Your ERP covers most of what the business does. What is left over sits in a spreadsheet one colleague has maintained for years, in a folder of PDF invoices, and in three steps somebody works through by hand every Monday morning. That is what most mid-sized companies look like on the inside, and for the larger part of it that is fine. It gets expensive where the manual work grows along with the business, with every new customer, every new site, every extra spreadsheet. ARSoftware builds software for exactly those places.
Off-the-shelf software is usually the right answer. This page is not written against it.
Accounting, payroll, stock control, time tracking: processes that look much the same in every company are covered better by a finished product than by anything you build yourself. The vendor spreads its development cost across all its customers, it keeps up with changes in the law, and if it goes under there are competitors offering the same feature set. Build your own in those areas and you pay for software you could have rented, then maintain it on your own as well.
Two points make that flip. Start with your process. Yours deviates from the standard, and for a reason you want to keep: the deviation is often exactly what customers buy from you and not from a competitor. A finished product then forces you to bend the business around the software instead of the other way round. Then there is your data. It sits in several systems, and no product on the market joins up precisely those. So the join gets made by hand, in a spreadsheet, again every week.
A finished project shows what that looks like. PrimeBoost, an Amazon marketing agency in Munich, drew its client figures from three sources: the Amazon SP-API, Seller Central and its own CRM. For every report an account manager pasted the numbers together and checked at the end whether the totals added up. Across more than 20 seller accounts that ate roughly 15 hours a week. No product you can buy joins up those particular three sources. Since May 2026 a system of their own does it, and the 15 hours are gone.
Usually both answers are right, just in different places. The finished product stays where it fits. What gets built is the remainder.
Three kinds of system, sorted here by the work they take away.
Pulling data out of one system, checking it and entering it into another, processing documents, writing results back. We settle up front which steps run without asking and which need a sign-off, and we log every one of them. Where a language model is needed for that, we use one. Where a rule is enough, we do not.
Your ERP, your CRM, the shop and the file from the tax adviser do not talk to each other. We build the connection between them, put the figures in a database and set the reporting on top of that. A language model comes in where it saves a step, at the querying for instance, when somebody types a question instead of clicking through filter menus. Model access is encapsulated, so switching provider later does not mean rebuilding.
Search across your own documents, drafts for letters that recur, analyses that would otherwise live in spreadsheets. We start with one use case a team uses daily, instead of a platform for everything. What proves itself, we extend. What sits unused, we drop.
If the data is not to leave the building at all, part of this exists as a finished product for hardware you own, with the processing staying in house.
A corporate has an IT department to take a new system over, and a budget that survives a failure. Here both are missing.
In a corporate, IT puts a new system into service. In a mid-sized company that is often done on the side by whoever looks after the computers, sometimes by an external IT firm whose actual job is the network and the desktops. So operating documentation and training come with the delivery, in the language of the people who run the system afterwards. If there is nobody in house for it, we run and maintain the system ourselves on request.
A project that runs for a year and produces nothing usable until the end is a write-off at a corporate. Here it is a serious problem. So we cut the first step small enough to be worth something on its own, and we put the effort estimate for it in front of you before you commit to more. After each step you can stop.
A corporate rarely has to ask its supplier that. You do. ARSoftware is a very small company, and anyone ordering a system they will depend on daily for years wants to know beforehand what becomes of it if we are no longer around.
The commitment on this has held since the first project and is not being reinvented here. You get the full source code, the documentation and the access, so another team can carry on without starting from zero. We build with widely used technology and encapsulate model access, so nothing is tied to us as a supplier. Response times and what happens during a longer absence depend on how critical the system is for you, which is why they belong in the contract rather than on a marketing page.
Whether this programme is open to you is decided by the size of your company. The threshold sits lower than the word Mittelstand suggests, which is why it comes first here.
The programme funds small commercial enterprises with an establishment in Bavaria, fewer than 50 employees and no more than 10 million euros in annual turnover or annual balance sheet total. At that threshold the word Mittelstand misleads. The IfM Bonn counts companies with 500 or more employees as Mittelstand too, as long as they are owner-managed, whereas the Digitalbonus means something considerably smaller. Employ 80 people and you are Mittelstand and still get nothing out of this programme, however well the project itself fits.
For everyone below it: up to 50 % of eligible expenditure, capped at 7,500 € under the Standard part of the programme and at 30,000 € under Plus. Plus requires particular innovation content, and the guideline names AI as an innovation focus. Nothing at all is funded below 4,000 € of eligible expenditure. Anyone planning to start with a potential analysis sits under that floor and has to file for the project that follows it.
Applications are filed online and, since May 2025, only through the applicant company's own ELSTER-Unternehmenskonto. Bavaria releases a fixed quota of applications each month, and filing opens on the first working day. How far the quota goes depends on demand. Per funding area you get one application across the whole programme, one for digitalisation and one for IT security, so applying for a project in stages is not possible. The programme runs to 31 December 2027, and what follows it has not been decided.
An application is decided by the responsible authority, not by us.
Last updated: August 2026. The Digitalbonus Bayern figures come from digitalbonus.bayern and were checked on 9 August 2026. The Mittelstand definition comes from the IfM Bonn. Funding conditions change, and this page is not a statement from the granting authority.
Company size is the wrong measure. What matters is how much working time a process costs every week and how much longer it is meant to run that way. A firm of 15 where two people each spend a day a week copying data between systems has the clearer case than a company of 200 whose processes a finished product covers cleanly. Weigh the hours per year against the one-off build cost plus running it. If the sums only work out after many years, we advise against it, and we say so.
We do not quote flat prices, because the price follows the scope. The day rate of 1,600 € net is published, and the ranges on our price page are calculated from it. What decides the figure is how tightly the use case is scoped, how many systems are connected and what condition they are in, the data-protection and hosting requirements, and running the thing afterwards including model costs. After the analysis you get an effort estimate for one clearly bounded first step, before you commit to anything larger.
Yes. We do not assume an IT team of your own. What it takes is one person in the building who knows the process that is meant to improve, not somebody who looks after servers. Operating documentation and training come with the delivery, in the language of the people who will run the system. On request we run, monitor and maintain the system ourselves, so nobody on your side picks up an extra job.
You keep it. Every project hands over the full source code, the documentation and the access. Because it is built with widely used technology and the model access is encapsulated, another team can carry on without starting from zero, and nothing binds you to ARSoftware as a supplier. That same commitment sits on the custom AI software page, which also answers what applies if we become unavailable.
Tell us briefly what you are working on. You get an honest assessment of whether AI is the right fit, and what a sensible first step looks like.
Or email us directly: contact@arsoftware.tech
