Legacy ERP without an interface: automate, don’t replace
An IT systems house lost 16 hours a month to copy and paste because its ERP has no interface. Why buying a new system is rarely the first answer.
In short
An ERP without an interface forces employees to copy and paste between systems. The reflex to replace the whole system is usually the most expensive answer. Automation around the legacy system is often cheaper. But it costs governance: whoever builds it must deliver traceability, access rights and ownership with it, otherwise a second, unsupervised process arises.
Auf Deutsch lesen: deutsche Fassung
An IT systems house worked out for me what a single missing data path costs per year. Two systems that do not talk to each other. An inventory management system from a time when interfaces were not yet a selling point. In between, a person who reads figures from one window and types them into the other. Around 16 hours a month. That is not an outlier, it is the norm in mid-sized companies, and almost nobody works it out.
I am not writing this from a consultant’s perspective. I run a precision engineering company myself, and I know the moment when you decide that this is simply how it works. The transfer does function, after all. It just costs time again every month, and because it never fails, it shows up in no list of faults. That is exactly why it stays in place for so long.
16 hours a month is not an IT question but a calculation
Work it through for yourself once, it needs no consultancy. 16 hours a month are two full working days. Extrapolated to twelve months that is around 192 hours, or roughly 24 eight-hour days. One month of work, every year, for retyping data that is already in the company. I am deliberately not naming an hourly rate, because you know yours better than I do. Simply multiply.
The bitter part is not the number. The bitter part is who does this work. It is rarely the cheapest head in the company. It is usually the one who has the expertise, because when transferring they recognise when something is not right. Precisely that person then sits there and types it out. In a labour market in which you can hardly replace this person, that is the most expensive way to burn time.
And there is an effect that no Excel sheet shows: anyone who transfers data for half the morning stops seeing things. After the thirtieth line item nobody looks any more. Copy and paste is not only slow. Copy and paste is also a source of errors that disguises itself as diligence.
Copy and paste is not work. Copy and paste is an open invoice that is issued anew every month.
The most expensive reflex is called: new system
When I open up this calculation, almost always the same suggestion comes back. Then we need a new ERP. And sometimes that is even true. Only it is almost never the first answer, but the last.
Replacing an ERP is not a software purchase. It is a project that ties up your master data, your processes, your document chain and your people for months. You pay not only for licences, you pay for migration, training, parallel operation and the mistakes you make in live business in the first weeks. Anyone who swaps an ERP because of 16 hours of transfer work is solving a two-day problem with a two-year project.
Before you talk about replacement, clarify three things:
- Is the process itself clean, or are your people typing out data that would not be needed in this form at all? Process first, then tool, then AI. In this order and in no other.
- Is it really a missing interface, or is the way out of the legacy system simply not known? Many systems can export more than the company knows, because nobody ever needed it.
- How long is the legacy system meant to keep running? For a system that will be replaced in eighteen months anyway, you no longer build automation. For one that will carry on for another five years, it pays off quickly.
Four ways around a system without an interface
If the legacy system stays, there are essentially four ways around it. I have sorted them by what they really cost in operation, not by how they look on a slide.
| Way | How it works | Where it tips over |
|---|---|---|
| Scheduled export and import | The legacy system regularly writes a file, the target system reads it in. | When the format changes with an update and nobody notices. The run reports success although it is transferring rubbish. |
| Read access to the database | A service reads directly from the legacy system’s database instead of going through the user interface. | When the vendor contractually prohibits the access or support takes it as a reason to refuse warranty. Clarify in writing beforehand. |
| User interface automation | A piece of software operates the screen like a person, clicks, types and reads off. | With every layout change. This solution is the fastest to build and the most exhausting to keep going. |
| Intermediate layer with its own storage | A small service collects the data, checks it and passes it on. | When nobody is named to maintain it. Then in two years there is a system in the company that only one person understands. |
Two pointers from practice that are in none of these rows. First: always build the transfer in one direction only, for as long as you can. As soon as two systems write to each other, you need a rule for which one wins in a conflict, and experience shows this rule only gets written down after the first data loss. Second: at the start, let the automation run alongside without writing. Two weeks of idle running show you the special cases that nobody thought of in conversation, and there always are some.
You see the pattern: with all four ways, the technical part is the easier one. What distinguishes them is the question of what happens when something changes, and who is then responsible. That is exactly where these solutions fail in mid-sized companies, not on the technology.
The calculation nobody opens up
Every automation around a system builds a second path along which your business data runs. That is entirely legitimate. Only, from the moment it runs, this second path is part of your processes, and so the same requirements apply as for the first. This is regularly forgotten because the solution begins as a small helper.
I ask these four questions in every audit in which such a solution turns up:
- Traceability: can you show for a single data record when it was transferred, from which state, and what was changed in the process? If the transfer touches anything relevant to the accounts, that is not optional. The German principles of proper bookkeeping require that a business transaction remains traceable in how it came about.
- Access rights: under which account does the service run? In practice I often see the personal account of the colleague who built it here, with full rights and no expiry date. That is a nonconformity as soon as you operate an information security management system, and a risk even if you do not.
- Error behaviour: what happens when the transfer goes wrong? Does the system notify someone, or does it carry on silently? A run that simply does nothing when an error occurs and reports success is more dangerous than no automation at all, because then nobody checks any more.
- Ownership: who maintains the thing when the colleague goes on parental leave or leaves the company? The bus factor is the most underrated metric of all in mid-sized companies. If the answer is a single name, you have not built a tool but a dependency.
