Process automation & enterprise AI
From a single workflow to your own agent team
Most businesses start with one workflow that eats time – incoming invoices, email enquiries, quotes. Some then realise they actually need something bigger: a model that knows the whole business, and agents that work with each other. We build both – at a fixed price agreed in advance.
How we work
Prompt engineering is the entry level. We build on top of it.
Being able to instruct a language model well is a start. For a business that needs to rely on it, that is not enough.
Prompt engineering
Explaining clearly to the model what it should do. Enough for a single answer. One attempt, no way back: if it is off, nobody notices.
Loop engineering
The agent works in loops: it checks its own result, fetches missing information, corrects itself – and only reports back when it is finished or genuinely stuck. That is how you get a reconciliation that chases instead of giving up.
Graph engineering
Several agents with clear roles and handovers, like an org chart: who researches, who decides, who checks – and where a person has to sign off. Nothing runs in circles, nothing passes unchecked.
It sounds technical. It is the difference between a chatbot and a system you can hand a working day to.
The next level
Your own enterprise AI model with an agent team
A language model that knows your business: your documents, your processes, your wording. Plus agents that do not just answer but work – writing quotes, maintaining data, producing reports, connecting systems. Some people call that an enterprise operating system; the name matters less than the division of labour behind it.
- Every agent has a remit – and sees only what it needs for that.
- Outbound steps such as emails, invoices or publications require sign-off.
- Hosted in the EU or on your own infrastructure. Your data stays yours.
- It grows with you: you start with one agent, not twenty.
The precondition
An agent team is only as good as its access to your data
An agent that cannot see the staff rota cannot re-cover a shift. One that cannot see supplier prices cannot check an order. Anyone serious about working with AI today has to give the team access to the whole business – not to a slice of it.
People
Staff rota, time records, employment contracts, payroll, holiday and sick leave.
Purchasing & stock
Orders, delivery notes, supplier prices, incoming invoices, stocktake and inventory.
Sales & POS
Point-of-sale system, revenue per item and day, bookings, cancellations, price lists and promotions.
Communication
Phone, email, WhatsApp, Instagram, Facebook, review sites – inbound as well as outbound.
Finance
Bank movements, open items, documents, reports for your accountant.
Operations
Opening hours, seasonal calendar, weather, reservations, maintenance and equipment.
Access alone is not enough
Data sitting as a pile of PDFs on a drive effectively does not exist for an AI. It has to be stored and described so an agent finds it in seconds: searchable, cross-linked, tagged with date and ownership. That is the unglamorous part of the work – and the part that decides whether it succeeds or fails.
Full access is not open access
Every agent gets exactly the areas its job requires – the Instagram agent never sees a payslip. It runs in the EU or on your own infrastructure, outbound steps require sign-off, and every access stays traceable.
We connect what is already there
No system change, no “that only works with our software”. If a program has an interface, we use it. If it has none – and with POS systems, rota tools and supplier portals that is the rule rather than the exception – we operate it the way a person would: through the interface, the export, the inbox. Nothing changes for the business, except that the work disappears.
Only once that is in place does an agent team really take work off your hands – at a scale you could not pay for in staff. The reach of the business multiplies while payroll stays where it is. What you pay for on an ongoing basis is compute, not salaries.
An example from daily operation
Every channel knows whether you are open today
For a weather-dependent business this is the single most common question. The team sets the status with one tap in the messenger – with a time if needed, “open from 4pm today”. From that moment every agent knows it.
- The phone assistant knows it from the first second of the call – no lookup, no waiting for the caller.
- The Instagram agent factors it into the day plan, and a widget on the website shows the same status.
- Every entry is tied to a calendar day: yesterday’s “open” can never carry over to today.
- Each morning the system checks the weather itself and asks if nothing has been set for today.
Two of them are already running – here in detail
Invoice and bank reconciliation
Document to transaction every morning, chased until everything is complete.
View →Instagram agent
A post a day by seasonal calendar, from real photos, with human approval.
View →What else businesses have us automate
Answer email enquiries automatically
Standard enquiries take care of themselves – your inbox stays free for the cases that really need attention.
Generate quotes automatically
A web form becomes a finished, ready-to-send quote – in minutes instead of days.
Custom workflows
If the problem exists, an automation for it usually exists too. We’ll find it in the discovery call.
Fixed price instead of a running meter
Fixed price after a free discovery call, typically €1.000–2.500, delivered in 1–2 weeks.
Project management
Interim project management
Experienced project management (classic & agile) for your initiative – flexible, remote or on site, billed hourly or per project.
Book a free discovery call