Case study: Lagerhaus Mieming Request received: 29 June 2026 · First production use: 2 July 2026 · Completed: 11 August 2026
Published with permission from Lagerhaus Mieming.
Background
On 1 July 2026, Austria introduced a new reduced VAT rate of 4.9% for selected basic foods (§ 10 Abs. 1a UStG 1994). This unusual figure stems from EU law: the VAT Directive generally allows only two reduced rates, which Austria already uses for its 10% and 13% rates. An additional rate is only permitted if it is below 5%.
That fractional rate was the central technical challenge of the project.
Lagerhaus Mieming runs an inventory management system built on DataFlex and developed over decades. It supported tax groups of 0%, 10%, 13% and 20%; a fifth group had never been envisaged. The item master needed to display the marker 5, while calculations had to use 4.9%. This distinction between the displayed value and the calculation rate ran through the entire system.
The request reached us on 29 June, two days before the change took effect.
The scope was larger than expected
The change affected the whole processing chain:
- Checkout and receipt handling
- Document lines and totals
- Invoice, consolidated invoice and delivery note variants
- Several forms of document printing
- Purchasing and ordering
- Posting batches and transfers to accounting
- Monthly reports and advance VAT returns
The application was undocumented. Its actual logic had to be reconstructed from screen fields, include chains, temporary records, printing routines and accounting entries — knowledge that had existed only in people’s heads and in the code for decades.
Approach: parallel versions
Each affected source file was maintained as a clearly marked parallel version. The original and updated versions existed side by side, making it possible to switch back to the proven version at any time.
The rollout proceeded in stages:
- From 2 July, the first checkout and document programs entered production.
- Next came invoicing, consolidated invoicing, ordering and the remaining areas, with each step tested and approved individually.
- In early August, after about a month of trouble-free parallel operation, the proven variants took over the production names. The original versions were backed up and archived first, keeping the release traceable and reversible.
- By mid-August, reporting had been extended to cover the new rate, including advance VAT returns and intra-Community acquisitions.
How AI was used in this project
The project was implemented with the support of an AI coding assistant.
Its scope covered the entire processing chain: document lines, totals, printing, invoice and delivery note variants, purchasing, posting batches, monthly reports and, finally, the advance VAT return.
The division of responsibilities was clear:
Developer
- Specified the special rule
5 -> 4.9 %. - Explained existing fields and old program logic.
- Made structural decisions, such as extending
DOCF. - Tested the programs using real workflows and documents.
- Reported observed errors immediately.
- Approved production steps and the subsequent release.
AI assistant
- Located relevant source code and followed data flows across includes.
- Compared working old programs with new variants.
- Developed small, localised changes.
- Preserved the bytes of legacy file encodings when editing.
- Compiled affected programs and checked error lists.
- Analysed real CSV, DAT, checkout and accounting data.
- Stopped when datasets were inconsistent rather than writing uncertain changes.
- Created audit, planning, test and repair tools with explicit preconditions.
- Documented backups and removed temporary tools from the production path.
Two aspects stood out:
The old VDF7 development environment was restored to working order. A compiler call initially failed because the historical development environment was incompletely configured. The AI reconstructed the necessary settings and paths and restored the environment after approval — a task that often takes longer than the actual change in systems of this age.
Domain knowledge and AI complemented each other. Several decisive clues came from the user: one data field had a special historical meaning relating to a long-abolished land value levy (Bodenwertabgabe), and the receipt routine actually in use was different from the one its name suggested. The AI corrected its assumptions accordingly. Success came through dialogue and repeated refinement.
Quality assurance
- Source comparisons between the old and new program variants
- Compilation after each small change
- Practical user tests with real receipt and delivery note workflows
- Recalculation to the cent from quantity, price, discount, price unit and tax rate
- Reconciliation of gross invoice totals against posting batches
- Backups of all original sources before release
- Follow-up checks in the next month
Result
The change was in production from the day after the new rate took effect and was fully completed by mid-August. Operations continued throughout.
What this project shows
Long-established business applications remain the foundation of many companies. They can be adapted under time pressure and without documentation, provided the source code is available.
AI accelerates code analysis, exposes hidden dependencies, helps create and check variants, and supports controlled changes within clear safety boundaries. People provide the domain knowledge, take responsibility, and make the decisions, tests and approvals.
The work continued beyond the first successful compilation to practical testing, error diagnosis, release, archiving and plausibility checks in the following month.