A process assessment that ends in working AI automation.
The assessment names the one process where the most manual hours disappear at your place, and it says so when this is not your next step yet. Whatever it recommends, the same team builds it and takes it live. For companies of 5 to 50 people with no dedicated IT team.
12+ projects delivered
every one on deadline, in production
Fixed price and deadline
in writing, and if we slip, the next payment waits
Reply within 24 hours
on working days, to every enquiry
In writingAfter the assessment
Two stages, two prices: the assessment first, then the build.
Where the manual hours actually go
The spreadsheet was the right call. Now five people write to it.
Someone copies a figure out of the invoicing tool, someone types in an order, and at month end two places disagree.
The same figure gets entered by hand, more than once.
Out of the invoicing tool, into the spreadsheet, onto the delivery note. Whoever does it knows which step it slips at.
How it works lives in one person's head.
While they are in, it works. When they are on leave, somebody else rings round asking where things go.
Everyone has heard of an AI project that stopped at the demo.
It looked good in the presentation and never went live. That story holds back more companies than the manual work does.
Our answer to each of those
We start with one process, the one carrying the most manual hours, and build it all the way to live.
One process, named
The assessment walks your working week and names the one process that costs the most manual hours. What works in the spreadsheet stays there.
The work we take over
Data read out of invoices and orders, inbound mail sorted with a draft reply, a quote drafted from your own price list. The proposal states which step the model sees and when the data is deleted.
The process leaves one head
What one person knows today goes into the system: which step runs when, and who approves it. It runs the same through annual leave.
A single process is the first step
The first process stands on its own, even if we never build on top of it. If the assessment finds it is not worth doing yet, the document says so, with reasons.
If it feels as though everyone else already runs on AI: Eurostat’s 2025 survey puts AI use at roughly a fifth of EU enterprises on average, and at 17% among the EU’s small enterprises.
What holds companies back
Of EU companies that considered AI and did not adopt it, this share names missing expertise as the first barrier.
"We have nobody who understands AI"
You do not need one. You know the process, we know the AI. Your team learns to use the system; running it stays with us, under monthly maintenance.
This share is held back by legal uncertainty.
"What about the legal side?"
There is a concrete rule now and we build to it: since 2 August 2026 the EU AI Act requires an AI answer and AI-written content to be labelled. We build to that transparency rule wherever the system runs, because it is the strictest regime we work under. The high-risk tier that gets the press starts at the end of 2027 and does not touch a typical SME process.
This share names data protection as the obstacle.
"Where does our data go?"
Where the proposal says. We record which model sees what and configure it so the provider does not train on your data. Every AI step is logged, so it can be traced afterwards.
Eurostat's 2025 EU enterprise survey. Industry research, not our own measurement.
What each role gets
For the owner
A written monthly figure: the hours the system replaces, and what it cost.
The return is a number you track, and the system belongs to the company, credentials included.
For office and finance
Invoices, requests and email arrive pre-processed, ready for a decision. Typing and copying leave the day, and the decision stays with you.
For whoever does the routine work
The system writes the first version: the reply, the summary, the translation. You read it over, and the decision about what goes out stays yours.
A worked example of a built process
An inbound quote request, once we have automated it for you
The steps are the ones you run today. Fewer of them go through a person.
Inbound email
The enquiry arrives in the shared inbox, exactly as it does now.
Extract
The system pulls out who is asking, what for, and in what quantity.
Draft quote
An itemised draft is assembled from your own price list, in your usual form.
Approval
A person reads it, corrects it, and decides what goes out.
Send
The approved quote goes back to the sender and is filed with the thread.
The draft reply, before anyone approves it
Three steps on one screen: the incoming email, the details read from it, and a draft built from the price list, which a person approves.
Incoming email
- Sorted as
- History
- Price list
Read from it
- Company
- Item
- Item
- To check
- Delivery
- Deadline
Draft reply
To: Example & Co., purchasing
Prices: Sample Ltd. price list, October
Who AI automation pays off for, and who it does not
This fits you if
- One process eats a lot of manual hours every week.
- The data the work runs on already sits in a system you own.
- You expect a measurable result, counted in hours.
We would point you elsewhere if
- The process runs a few times a month. Automation costs more than it returns there, and we say so.
- You want a chatbot on the website. That rarely pays for itself on its own.
If the process needs a system before it needs AI, this is where we would start.
The package
First process
- One chosen process, assessed and built
- A clickable preview on your own data, at every stage of the schedule
- A monthly written measurement: the process's hours, before and after
- Human approval on the sensitive steps
- Labelling and logging built to the EU AI Act standard
- The system and every credential are yours
If an off-the-shelf workflow would serve your process, we say so on the call.
In writing
After the assessment
How many steps does the process have, and how many systems does it touch? Sorting incoming email costs less than document processing wired into invoicing.
Then: maintenance and growth
A monthly plan with a fixed response time: oversight, updates to the model and its instructions, the next process. With a rented AI tool the fee runs per user, and it runs for as long as the company uses it. This fee is set in the first proposal and it is optional: the system is yours, credentials included, and runs without us.
Request a process assessmentWhat we have already built
Gloster Content Engine
We built an internal AI content platform for a software company, and it runs in production. It drafts from a brief and from research, in several languages, and a person decides what goes out.
There is no figure beside it on purpose: the return is measured by the client on their own data, and we do not claim what we did not measure.
Companies we've built for
How much does AI automation cost?
The price list says the price comes after the assessment, and two things decide the number: how many steps the process has, and how many systems have to be touched. The assessment is its own stage with its own price, so the build is a decision you make afterwards.
Two separate products. The free audit answers one question: is your company ready for AI today, and is it worth starting at all. You get a short written answer to that, and an offer of a call. The assessment is the paid stage and it is the full walkthrough: which process burns how many manual hours, what has to be connected to it, and what order to take them in.
About 40% of enterprise AI projects that start ever reach production (Gartner data, via Portfolio.hu's 2026 summary). That is what the assessment is for: it names the process, the integration points and the hours you can expect to save before anybody writes code. What gets built after that runs in production.
If you have the capacity and the experience, build it in-house, that is the cheapest route. A 2025 MIT study found that externally purchased AI solutions succeed about 67% of the time against roughly 33% for internal builds. In-house work sits alongside the day job, and urgent tasks tend to take the time back. We take it through at a fixed price and a stated deadline, then hand it over so it runs without us.
One week. We walk the processes, find where the manual work goes, and check which step has usable data behind it. You end up with a written document: what is worth automating, in what order, and what it costs. The build after the assessment typically takes 2 to 4 weeks, on the schedule the proposal sets.
The company is, the same as today when a colleague sends the email. So the system is built to keep that manageable: at a step that goes outside, a person approves it, and there is a written trace of what went in, what the model returned, and who let it through. When something does go out wrong, that trace is what shows where to fix it.
It is the most common situation there is, and it is not the team's fault: nobody gave them an approved route to work on. The first step of a rollout is that route. One process where it is written down which model sees what, where the decision stays with a person, and what never leaves the company at all.
The material is yours either way, and it is written so somebody else can work from it: which step connects where, what data it needs, in what order. There is no pre-contract behind it, and nothing to pay to take it with you.
Start with the process assessment
It starts with a 30-minute call: we pick the process where the most manual hours disappear, and if this is not your next step yet, you hear that too.
- 12+ projects deliveredevery one on deadline, in production
- Fixed price and deadlinein writing, and if we slip, the next payment waits
- Reply within 24 hourson working days, to every enquiry