What does custom software cost?
Most suppliers answer this with "it depends". That is even true, and it helps nobody who has to put a budget together. This page gives the figures ARSoftware calculates with, and says what moves them.
The short answer
A consulting day costs 1,600 €. All prices are net and exclude VAT. Every line we bill by time is calculated from that rate. It sits at the top because it lets you check the rest of this page.
Two pricing models sit side by side here, and anyone who conflates them misreads every figure below. Consulting is billed by time. You buy days, and what a day costs has nothing to do with the size of your company. The local AI suite is licensed instead. Whether a company falls into band S or band M is decided by its user count, not by the work we put in.
| Service | Price |
|---|---|
| Consulting day (workshop, concept, audit) | 1,600 € |
| Half day | 950 € |
| Travel day outside Bavaria | 800 € |
| Change requests after acceptance (hourly rate) | 200 €/hr |
| Support outside the maintenance contract | 220 €/hr |
How the size bands are cut, and what the introduction and the annual licence cost in each, is here.
| Segment | Users | Introduction (one-off) | Annual licence |
|---|---|---|---|
| S | up to 20 | 9,000 €–15,000 € | 4,000 €–6,000 € |
| M | 20–50 | 15,000 €–25,000 € | 7,000 €–12,000 € |
| L | 50–150 | 25,000 €–40,000 € | 14,000 €–24,000 € |
| XL | from 150 | from 40,000 € | from 30,000 € |
What that prices is the setup work, not the functional scope. The modules you run are billed on top, and their prices are on the local AI suite page.
What the suite does is on the local AI suite pageWhat is known up front and what is not
You know what the analysis costs before you commission it. Three packages are on offer. One day on site with a short report is the AI potential analysis. Next up is the Voranalyse, which delivers architecture, data flows, a data protection concept and the effort estimate. Widest of the three is the extended concept, which also prepares the data protection impact assessment and includes a draft Betriebsvereinbarung.
| Package | Scope | Price |
|---|---|---|
| AI potential analysis | 1 day on-site plus short report | 1,900 € |
| Voranalyse & architecture concept | 3-5 days, architecture, data flows, data protection concept, effort estimate | 5,500 €–8,500 € |
| Extended concept incl. DSFA preparation and Betriebsvereinbarung draft | 6-9 days | 10,000 €–14,000 € |
If you commission the overall system within 3 months of the Voranalyse, 50 % of what the analysis cost is credited against it.
Full custom development with no product underneath it appears in no table. That one is priced on request, and this is not a dodge. Effort depends on things nobody knows before the analysis, so a number written here would either be too high or a loss for us. What we can name up front is the day rate and the price of the analysis the estimate comes out of.
When it really is built from scratch
Projects with no product underneath them always start the same way here, with a workshop. One day, when a single team and a single process are at the table. Three, when several departments are affected, legacy systems have a say, and nobody has yet written down what the finished system is meant to do. Its length is decided by the scope and the complexity of the undertaking, and we settle that in conversation beforehand rather than on the invoice afterwards. The workshop is billed by the day, at the rate in the table further up.
The analysis follows. We look at the systems already running at your company, at the state of the data, and at the processes the new system will touch. Out of it comes the architecture concept with data flows and a data protection concept. What that costs is in the table directly above. The Voranalyse covers the ordinary case. The extended concept covers the one where the works council and the data protection officer need documents that somebody has to write.
The effort estimate comes at the end of the analysis. There and not earlier, because before that nobody knows what your database looks like from the inside. The assumptions it rests on are written next to the number. Drop one of them and we recalculate, and you can see why. Anyone who wants to stop after the workshop stops.
An example from practice
PrimeBoost is an Amazon marketing agency in Munich. Between October 2025 and May 2026 we built the reporting system its team uses today to analyse more than 20 seller accounts. What the project cost is not on this page. We are cleared to name the client, not the figures.
What drove the effort can still be named. Three data sources had to come together, one of them a CRM that was already running. Connecting to the Amazon SP-API was built specifically for this. And October to May is calendar time, not a statement about effort.
What a comparable undertaking would cost at your company is something we say after the Voranalyse. Saying it earlier would be guessing.
What drives the price
The ranges in the tables are wide. Five things decide where a project lands inside them.
-
How tightly the use case is scoped
A use case you can write down in two sentences can be estimated. "AI for sales" cannot. The distance between those two phrasings costs money before the first line of code exists.
