Repetitive work, handed to a computer
We do not sell AI. We take away the keying-in that costs someone half a day every week. We start with the one thing that eats the most hours, and you see the result within the first month.
What this actually is
Automation is not a robot and not the AI from a conference stage. It is software that does a thing somebody clicks through by hand today. It opens the file, moves the data, issues the document, sends it on.
Language models come in where data cannot be described by a fixed rule. Every supplier's invoice looks different, and a customer email can be written any way at all. There a model reads the content and pulls out what is needed. Everywhere else a plain rule is enough, cheaper and more reliable.
We build on n8n, which we install on the client's own server, or we write the software directly when the process is too specific for that. It is a technical decision and we make it after looking at the process, not before.
Where we start
Never with a tool. We start by working out where the hours go, because that is the only way to know whether it is worth doing at all.
- 1We count what takes how longOn the free call we walk through a week in the life of whoever does the most clicking. How many times a day, how many minutes, how often something has to be fixed after a mistake. That is usually the first time a company sees the total.
- 2We pick one process, not fiveWe take the one with the best ratio of hours recovered to difficulty. Usually it is not the most annoying one, it is the most frequent one.
- 3We build it and run it beside the manual workFor the first few days the automation runs in parallel with the person and the results get compared. Only once they match a dozen times in a row does anyone stop doing it by hand.
- 4We measure, then take the next oneAfter a month you know how many hours actually came back. That number decides whether we go into a second process and which one. Without it, it is guesswork.
The whole first process usually takes 2 to 6 weeks. We do not run six-month pilots, because after six months nobody remembers what it was for.
What is actually worth taking off people
This is the list that comes up most often on a first call. We put the usual saving against each one, because without that you cannot decide.
- Retyping orders from email and PDFs3 to 8 hours a weekAn order arrives as a file or an email body and somebody moves it into a spreadsheet or a system. A model reads the content, pulls out the line items and puts them where they belong, and a person only approves.
- Issuing repetitive documents2 to 6 hours a weekQuotes, confirmations, reports and summaries assembled every week from the same parts. The document gets built from data already in the system instead of being glued together by hand.
- Connecting systems that do not talk4 to 10 hours a weekSales in one, stock in another, accounting in a third. Data travels by itself, one way or both, instead of through the clipboard.
- Reading supplier invoices2 to 5 hours a weekEvery supplier lays an invoice out differently, so a fixed rule will not do. A model pulls out line items, amounts and dates, and parks anything uncertain for a person to check.
- Searching company documents1 to 4 hours a weekContracts, manuals and old quotes where nobody can find a clause from two years ago. The search answers in a sentence and shows which document it came from.
- Watching deadlines and shortages1 to 3 hours a weekInstead of a person going through a list once a day and phoning around, the notification goes to the right person by itself, the moment something crosses a threshold.
If the count shows the problem is not retyping but the lack of one shared place for data, it is more honest to say so and build software around your processes rather than automate the mess.
How many hours usually come back
An example assembled from the cases we see most often in companies of 20 to 60 people. One person in the office, three processes taken off in sequence.
- Retyping orders from email
- 5 hours a week
- Issuing repetitive documents
- 3 hours a week
- Moving data between systems
- 4 hours a week
- Fixing mistakes caused by retyping
- 2 hours a week
- Total
- 14 hours a week
- Build time for three processes
- 6 to 10 weeks
At fourteen hours a week, automation usually pays for itself in 2 to 4 months.
This is an example, not your quote. We do the real arithmetic on the free call, on your numbers, and only then give you one fixed price in writing.
Where automation does not pay off
We say this on the first call, before you work out a budget. In these four cases we advise against it and explain why.
- When the process changes every monthAutomation pays off on repetition. If the rules keep shifting, maintenance eats more time than it returns. Better to stabilise the process first.
- When it happens once a quarterA task taking two hours four times a year is eight hours annually. Building and maintaining the automation will cost more. The numbers have to be counted, not felt.
- When the decision needs a personA model reads and transcribes well but takes responsibility badly. Wherever a mistake has a real cost we leave a person to approve, and we say so plainly instead of promising it runs untouched.
