AI meets existing system: why established IT landscapes are not an obstacle to AI, but often an advantage

According to a Slalom study from 2025, 61% of German companies are running the majority of their main business applications on outdated platforms. According to the study “Legacy Modernization 2024” (CIO/Computerwoche/Thinkwise), 72% are facing the challenge of updating outdated but business-critical legacy systems. This is not a marginal problem, but the normal state of the German corporate landscape.

What these figures do not show: Many of these companies are waiting. They are waiting for their system landscape to be “clean enough” to introduce AI. This logic reverses the causality. Because the real hurdles to introducing AI are rarely the age of a system. They are missing integration points, unclear data access and the question of governance: who is allowed to do what with which data?

These are solvable problems. And often, not a single core system needs to be touched.

Here is an idea that surprises many: large language models, the AI technology behind applications such as internal assistants, automated document processes or intelligent search, are designed from the ground up to deal with precisely the data that typically accumulates in grown corporate landscapes.

LLMs are designed to understand human speech input and produce human-like text output. They can analyze and process large amounts of unstructured data such as texts, emails or documents – recognizing complex relationships and identifying patterns that are often difficult for humans to recognize.

Legacy systems produce exactly this type of data: Orders that arrive by email. Maintenance logs that are stored somewhere as a PDF scan on a file server. ERP histories that contain decades of transaction data but have never been systematically analyzed. Contracts, inquiries and documentation are all text, all human language, all processable.

The key point here is that AI is not “plugged onto” a system. It is about intelligently expanding existing processes without shaking the foundations. AI modules are integrated into the process chain, they read inputs, enrich them, forward them or return an answer. The core system remains unchanged. It delivers data and receives results as it always has, only now also from non-human agents.

The question that must be answered first in every project is not: “How do we replace this system?” Rather: “Where in this process can an AI module intervene in a meaningful way without having to change the core system?”

The answer almost always lies in integration layers. The core system, whether ERP, CRM, proprietary industry solution or database from the 1990s, remains what it is: the reliable repository for transactions and master data. An AI module is operated alongside it as an independent service. It obtains data from the existing system, processes it and delivers structured results back or directly to the user.

Das vermeidet die größten Risiken klassischer Modernisierungsprojekte: Kernprozesse werden nicht unterbrochen, kein jahrelanges Migrationsprojekt mit unsicherem Ausgang, kein Wissensverlust durch Systemwechsel. 91 % der Unternehmen priorisieren laut aktueller Studien die Modernisierung ihrer Kernsysteme gegenüber einer riskanten Neuentwicklung. Das ist eine „Modernize over Replace“-Strategie, die vor allem durch den Wunsch nach Stabilität getrieben wird. KI-Integration über externe Module geht genau in diese Richtung.

In concrete terms: a mechanical engineering company continues to use its existing ERP for warehousing and orders and receives an AI component that automatically categorizes and pre-qualifies incoming supplier inquiries. An insurance company does not change its core system, but lets an LLM read claims notifications in free text, extract relevant information and feed it into the existing processing in a structured manner. A service provider opens up decades of contract data for targeted evaluations without ever migrating the data.

A particularly effective approach for companies with complex inventory landscapes is Retrieval Augmented Generation, or RAG for short. The principle is simple: a language model is not retrained with internal data, as this would be expensive and time-consuming. Instead, it receives the relevant information for each query from an indexed knowledge base that is fed from existing sources.
Typical sources are internal documents, wikis, support databases, product documentation, CRM systems or repositories such as SharePoint and Google Drive. In addition, there is ERP export data, scanned legacy documents, email archives, handwritten maintenance reports – everything that could not previously be systematically queried.
RAG gives companies the opportunity to feed language models with their own data without retraining or fine-tuning and thus provide specially adapted AI functions at a fraction of the time and cost. This is particularly relevant for regulated industries: Access rights can be controlled at document level, GDPR-compliant operation can be planned, and data does not leave your own infrastructure if that is what you want.
RAG is not a substitute for modernization. It is a way of accessing knowledge that is already there and has been lying fallow until now.

There is a third perspective that is often overlooked: AI can actively accelerate the modernization of your own system landscape.

AI-supported tools such as automated code analyses make it possible to carry out complex migration and modernization processes with reduced manpower and minimize risks. Generative AI can analyze existing code, generate documentation and convert legacy code into modern languages while preserving the business logic. According to recent field reports, the time required can be reduced by around a third compared to manual refactoring.

This is particularly important because knowledge of legacy systems is dwindling. 61% of IT decision-makers see demographic change as a risk factor, as know-how about older technologies such as COBOL or mainframe structures is increasingly being lost. AI can serve as a bridge here, not to replace this knowledge, but to secure it and make it accessible before it is lost.

The result: no either/or between “continue to operate the system” and “build everything from scratch”. Instead, a third option: step-by-step, AI-supported further development in which every step can be used immediately.

The initial question is not a technology question, but a process question: Where in your system landscape does data accumulate today that no one systematically analyzes? Where are there steps in the process that are completed manually because no system links them automatically? Where is there knowledge in documents that no one can query?

These answers result in concrete pilot projects with a narrow scope, clear metrics and real ROI. No big bang. No waiting for the perfect infrastructure.

At BAYOOTEC, we bring together precisely these two competencies: many years of experience in working with mature system landscapes and the technical know-how for AI integration in complex, regulated environments. We build the connecting layer between what already exists and what is possible with AI today.

Would you like to know where AI is already possible in your system landscape without risking your current processes? Please feel free to contact us. We look at your specific situation together.

AI is here to stay. It is becoming more powerful, more accessible and more deeply integrated into engineering processes. This is not a threat, but it is not a sure-fire success either. The decisive factor is how companies and teams deal with it.

At BAYOOTEC, we have deliberately decided against accepting AI as a black box or using it as a shortcut. Instead, we rely on clear policies, human approval processes, secure prompting, structured logging and regular training. That takes time. It also costs some speed in some places. But it is the basis on which we can deliver code that we stand behind, for customers who rightly expect it.

Analysis is not responsibility. Responsibility lies with us.