Florian Seidl
Florian Seidl
Industrial Informatics · · 7 min read

Industrial data fabric: what it is, what it is not – and what it really delivers

“Industrial data fabric” is one of those terms that means everything and nothing. Before talking about what such a layer can do, the more uncomfortable question is worth asking: what is it, and what is it not?

Pharma MES Europe 2026 series, part 4 of 6: The posts were written for the joint appearance of cts Group and Xenium AG at Pharma MES Europe 2026 in Berlin. Previously: part 3, Why MES projects fail.

Every vendor means something different when they say “Data Fabric.” A good deal of the confusion in data projects comes from that — and the occasional bad purchase. So let us begin with the definition.

In short, an Industrial Data Fabric (IDF) is a foundation and connectivity layer between OT (Operational Technology, the technology of production) and IT (Information Technology). It turns fragmented production data into one coherent, contextualised layer. Not a new system on top — a layer underneath.

If that sounds abstract, the picture gets sharpest through contrast. So first, what an IDF explicitly is not.

What it is not

Distinction 1

Not a process control system

It is not a classic OT system, no SCADA (Supervisory Control and Data Acquisition), no PLC and no DCS (Distributed Control System). It does not control or regulate the process.

Distinction 2

Not an MES

It is not a Manufacturing Execution System, does not execute production orders and does not replace the execution layer.

Distinction 3

Not an HMI layer

It is not a classic operator interface (HMI, Human-Machine Interface) as in SCADA. Visualisation is possible — but it is not the core.

Distinction 4

Not just a historian

A historian stores values; an IDF connects and contextualises them and makes them available across systems.

And it replaces none of these systems but sits underneath them. Anyone selling you a “Data Fabric” that controls the process or replaces the MES is selling something else.

What it is

At its core it is a pure connectivity layer: messaging, data reception, interconnectivity. A foundation between IT and OT — not a replacement for either. From that role follow a few properties that make a dependable IDF:

Property

Vendor-neutral

No built-in cloud dependency, no lock-in to a single vendor. It stays open to the existing installed base.

Property

Centrally managed, run locally

Autonomous instances per site, technically self-sufficient — connected into one central level. One system enterprise-wide, instead of many islands.

Property

Extensible

Where a native standard interface is missing, a custom driver reads it. Scripting (for example in Lua) on every component normalises data at the edge.

Property

Built for enterprise scale

Connectivity across many sources and sites, without extra tools — with a web toolset for visualisation and apps on top.

Tools such as AspenTech Inmation put this principle into practice. But the IDF is an architectural pattern, not a product — which tool implements it is decided by the installed base, not the brand. cts works deliberately system-independent here.

What it can do

The practical difference lies in the connection model. Instead of building each connection individually — MES to process control, line to cloud, site to site — a layer emerges over which these connections run in a standardised way. An IDF can act as a translation and standardisation layer: it translates a standard message into whatever the respective target system understands.

The effect shows less in the first step than in every one after it:

1
The first connection takes effort. The layer has to be in place, the setup takes time — there is no way around that.
2
Each further one gets faster. The next data source, the next PLC, the next cloud connection draws on what already exists — templating and reuse.
3
The effect compounds across sites. Once the layer exists, OT connectivity can be established across sites — without duplicating a historian and MES per site.

What it delivers

The real benefit is not a single number, but a shift: away from one-off, manual integration work — towards a scalable, repeatable model that speeds up every further use case.

Where this effect could be measured, it is clear. In a pharma project, we compared an initiative from 2025 with a comparable reference project from 2020, that is, before and after the shared connectivity layer was built:

Around 2.5 times faster delivery of new use cases after the shared layer was built. This is an experience value from one project, not an industry average; how large the effect turns out elsewhere depends on the starting point and system landscape. The reason is the same in every case: the effort per further connection drops, because templates, drivers and the data model are reused.

Anyone who wants to prove the effect in their own project is best off measuring a few simple figures from the start: the effort to connect a data source, the time from request to first use, and the number of templates reused for a new line or a new site. If these values are recorded from the first use case on, it can later be shown what the shared layer actually changed.

And to keep it honest: the first implementation takes time and setup. The gain lies in the repetition, not in the first step. For anyone who only wants to connect a single data point once, the effort is not worth it.

After 18 years on the OT side — with PLCs, process control systems and data integration — and around ten years with connectivity software, I have reached a point that is almost uncomfortable for a technical person: the technology is no longer the bottleneck. It has existed for a long time.

What is missing is rarely the capability — it is someone who has already walked the path. Many hurdles that were hard years ago are easy today, because they have already been solved in other projects. You do not have to start from scratch. And honestly, I find that the more interesting message than any feature list.

Three test questions for an industrial data fabric proposal

Anyone evaluating a proposal for an industrial data fabric gets to the point quickly with three questions. They follow from the distinctions above:

01

Does the solution write back to the controller or control system?

If so, it is part of process control and not a foundation layer. In regulated plants, that has direct consequences for validation.

02

Does it replace the MES or historian, or connect them?

A foundation layer leaves existing systems in place and makes their data available. If something has to be replaced, you are planning a different project.

03

How is it run per site and managed centrally?

Autonomous instances with a central level scale differently from one installation per site with no connection between them.

In numbers, the difference between individual interfaces and a shared layer shows up with only a few systems; this is covered in the article How Production Data from Different Systems Is Consolidated.

What an industrial data fabric does not solve

An industrial data fabric connects and contextualizes data, but it does not decide what the data is used for. Unclear goals, missing data ownership and poor data quality at the source remain unless someone addresses them, and the layer itself needs operation, updates and clear responsibility. This calls for a self-critical look: as an integrator focused on connectivity, it is tempting for us to answer data problems with a platform. For a single connection between two systems, a direct interface is often the better choice.

When it is worth it — and when it is not

An IDF is not the right path for everyone. For a single, self-contained integration or a manageable system landscape it is oversized. It shows its value where data has to come together across many systems and several sites — and where every further use case is meant to build on the same foundation.

From experience: the fastest way to test a “Data Fabric” in a proposal is to ask what it does not do. Does it control the process? Does it replace the MES? Then we are not talking about a foundation layer, but about another system on top — and the silo question all over again.

The term will keep being used loosely — no blog post will change that. But now you have the test for it. Where the line runs for you between “foundation” and “yet another system” is best clarified on the concrete case. That is exactly what we are happy to clarify in a conversation.

Next in the series, part 5: The first use case that counts

Foundation or yet another system?

cts Group and Xenium AG clarify with you, on your concrete case, when an industrial data fabric holds up and when a simpler path is the better one.

Arrange a conversation