- Invoice receipt: Receive PDF invoices or Errors continue to be checked by humans instead of being automatically included in the booking. The typical implementation time for this is (in manageable setups). two to four weeks.
- Offer management: A customer inquiry is received by email, AI extracts the scope of services, checks against price lists, generates a draft offer and forwards it to a clerk for approval. The customer receives his offer more quickly and processing takes less time. There is no ERP module involved – and there is no seven-figure project sum involved.
- Client mail from tax advisors: Incoming documents are classified according to type and urgency and responsibilities are assigned. What causes daily bottlenecks in law firms with high volumes of mail can be mapped out as a rule-based workflow – without any major DATEV implementation project.
Why low-code and AI fail – and how you can prevent it
The biggest operational risk in low-code projects is shadow IT or AI. The digital association Bitkom found out in a survey from October 2025 that 17 percent of survey participants assume that their employees use AI tools for business purposes via private accounts. With more 17 percent There were isolated cases in this regard – and at eight percent this behavior is widespread. If customer data is involved in such a case, it is a GDPR violation without an order processing contract as a basis. In this respect, a low-code strategy needs clear usage guidelines – otherwise you would be exchanging an ERP risk for a compliance risk, which would not be progress.
In addition, low-code platforms can reach structural limits when it comes to highly specialized logic structures (multi-level production planning, group consolidation, ISO-certified audit trails). And: Vendor lock-in is real – proprietary platforms are difficult to migrate. Open source and self-hosting solutions have a structural advantage here. However, this is no guarantee that there will be no dependencies in the future.
Against this background, I recommend that medium-sized companies follow the following three steps in order to sustainably realize their low-code AI ambitions.
- Select pilot process according to strict criteria: The first candidate should meet four criteria at the same time: high volume, clear technical rules, no write access to the ERP in the first expansion stage, and a person responsible who already knows the process in detail. The areas of invoice receipt and offer preparation almost always meet these requirements. A process that is both production-critical and poorly documented is almost never the case. Anyone who chooses the wrong pilot at this point must expect that acceptance will be significantly lower on a second attempt.
- Clarify data protection before the first workflow: A data processing agreement with the model provider is something that should be done before the go-live, not after the fact. This review must also be done with the low-code platform itself in mind, not just the language model behind it. Anyone who works properly here eliminates the shadow IT problem described at the beginning: If the official solution is ready for use more quickly than the unofficial one, employees are also less likely to use private AI accounts.
- Define scaling limit in advance: Determine the level of complexity at which a process moves back onto the ERP roadmap before you need it. Introducing this limit later costs more time than the original ERP decision saved – in the midst of an evolved workflow with twelve branches. A simple signal works well in practice: As soon as a workflow requires more than three nested conditions to represent a single decision, the logic should be checked – not extended.
The actual utility for SMEs
Six months less implementation time doesn’t just mean being productive six months earlier. It means:
- To be able to demonstrate results before the ERP budget is released.
- To be able to test processes before they are scaled.
- Being able to fail because it’s a workflow for 4,000 euros and not a project for 200,000 euros.
Your ERP project can remain on the roadmap. But for the next twelve months there is usually a quicker way. (fm)
This article was published as part of Foundry’s German-speaking expert network. Would you like to join in? Apply now!