-
Which systems get connected
Every connected system needs a connector, and what that costs is decided by the interface. A documented REST API is configuration work. A twelve-year-old database with no documentation is archaeology. How many systems there are matters less than the condition of the worst one, because a single legacy system pushes the range up on its own.
-
What data protection demands
Whether data may leave the building decides the architecture and with it the hardware. Local operation saves contracts and costs machines, and EU hosting works the other way round. Once the works council and the data protection officer are at the table, documents come with them, and somebody writes those.
-
How thoroughly you test and document
Acceptance against test cases agreed in advance takes longer than "well, it runs". With AI systems it is still the only way to find out whether the output is any good. Ask for an evaluation built from a hundred real questions and you get one, and you pay for it.
-
What running it costs afterwards
A system nobody looks after is dead in two years. Maintenance, model costs and the question of who is reachable when it breaks belong in the first budget round, not the second.
Custom software or off-the-shelf
Off-the-shelf software almost always bills per user per month. That invoice grows with headcount and never stops. You pay for a build once and run it afterwards, and running it costs less than building it. In the table above, the annual licence sits below the introduction in every size band. Which of the two invoices is smaller over three to five years depends on your user count and on your vendor’s licence price. We do not know the second one.
Buy off-the-shelf when your process is the one everybody has. Accounting, payroll, an ordinary CRM: a vendor has spent ten years building in the special cases you would otherwise have to find and pay for yourself. Buy it when your team is small, because the one-off cost then spreads across few heads. And buy it when nobody in the building wants to own a system of your own. A subscription somebody can cancel is more honest than an application nobody touches after two years.
The same applies to us. We sell the AI suite as a licensed product rather than a one-off build, because most of it cannot be built again for every customer. What becomes custom is what makes your company different from the next one. That is the connectors to your systems and the rules your decisions follow.
Running costs
The invoicing does not stop at acceptance. Operation is the first item: maintenance, support, updates. With the suite the base scope sits inside the annual licence, and what the higher maintenance tiers and the maintenance of your own connectors cost is on the local AI suite page.
Model costs are the second, and they are the line most often missing from a budget. They move with usage and with the provider’s pricing. Buy a model through an API and you pay per request, so if usage doubles, the invoice doubles. Run the model on your own hardware and the item shifts into electricity, depreciation and eventually a new server. It does not go away.
The third item is further development, and that one you control. New fields in the CRM, a changed process, a function the first scope left out. That runs as a change request billed by time. And who looks after the system in year three, once the person who introduced it has left? That question costs more money than most technical ones.
How we estimate
We estimate once we have seen something. Discovery comes first: the process, the state of the data, the systems already in the building. Out of that comes an effort estimate for one clearly bounded first step rather than for the whole undertaking, because a single number covering two years of work would be a guess and would do you no good. How a project runs after that is described under AI software development.
Common questions about cost
What does custom software cost compared with off-the-shelf software?
Off-the-shelf software is usually paid per user per month, indefinitely. Perpetual licences, per-device licences and consumption-based pricing exist too. A build is paid for once and run afterwards. Which of the two invoices is smaller over three to five years depends on your user count and on your vendor’s licence price. For small teams, and for processes that look the same in every company, off-the-shelf wins nearly every time. Building becomes interesting where the licence sum returns every year and the process is your own.
Why do you not quote flat prices?
Because a flat price set without knowing your systems either carries a markup for our risk or turns into our loss. What is quoted instead is the day rate, fixed prices or narrow ranges for the entry packages, and ranges for the introduction of the suite by company size. A dependable effort estimate follows the Voranalyse, once somebody has looked inside your database and your processes.
What does a day of consulting cost?
A consulting day costs 1,600 €. All prices are net and exclude VAT. Half days and travel days outside Bavaria carry their own rates, which are in the table on this page. Workshops, architecture concepts and audits are billed against it. The Voranalyse does not run on the day rate. It has a package price.
How high are the running costs after go-live?
Three items continue: maintenance and support, model costs, and everything you have changed later on. With the AI suite the annual licence covers the base scope. Higher maintenance tiers and maintenance for your own connectors come on top. Model costs follow usage and the provider’s pricing, and running locally trades them for hardware, electricity and depreciation. Changes after acceptance run as change requests at the hourly rate.
Can we start with a small project?
Yes, and that is the usual route. The AI potential analysis is one day on site with a short report afterwards, and you can stop there. Anyone going further takes the Voranalyse with its architecture and data protection concept; commission the overall system within 3 months and 50 % of what the analysis cost is credited against it. Within the suite the smallest sensible cut is the base plus the chatbot, with the email module the year after.
Let's talk about your use case.
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