Last updated: June 22, 2026
Your company's data is exploding – but your ERP database knows nothing about it. Every telematics sensor, every IoT device, every AI model and every robot produces more data points per day than a classic ERP system was ever designed to store. Anyone who wants to use AI meaningfully in the coming years cannot avoid one central question: where to put all this data? The answer is a data warehouse. In this article I explain why building a data warehouse is becoming a must today, how we implement it at PixelMechanics with open-source technologies without license costs – and what a recycling company achieved with it in concrete terms.
The key facts at a glance
- Classic ERP databases are not designed for the data volumes from telematics, IoT, AI and robotics.
- A data warehouse plus middleware bundles all data centrally and makes it AI-ready.
- PixelMechanics relies on open source without license costs: Airbyte, Apache Kafka and PostgreSQL.
- Practical example: at a recycling company we connected HubSpot, Basion ERP, YellowFox and IoT devices with the data warehouse and AI.
- Result: AI solutions can be introduced in days instead of months.
Why the ERP database is not designed for the future
An ERP system is built to map business processes: orders, invoices, inventory, bookings. Its database is optimized for clean, transactional data – not for the data volumes that telematics, IoT, AI and robotics generate today.
The difference is enormous. An order may generate a few dozen records. A connected vehicle fleet with telematics sends GPS position, fuel consumption, load and machine states every second. IoT sensors in production deliver continuous measurement series. AI models need historical data over years in order to learn. Robotics generates movement and status data in real time.
If you try to force all of that into the ERP database, one of two things happens: the system becomes slow and unstable, or the valuable data ends up scattered across isolated silos that no one can access centrally. Both are a killer for any future strategy. That is precisely why a dedicated data layer is needed – a data warehouse, complemented by middleware that connects all systems.
What is a data warehouse – and why middleware?
A data warehouse is a central database that brings together data from many different sources, stores it permanently and makes it usable for analysis and AI. Unlike an ERP system, it is designed to process huge amounts of historical and ongoing data efficiently.
The middleware is the connecting layer in between: it ensures that data from ERP, CRM, telematics and IoT devices flows automatically and in real time into the data warehouse – without every system having to be individually coupled with every other. This creates a single, reliable data source instead of a patchwork of individual interfaces. Anyone who takes this step lays the foundation for a real digital operating system in the company.
Our open-source stack: Airbyte, Kafka and PostgreSQL – without license costs
At PixelMechanics we implement the data warehouse and middleware directly at the customer's site – consistently based on open-source technologies. That means: no ongoing license costs, no dependence on a single vendor and full data sovereignty for the company.
Our proven stack consists of three components:
- Airbyte handles data integration. Via hundreds of ready-made connectors, it automatically pulls data from ERP, CRM, telematics and other sources into the data warehouse.
- Apache Kafka is the real-time nervous system. It streams events and sensor data reliably and at high volume – ideal for IoT and telematics.
- PostgreSQL serves as a robust, highly scalable data warehouse in which all data converges and is available for analysis and AI.
The appeal of this architecture: it is proven, scales with the company and costs no license fees. The budget flows into value creation instead of software licenses. You can find more about how we connect ERP, AI and IoT on our page on intelligent ERP solutions with AI and IoT.
Practical example: data warehouse for a recycling company
What this looks like in practice is shown by our most recent project at a recycling company. The starting situation was typical: many systems, lots of data – but no central access. This is exactly where we built a data warehouse with middleware and connected the key systems with one another:
- HubSpot as CRM for customer and sales data
- A legacy ERP system for the operational business processes
- YellowFox for the telematics of the vehicle fleet
- additional IoT devices from ongoing operations
Today all these sources flow via Airbyte and Kafka into a central PostgreSQL data warehouse – and are connected to AI there. Isolated data pools became a continuous data foundation. For the waste and recycling industry we have compiled this approach in our industry solution for the waste management sector .
The result: introduce AI solutions quickly instead of preparing for years
The decisive effect showed up immediately afterwards: because the data is now centrally located in the data warehouse, the company was able to quickly introduce AI-based solutions and access the data directly. What would previously have meant months of preparatory work and interface tinkering is today a matter of days.
