Caching SAP metadata wisely: fast for the user, light on SAP
An SAP system has tens of thousands of tables. If you build tools that work with them, you do not copy everything; you fetch knowledge when it is needed and build it up step by step.
Short explanations of concepts from our field, ordered by topic.
Concepts around SAP, S/4HANA and integration.
An SAP system has tens of thousands of tables. If you build tools that work with them, you do not copy everything; you fetch knowledge when it is needed and build it up step by step.
Which SAP integration is best depends on the version and the type of SAP system. By first letting the system tell you what it can do, you choose the most modern standard it supports.
Years of in-house ABAP code make every SAP upgrade expensive and risky. By first mapping out what is still used, the core becomes clean again and the move becomes smaller.
An interface that may read any SAP table is quick to build, but security teams rightly block it. A dedicated object per data question is safer, auditable and survives upgrades.
SAP offers data through OData services. Version 2 is the familiar standard, version 4 the more modern successor, and RAP is the way SAP builds new services.
In SAP, functions can share a common container. One error in one function then makes every function in that container unusable. Strict separation prevents this.
Mainstream maintenance for SAP ECC ends at the end of 2027. Many businesses still have no route. Do not start with a migration plan, but with a factual picture of what your system really is and does.
If you connect other systems to SAP, you may face extra licence costs. This is called indirect access. SAP now often charges for it per document: digital access.
In SAP, a CDS view describes a data model with meaning: which tables belong together, how they relate and what the fields mean. Modern SAP apps and APIs are built on them.
A BAPI is a standard function maintained by SAP to read or change a business object, such as an order, customer or material. It is the safe way to control SAP from outside.
An RFC (Remote Function Call) is the way another system calls a function in SAP, as if it were a piece of SAP itself.
SAP S/4HANA is SAP's current ERP system, built on the SAP HANA database.
General AI models write tidy ABAP syntax, but they do not know the SAP-specific tables, fields and modern patterns well enough. That is where it goes wrong in practice.
Concepts around low-code and OutSystems.
Licences that grow with every application turn the moment of renewal into the moment of dependence. If you know in advance what a switch takes, you negotiate from strength.
Low-code platforms such as OutSystems and Mendix promised fast development without deep programming knowledge. With AI-assisted development, that speed is no longer a unique advantage. What is left, and when is a low-code platform still the right choice?
Teams that need SAP data do not have to become SAP specialists themselves. Describe which data you want in a neutral model, and only choose the technology at execution.
OutSystems can retrieve and update SAP data through SAP's OData services. With a few fixed agreements, that integration stays clear, secure and easy to maintain.
If you want to move away from a low-code platform, you find that the applications are tied to the platform. Rebuilding them step by step into clear software of your own gives you back control.
Mendix is a low-code platform of Dutch origin, part of Siemens since 2018. You build applications with models instead of code. What it delivers, and where the licence starts to pinch.
OutSystems is a low-code platform for building business applications with visual models instead of only with code.
Concepts around artificial intelligence and language models.
Apps built quickly around one AI provider such as OpenAI or Anthropic are cheap to make and expensive to untangle. Switching models breaks the software, and the provider knows it.
Many AI experiments get stuck in the pilot phase. That is rarely down to the technology, but to the choice of process. Start with one well-defined task with a clear return.
An AI agent can look up data, run calculations and make proposals on its own. By setting out precisely what it may and may not do, you get the convenience of AI without losing control.
Language models sometimes make up a source or an amount, and state it just as firmly as a correct answer. You cannot prevent that, but you can control it: with checks, comparison and a system that dares to admit doubt.
An AI assistant does not know the quirks of SAP. When the platform itself supplies the rules, pitfalls and working patterns, the AI does not have to guess.
The Dutch data protection authority received a report of such an incident at a GP practice. With AI that runs in your own environment, data does not leave the company.
A language model is strong at language and structure, but not at calculating, counting or looking things up. Leave that to ordinary code, and use AI to structure questions and explain results.
A large share of employees use AI tools whose use their employer does not know about or has not approved. Banning them does not help. A safe alternative that works better does.
A Two Swallows whitepaper for business and IT leaders. Who can still change an application when the original team or the original supplier leaves?
Converting speech to text can be done entirely on your own hardware, with open models such as Whisper and Parakeet. We measured how well and how fast that works.
