Why the question is being asked afresh in 2026
The old rule of thumb was right: buying is cheaper than building, unless you have a very good reason. Custom software meant a team of developers for months, a specification that was already out of date at acceptance, and a dependence on the people who knew the code.
Two things have shifted. First, software is produced considerably faster with AI-assisted development: a tool for a clearly defined workflow is usable within weeks, and tests and documentation are produced along the way instead of at the end. Second, most standard software providers charge per head per month, permanently, and AI features often come as an add-on module. The costs grow with every employee; the benefit does not necessarily. So the outcome is no longer a foregone conclusion. It has to be done, workflow by workflow.
What standard software really costs
The licence is the visible part. Here are the public list prices and third-party estimates, as of October 2026, with a simple projection:
| Area | Example | Price according to list or estimate | Projection |
|---|---|---|---|
| HR | Factorial | from €8 per employee per month, modules extra | 300 employees: around €29,000 a year, around €144,000 over five years |
| HR | Personio | an estimated €5 to €15 per employee per month; the vendor only gives prices on request | 300 employees: €18,000 to €54,000 a year |
| HR, suite | Workday, SAP SuccessFactors | implementation 12 to 24 months, projects in the millions | usually oversized for companies with fewer than 1,000 employees |
| Service | Zendesk | $19 to $115 per agent per month | 20 agents: around $4,600 to $27,600 a year |
| Knowledge | Confluence | from around $8 per user per month | 300 users: around $28,800 a year |
The projections are base prices. They do not include add-on modules, implementation consulting, interfaces to your other systems or price adjustments in subsequent years. Check current prices with the vendor; they change.
On top of that, there are costs that never appear on an invoice: the Excel spreadsheet next to the system, because a step is not provided for. The time employees spend in screens they do not understand. The colleague who needs a licence of their own for two transactions a month, or shares one. These costs are real, but only measurable in your business.
What a custom tool costs
Building your own also involves more than the build itself. Calculated honestly, there are four: the build itself, migrating the data from the old system, operation (servers, updates, security, model costs for the AI parts) and further development when the workflow changes. We build at a fixed price that is binding after the analysis, and run the tool under an ongoing operation arrangement that can be cancelled monthly; the price levels are under Approach and prices.
The difference from the licence lies in the shape of the curve. The build is a one-off cost, and operation is a fixed monthly amount, regardless of whether 30 or 300 employees use the tool. The more people occasionally deal with a workflow, the more this counts.
Decision matrix: buy or build
Costs are one criterion, and often not the decisive one. The following matrix helps you classify a single workflow. Go through it row by row. If the majority falls into one column, you have a direction. If it is close, the mixed approach described further below is usually right.
| Question | In favour of standard software | In favour of a custom tool |
|---|---|---|
| Does your workflow match what the software provides for? | yes, with small deviations | no, you are already building workarounds today |
| Is the workflow an advantage over competitors? | no, it is mandatory and the same everywhere | yes, it is one reason customers buy from you |
| Do legal requirements change frequently, and does the vendor maintain them for all customers? | yes, for example in financial accounting or statutory reporting | no, you set the rules yourself |
| Who uses the system, and how often? | a few main users, daily | many occasional users who become expensive per head |
| Are there Excel or email shadow systems alongside it today? | no | yes, and they carry important steps |
| How many other systems need to be connected? | few, with ready-made interfaces | several, with custom connections |
| Where should the data be held? | with the vendor is fine | on your own servers or in a data centre of your choice |
| Should AI work in the workflow, for example reading or pre-sorting documents? | the vendor's module is enough | you want to choose the model and data location yourself |
| Is there someone responsible for rules and approvals? | not necessary, the standard prescribes them | yes, a person with decision-making authority |
| Is there any software to buy for the workflow at all? | yes, several vendors | no, there is only the spreadsheet |
The third row deserves a second look. Where a vendor implements legal changes for thousands of customers at once, buying is almost always cheaper, because you share this maintenance with everyone else. With a custom tool you bear every change alone. That is not an argument against building your own, but one for drawing the line in the right place.
The mixed approach is often the right answer
Most businesses do not face an all-or-nothing choice. Financial accounting stays with standard software, and so does the ERP. The custom tool takes over the workflow that currently runs alongside: onboarding across four departments, the approval chain for quotations, the portal in which suppliers upload certificates. It reads from the standard system and writes back to it. Our principle for this is: connect where the system fits, replace where it does not. How we build such tools is described under Tools that fit you.
The risks of building your own and how to limit them
The legitimate objection to building your own is: then we depend on whoever built it. This risk can be limited contractually and technically. The following seven points should be in every offer for a custom tool, no matter who it comes from:
- Rights of use: code, rules, prompts and documentation belong to you after payment, without restriction.
- Source code you can access: in a repository that belongs to you, not just with the service provider.
- Documentation to continue working with: structure, operation, data model and interfaces described so that your IT or another company can take over.
- Automated tests for the important workflows, so that a change does not break anything unnoticed.
- Operation of your choice: your server, a German data centre or your cloud.
- Interchangeable models: the AI parts do not depend on a single provider.
- Cancellability: ongoing operation can be cancelled monthly without the tool ceasing to run.
Anyone who gets these seven points is often more independent with a custom tool than with standard software whose price and range of functions are determined by the vendor alone.
Examples from practice
For a listed fuel cell manufacturer, while checking environmental claims under the EmpCo Directive, we built a tool in which the company can see what needs to be done on each page. It shows how many affected passages each page has and which have already been dealt with. Not a spreadsheet by email, but a status that updates as the work progresses. There was nothing to buy for this (Case study: regulated texts across ten domains).
In our own operations, custom tools replace project software and timesheets: work sessions are automatically recorded as tasks, named and prepared for invoicing. How this is set up and what went wrong at first is described in the case study on our own operations.
How to work through your case
- Choose one workflow, not an entire system.
- Add up the licence costs for five years, with the modules you really need and the expected number of employees.
- Estimate the time that currently goes into workarounds: spreadsheets, double entries, follow-up questions. A week of tally marks is enough.
- Go through the decision matrix and note where the majority points.
- Get a fixed-price quote for building and operating the custom tool, with the seven contract points.
- Compare over five years, not over the first year.
How the costs of an AI project break down in detail, from the analysis to the running model costs, is described in the article What an AI project costs a mid-sized company.
What this means for you
Standard software is not wrong; it is just no longer automatically the cheaper choice. Where your workflow matches the standard, buy. Where it is your advantage, or where a spreadsheet carries the business today, building your own often pays off in 2026, with code that belongs to you. Decide workflow by workflow, not system by system.
Your next step: use the cost calculator to work out the order of magnitude for your workflow before you renew the next licence.
Further reading
- What an AI project costs a mid-sized company: analysis, build, operation, model costs
- AI potential analysis: which workflows pay off with AI, and which do not
- Service: tools that fit you
- Case study: tasks, email, calls and checks in our own operations
- Industry: AI for professional services, with time recording and invoicing without standard software