- When the data is in no state to be trustedAutomation on dirty data produces errors faster than a person did. Order in the data first, automation second. Sometimes that means starting not with automation but with software that keeps the data straight.
We have talked several of these out of being built. We would rather lose the job than leave a company with automation nobody trusts, switched off six months later.
Where your data ends up
Almost everyone asks this, and rightly. Automation with a language model means the content of a document travels somewhere, so you need to know where.
By default we install n8n on the client's own server rather than in a vendor's cloud. The flows, the data and the history stay with you. We send the model only the fragment needed for the specific task, never the whole database.
- Fully in-house optionWhere data is sensitive, the model itself can also run on your hardware. It is slower and weaker than the best available, so we say honestly when that trade is worth it and when it is just more expensive.
- Only the fragment neededWe do not send whole documents when one section will do. Less data leaving means less risk and a lower cost per call.
- You can see what happenedEvery run leaves a trace: what went in, what came out, what failed. Without that you can neither trust the automation nor fix it when it starts getting things wrong.
- Personal data handled separatelyIf employee or customer data runs through the process, we settle at the start who has access and how long data is kept. The same approach we took in Auresta RCP.
Questions we hear most often
Q.1How much does process automation cost?
You get one fixed price in writing after a free call. The figure depends on how many processes are involved and whether the systems we connect have an integration interface. On the call we work out how many hours a week manual work costs you and how many months until it pays back. Market ranges are covered separately on the blog.
Q.2How long does it take?
The first process usually 2 to 6 weeks. Shorter when it is about moving data between systems that have an integration interface. Longer when documents with varying layouts have to be read and the setup has to learn from your examples.
Q.3Are you an n8n partner?
No. We build on n8n and install it on the client's server, but we are not an n8n partner and not certified by them, and we will not describe ourselves that way. You can check the partner directory in under a minute, so any other answer would be foolish.
Q.4Does my data go to an external model?
Only the fragment needed for the task, and only when a model is needed at all. n8n itself we install on your server by default, so the flows and history stay with you. Where data is sensitive the model can run on your hardware too, and we say honestly what you lose by doing that.
Q.5What if the automation gets something wrong?
We assume it will and build for that. Anything uncertain goes to a person to approve, and every run leaves a trace of what went in and what came out. For the first few days it runs alongside the manual work and we compare results before anyone stops doing it by hand.
Q.6Do we have to replace the systems we use?
No, and usually there is no reason to. Automation is designed to attach to what you already have. Replacing a system is a separate decision and a separate project, which we only discuss when the old system genuinely blocks any sensible connection.
Q.7Who maintains it when a form or a supplier changes?
Automations need looking after, because the world around them shifts. You can maintain them yourself, since you get access from day one, or take a care plan from 490 zł a month. We raise this before the start, not after the first breakage.
Q.8Where do I start if I do not know what to automate?
By counting hours, not by choosing a tool. On the free call we walk through a week for whoever does the most clicking, and out of that comes a list sorted by how much can genuinely be recovered. Very often the most annoying task turns out not to be the most expensive one.
Q.9Will AI replace my staff?
Not in the uses we are talking about here. We take away the retyping and clicking, the part of the job nobody enjoys and where mistakes are easiest. The usual result is that the same person stops sitting over a spreadsheet until the evening, not that they stop being needed.
Q.10Will I see how much I saved?
Yes, and we treat that as part of the job. We measure the number of runs and the time that used to go into doing it by hand, so after a month there is a concrete number of hours. That number then decides whether a second process is worth doing at all.
Specific cases, one at a time
Six narrower topics, each about one specific kind of automation we build most often: the tool itself, the general pattern for connecting systems and the two most common integrations on that list, plus document flow and quote generation.
See how it works inside
We document our n8n workflows the way they actually behave: what they do, what they will not do, which version they were verified on and what breaks in the edge cases. The files are there to download, no form in the way.
Before you ask for a quote
Pieces answering the cost questions we get most often before a first call.
Tell us what still gets done by hand
A free call, where we count the hours. Then a fixed price and a date in writing, with no commitment.