With model escalation, a small AI model answers the question first, and a larger model only takes over when the answer scores too low in a check.
A conversation with an AI service in the cloud is not private by default. A case in Canada shows what a provider can see and share. What does that mean under the GDPR, in the United Kingdom and in Switzerland?
With RAG, a system first finds the relevant passages in your own documents and gives only those to the language model. When the system finds the right passages, the model can base its answer on them.
Generated ABAP code can look correct and still work differently from what was intended without any error message, because of limits in SAP that are not properly documented anywhere. If you know these pitfalls, you can catch them systematically.
With LoRA you teach an open AI model your own field without retraining it from scratch, and with quantisation you make it small enough to run on your own hardware.
A language model is an AI model trained on text that can write, summarise and translate text itself.
AI stands for artificial intelligence: software that performs tasks which normally require human thinking.
An MCP server gives an AI assistant live access to your systems and data, through an open standard: the Model Context Protocol.
Local AI runs on your own hardware instead of at a supplier in the cloud, so your data stays in-house.
Ollama is free open-source software that lets you run AI language models on your own computer or server, without the cloud and without costs per question.
The same question, the same model, the same settings and even the same seed, and still a different answer. In the trial we describe here, rounding in the compute kernels played a part. For testing, checks and accountability you need to know this.
How we work: architecture, advice and delivery.
In many SMEs, one person or one supplier knows how the IT fits together. If they drop out, work can grind to a halt. Documented knowledge and control in one pair of hands prevent that.
Not every piece of development is equally predictable. That is why we agree per backlog item how it is paid for: a fixed price, a fixed budget with a review point, or a timebox for research.
When a spreadsheet is used every month for the same planning or report, one wrong formula can carry through time after time. Repeatable, verifiable processing makes the work faster and the figures more reliable.
If you want to know what a decision means financially, you have to keep facts, plans and scenarios strictly apart. That way you can work through a scenario without changing the booked figures.
With a fixed price we agree in advance what will be delivered and what it costs. After that the price does not change, however many hours we put in. No timesheets, no discussion afterwards.
The classic DTAP pipeline starts with Development. With AI, software increasingly starts before IT, with the business itself. That is why we add a B: Business, Development, Test, Acceptance, Production. In Dutch we call it BOTAP.
When you move to a new SAP system, the integrations you keep using have to come along. By first taking stock and then moving in layers, you can check per object whether the move succeeded.
AI coding assistants write software fast. House rules give projects a fixed structure and help developers maintain the code later on as well.
We do not need to be the hosting provider, the cloud platform or the AI supplier. By leaving specialist tasks to specialists, we can advise what is best for you, not what earns us the most.
If you place integrations and custom code in SAP, you must always be able to say what is there, which version, and whether it is still correct. A central register with version history and checks makes that possible.
Before a website or app goes live, a short security review pays off. We find these mistakes on small websites almost every time, and they are usually quick to fix.
A separate SaaS subscription for every problem: after a few years you are paying for dozens of tools, some of which are hardly used. First look at what they deliver, and replace only what pays for itself.
Instead of annual plans, project portfolios and roadmaps, we work with one shared backlog. Together we decide which wishes come first.
ETL stands for extract, transform, load: taking data out of one system, converting it and loading it into another. With a fixed approach, an error-prone chore becomes a repeatable process.
Full stack means every layer of an application, from the screen to the database and the server. A full-stack developer or team works across those layers.
The Definition of Done records in advance when something is really finished. That way everyone knows in advance what the delivered work will be checked against.
CIOs and IT managers divide their time between budgets, suppliers, projects and people. A sparring partner with current technical knowledge makes their decisions stronger, and gives the board confidence.
With an hourly rate, the supplier earns more the more work is needed. In the age of AI, that grates more and more. We turn the incentive around: the more efficiently we work, the better for you and for us.
AI writes in hours what used to take days. As a result, the value in IT is shifting from producing code to making the right choices: what do we build, why, and what do we deliberately not build?
A copy of your accounts can be internally perfect and still be out of date. So check the consistency within the copy and also compare the figures directly with the source accounts.
The custom software we deliver for you and that you pay for is yours to keep, change and have maintained by another party. A partnership should last because it adds value, not because you cannot leave.