The historian: the wealth of data many factories leave unused

The short answer. Many factories already have the data to explain their recurring downtime: in the historian or in the archive of their SCADA system. Yet 43% of manufacturers say they are not using their collected data effectively, according to research by Rockwell Automation among 1,560 decision makers in 17 countries. The problem is rarely a lack of data. It is that nobody has the time and the context to get answers out of it. So do not start with a big data platform, but with your most expensive recurring breakdown, and first test whether the data holds the answer.

What is a historian?

A historian is a database that stores process data from the plant with a timestamp: temperatures, pressures, speeds, states and alarms, often going back years. AVEVA PI System is a well-known example. SCADA systems often keep such an archive too. Siemens WinCC, for example, stores measured values and alarm messages, which external software can read through the OLE DB Provider in the WinCC Connectivity Pack.

The bottleneck is not storage, but access

A historian records what happened, but not why. Which product was running, which mould was in the machine, which order, which breakdown followed and what the technician did: that context sits in the MES, in the maintenance system and sometimes still in a shift log. As long as those sources are separate, every root cause investigation starts with manual work.

Rockwell Automation’s report Scaling MES Across the Enterprise (July 2026) shows how big that gap is:

• 93% of manufacturers have an MES, but only 23% have fully integrated it with ERP, PLM, quality and OT.
• 33% call the MES their biggest data integration problem.
• 43% are not using their collected data effectively, while Rockwell itself calls that data the foundation AI needs to perform.

Keep in mind that this is self-reported, in research by a vendor of MES software. As a direction for the sector, it is recognisable. How to connect those sources without replacing everything is covered on our page on connecting systems.

What downtime costs

According to Siemens’ The True Cost of Downtime 2024, the world’s 500 largest companies together lose almost 1.4 trillion dollars a year to unplanned downtime, 11% of their revenue. According to the same report, an average large plant loses 27 hours a month to unplanned downtime. This too is vendor research, and it covers the very largest companies. For a mid-sized plant, the damage lies less in the absolute amounts and more in three patterns:

Repeat breakdowns. If only the big incidents are investigated, the smaller ones keep coming back. They drag down OEE without ever making the agenda.

Dependence on key people. Knowing which tags to look at for which breakdown is experience. If that experience sits with one process engineer, your breakdown analysis is only as strong as that one person’s diary and retirement date. Capturing that knowledge is exactly what knowledge management with AI is about.

Treating symptoms on the shop floor. Operators who adjust settings to keep a line running solve the immediate problem, but often mask the trend that could have pointed to the cause.

Do not start with a data platform, start with your most expensive breakdown

The reflex is often to collect all data centrally first. That takes a long time to pay off, and it requires investment before it is clear whether the data holds the answer. A smaller approach gives certainty sooner:

1. Pick one problem with a business case. The recurring breakdown that costs the most downtime, scrap or rush work. That also makes it clear what you will judge the result on.
2. First test whether the data holds the answer. With an export of historian or SCADA data and the breakdown log for the past few months, without integrations. After a few weeks you know whether there is a pattern, and you can decide to continue or stop.
3. Connect and automate what works. Only once the pattern is proven are the historian, MES and maintenance system connected, so an analysis is ready automatically for every new breakdown.
4. Scale up to more lines. With the experience of the first line and the confirmed causes as the basis for the next.

The difference is in the order: not investing first and hoping for insight, but proving insight first and then investing. That first test is in effect a proof of concept on your own data. What is needed afterwards to make the data structurally usable is covered on our page on the data foundation.

What to watch for with AI on process data

AI can take over the searching: show for each breakdown which tag deviated first, find earlier breakdowns with the same pattern and pull in the maintenance history. Four conditions decide whether that is responsible in your plant:

Read-only. The solution reads data from the historian, SCADA archive or MES and has no access to the controls. That keeps OT security intact and makes the conversation with your automation engineers easier.

Traceable. Every suggestion refers to the underlying measured values, so your engineer can check what it is based on.

The engineer decides. AI proposes causes, a person confirms them. This is called human-in-the-loop. Only after that confirmation does the knowledge go into the knowledge base.

Your data stays yours. Agree up front who has access and where the data is stored: wherever possible in the EU or on your own servers. What that involves is covered in where your data lives with European and sovereign AI.

What this means for manufacturers and technical companies is covered on our page on AI for technology and industry.

Sources

• Rockwell Automation, 93% of Manufacturers Have MES, But Only 23% Have Fully Integrated It (press release for the report Scaling MES Across the Enterprise, 14 July 2026)
• Siemens, The True Cost of Downtime 2024
• Siemens, WinCC/Connectivity Pack system manual: access to archive data via the OLE DB Provider

Frequently asked questions

What is a historian?

A database that stores process data from the plant with a timestamp, such as temperatures, pressures, states and alarms. It lets you see afterwards what a machine or line did before, during and after a breakdown.

Do we need a historian to get started?

Not necessarily. A SCADA system often keeps an archive too. Siemens WinCC, for example, archives process data and alarms that external software can read. An export from the MES or machine monitoring can also be enough for a first test.

How many manufacturers are not using their data?

In Rockwell Automation’s research among 1,560 decision makers in 17 countries, 43% say they are not using their collected data effectively.

How quickly will we know whether it pays off?

That depends on the data available. With an export and a clearly scoped problem, it is clear within a few weeks whether the data contains a usable pattern, before anything is connected or built.

Can AI also adjust the installation itself?

We advise against it. The controls are safety-critical and should stay shielded. The value lies in finding the right cause faster and more often, not in taking over the control loop.