under the retainer. If it turns out your problem does not need AI, just a plain workflow automation, the proposal will say that, because it is cheaper and more reliable.
Will the assistant hallucinate things to our clients?
Any language model can be wrong, which is why mine answer with sources from your own documents, act only through defined tools, and get a human handover the moment a conversation leaves the script. Where an error would be expensive, the assistant asks instead of assuming; that boundary is part of the design, not an afterthought.
Employees already paste things into free chatbots. Is that a problem?
It is the biggest AI risk in most companies today: client data flowing into tools nobody vetted, under terms nobody read. The fix is not a ban that everyone ignores, it is a sanctioned alternative that is genuinely better because it knows your business. The audit can map your current exposure.
Start a conversation
Tell me what eats your team’s hours and where your data goes today. You will get a straight answer, including whether your problem actually needs AI at all.