That is the real point: a data warehouse is not an IT end in itself, but the launch pad for AI. Anyone who has built the data foundation cleanly can simply dock on new AI use cases – from predictive maintenance and route optimization to automated analyses. This is exactly what we support with our tailored AI implementation.
What kind of consulting helps with building a data warehouse?
Building a data warehouse requires consulting that understands ERP, CRM, telematics and IoT equally and relies on vendor-independent open-source architecture. This is exactly the core competency of PixelMechanics: we connect your systems into a central data foundation and make them AI-ready – without forcing you into expensive licensing models.
If you want to make your company future-proof for AI, IoT and robotics, building a data warehouse is the logical first step. Talk to us about your AI and data warehouse strategy – or take a look at our reference projects.
Frequently Asked Questions About the Data Warehouse (FAQ)
Why is an ERP database not enough for AI and IoT?
An ERP database is optimized for transactional business data, not for the enormous data volumes from telematics, IoT, AI and robotics. It becomes slow and unstable with these data streams. A data warehouse therefore takes over the storage and analysis of large data volumes.
Which technologies does PixelMechanics use for data warehouse and middleware?
We rely on open source without license costs: Airbyte for data integration, Apache Kafka for real-time streaming and PostgreSQL as the data warehouse. This keeps full data sovereignty with the company.
How quickly can AI be introduced afterwards?
As soon as the data is centrally located in the data warehouse, AI applications can be docked on in days instead of months, because the data foundation is already in place and cleanly connected.
About the author: Michael Rohrmüller is the founder and managing director of PixelMechanics. He supports companies with digitalization – from ERP and CRM integration through data warehouse architectures to the introduction of AI. His focus is on vendor-independent open-source solutions that make companies independent and future-proof.
Further resources
- AI Consulting & Strategic AI Planning – How we develop your AI roadmap
- AI Implementation for Companies – Tailored AI solutions for SMEs
- Digital Operating System – Integrated ERP, CRM & AI from a single source
Frequently Asked Questions
What is a modern data warehouse and why do AI and IoT demand it?
A modern data warehouse centralizes structured and unstructured data from ERP, IoT sensors and production systems. AI and IoT applications need clean, consistent and historically complete data – the warehouse forms the unified layer that makes machine learning models, real-time analytics and predictive maintenance possible in the first place.
How is IoT data integrated into a data warehouse?
IoT devices generate continuous time-series data from sensors and machines. Integration requires data pipelines – via tools such as Apache Kafka, Azure Event Hub or AWS IoT Core – that capture, filter, transform and load data into structured tables.
How does PixelMechanics approach data warehouse projects?
PixelMechanics follows a five-step process: (1) analyze the data landscape, (2) design a cloud-native or hybrid architecture, (3) develop ETL pipelines, (4) set up a data governance framework, (5) configure the analytics and AI model integration layer.
Do we need a central data platform at all?
Only when data from several systems has to be combined for decisions, or when volumes exceed what the ERP was built for. If two reports out of a spreadsheet answer your questions, you do not need one yet.
What is the difference from our ERP?
The ERP runs the business, the platform analyses it. Reporting queries against a live ERP slow down the people trying to work in it, which is usually the first practical reason to separate the two.
What does it cost to run?
Storage is cheap; modelling and maintenance are not. Budget for the people who keep the definitions current — a platform whose figures nobody trusts costs the same and delivers nothing.
Where does AI fit in?
After the platform, not before it. Models need consistent, well-labelled data. Without it, every AI project spends its budget on data cleaning and still calls itself an AI project.
Cloud or on premise?
Mostly a compliance and cost question rather than a technical one. Check where your data may be stored and which skills you already have before you choose a product.
How do we start small?
With one question management asks regularly and that currently takes days to answer. Build the pipeline for that single question end to end, then extend it.
What is the most common mistake?
Loading everything before knowing what it is for. Data collections without questions become expensive archives that nobody ever queries.