The same applies when AI is involved. Whether document recognition reads out invoice data, a vision inspection sorts parts or a language model pre-structures orders: the question stays the same. Who looks at it, and how do they see that it is right? Automation without a final check is not progress. It is a process that nobody supervises any more.
A run that carries on silently when an error occurs is worse than no automation at all. Because then nobody looks any more either.
When changing the system is the right answer after all
So that no false impression arises: I am not against replacement in principle. I am against treating it as a reflex. There are situations in which every automation around the legacy system only throws good money after bad.
The clearest of these is missing support. If the vendor has discontinued the product, there are no more security updates or the system only runs on an operating system that is itself out of support, then you are building on sand. An automation then only extends the time in which nobody takes the uncomfortable decision.
The second situation is the data itself. If your master data is so tangled that every second evaluation needs manual work, no interface will help. An interface transports your data quality, it does not improve it. Anyone who transfers bad data faster has bad data in more places faster.
And the third: if you are building the third or fourth workaround for the same system, the decision has long been made, only nobody has said it out loud yet. From then on you are no longer maintaining an ERP but a patchwork that only two people in the company still have an overview of.
- No more support and no more security updates: replace, do not rebuild.
- Master data so tangled that every evaluation needs manual work: first the data, then everything else.
- Third workaround for the same system: the decision has been made, say it out loud.
- The vendor demands a project for every little thing: include the dependency in your calculation, not just the daily rate.
In all other cases the simple rule applies for me: the smallest solution that really eliminates the problem is the right one. Not the most complete.
How I would proceed before anyone builds anything
The order is more important than the tool. I never start with the software, but with the question of what is actually being transferred there and why.
In the case of the systems house, incidentally, the biggest lever was not the automation. It turned out that some of the transferred fields were no longer evaluated in the target system at all. They had been maintained for years because that was how it had always been done. That is not an exception, it is the rule. Anyone who automates first and asks afterwards automates their own ballast.
And one more thing that matters to me: you do not need a service provider for the first step. Your own specialist department measures which fields are really needed in two hours, better than any outsider. Get help for the build and for securing it, not for the listening.
This is my assessment as an auditor and managing director, not legal advice. If you want to know whether your home-grown solution holds up when it matters, that is exactly the kind of question an internal audit answers before someone else asks it.
Do you want not only to plan an automation around your legacy system but to have it built, with traceability and an access rights concept from the start?
Implementing AI projectsPrimary sources
- Section 238 of the German Commercial Code (HGB), duty to keep accounts
- GoBD (German principles for proper bookkeeping in electronic form), Federal Ministry of Finance circular as amended on 14 July 2025, section 3.1 Principle of traceability and verifiability (Official Fiscal Code Handbook 2026, Annex 33)
- ISO/IEC 27001:2022, Annex A, controls 5.18 Access rights and 8.2 Privileged access rights (iso.org)
Getting rid of copy and paste between two systems, in six steps
Measure instead of estimating
Have a record kept for two weeks of how much time the transfer actually costs and how often it arises. Without this number you are talking about feelings, and feelings lose against every other budget topic.
Sort out fields
Go through every transferred field with the specialist department and ask who actually evaluates it in the target system. Experience shows that part of the work disappears here without replacement, without a single line of software being written.
Clarify the way out
Check what the legacy system provides of its own accord: scheduled exports, reports, file outputs. And clarify in writing with the vendor which access path is permitted and what it means for warranty and support.
Choose the smallest way
Take the simplest solution that holds up, not the most elegant one. A scheduled export with import is boring and lasts longer than a user interface automation that breaks with every layout change.
Build in control
Define which service account the run works under, what it logs and whom it notifies when something goes wrong. A run that silently passes over errors is not a tool but a risk.
Name the owner
Write down who maintains the solution, who steps in during holidays and where the description is kept. One page is enough. Without this step you have not built an automation but a dependency on a single person.
Frequently asked questions
Is automation worthwhile if the legacy system is going to be replaced anyway?+
That depends on the remaining lifetime. If the system runs for another eighteen months or less, I would build nothing more and put the time into a clean migration instead. If it carries on for three to five years, automation usually pays for itself in the first year, because the transfer work arises permanently.
Can you simply access the database of a vendor system directly?+
That is a contractual question, not a technical one. Many vendors prohibit direct access or tie warranty and support to it. Clarify this in writing before anyone builds. An access that works technically but is contractually prohibited costs you the vendor’s support in an incident.
What is the most common mistake with home-built automations?+
That they run under the personal account of the person who built them and nobody but that person understands them. Neither is noticed for years, and both are noticed precisely when the person is not there. A named owner, a dedicated service account with appropriate rights and a short description solve ninety per cent of this.
When does such a solution become a topic in an audit?+
As soon as it moves data that is relevant to the accounts or personal data. Then traceability, access control and ownership apply just as they do for the main system. I do not check whether the solution is nicely built, but whether you can show for a single transaction what happened when and with what.
Author & expert review: Lars Zimmermann · ISO/IEC 42001 Senior Lead Auditor & Senior Lead Implementer · ISO/IEC 27001 Lead Auditor & Lead Implementer (PECB)
Last updated: 25 July 2026. Researched and reviewed to the best of our knowledge; not a substitute for individual legal advice.
Questions about your own case?
In a free 15-minute intro call we assess where you stand on ISO 42001, ISO 27001 and the EU AI Act, honestly and without a sales pitch.