The $3.1 Billion Admission: Data Was Never the Problem

Schneider Electric just paid $3.1 billion for a company most plant managers have never heard of. The purchase says something uncomfortable about a decade of digital transformation: the industry collected the data and still could not act on it.

A very expensive purchase of something invisible

In mid 2026, Schneider Electric announced the acquisition of Cognite, a Norwegian industrial software company, for $3.1 billion in cash, with plans to integrate it into the AVEVA ecosystem. Cognite does not make sensors, robots, or control systems. It makes something far less photogenic: software that connects industrial data to its context.

An equipment tag in a data historian gets linked to the pump it measures, the maintenance history of that pump, the process diagram it sits in, and the production it affects.

Paying roughly eighteen times revenue for that capability is a statement. The largest players in industrial automation have concluded that the scarce asset in manufacturing technology is not data collection, not analytics, and not dashboards. It is the connective tissue that turns a signal into a decision.

We think they are right about the diagnosis. We also think the conclusion looks very different depending on the size and shape of the factory you run.

The dashboard graveyard

Walk through almost any mid sized plant that went through a digitalization program in the last ten years and you will find the evidence on the walls: OEE screens above the lines, downtime Pareto charts in the meeting room, a monthly loss report that lands in inboxes with a thud. The data is there. The losses are also still there, often the same ones, year after year.

This is not a people problem. Plant managers are among the most action oriented professionals we know. It is a structural problem: the systems that see the losses are not the systems that fix them. An analytics tool can tell you that changeovers on Line 4 have drifted 18 percent over standard.

It cannot open a work order, reserve the parts, find a window in the production schedule, or update the changeover procedure. Somewhere between the insight and the action, the thread breaks, and a human is supposed to carry it across three or four disconnected systems by hand.

We call this the action gap, and we believe it is the single largest source of unrealized value in manufacturing today. Not missing data. Unacted insight.

Why every vendor stops at the same wall

The gap exists because every category of software stops at the boundary of what it owns. Analytics platforms see everything and control nothing. Maintenance systems execute work orders but have no idea which ones matter most, because they cannot see production performance or the schedule.

Manufacturing execution systems record what happened with great precision and treat maintenance as an external interruption. Each vendor optimizes its own box, and the decision, which by nature crosses all the boxes, belongs to no one.

Closing the loop therefore requires one of two things: either an enormous integration effort that stitches the boxes together after the fact, or a platform where the boxes were never separate in the first place.

What context actually means: one stop, told twice

The word contextualization sounds abstract, so here is the whole concept in a single event.

Industrial data contextualization is the systematic joining of a machine signal to the operational, temporal, asset, visual, commercial and organizational information that gives it meaning. The signal tells you something happened. The context tells you what to do about it.

Without context: Line 4 stopped at 14:32 for six minutes. That is a row in a downtime log. It will be reviewed at the end of the month, aggregated into a chart, and acted on never.

With context: the line was running its highest margin SKU on an order due Thursday for a key customer. The stop happened 40 minutes after a changeover, still inside the ramp up window. The maintenance system shows this is the third stop involving the same valve block this month.

The camera replay shows the jam actually started at the upstream capper; the filler that reported the stop was simply starved. Same six minutes. Completely different meaning, and a completely specific action: a work order on the capper, scheduled into Thursday's planned stop, before the deadline it threatens.

One event, two readings: the log entry versus the contextualized picture.

Two roads to context

For a global enterprise with thirty years of accumulated systems, context has to be reassembled. The engineering data lives in one archive, maintenance in another, operations in a third, and the relationships between them exist mostly in the heads of people approaching retirement.

Rebuilding those relationships as a knowledge graph is genuinely valuable and genuinely expensive; it is the business Cognite built and the reason Schneider paid what it paid. If you operate refineries or power networks, that road makes sense.

For most manufacturers, there is a second road: capture the context at the source instead of reconstructing it afterwards. When the machine signal, the work order, the production order, and the schedule window are born in one platform with one data model, most of the joins exist by construction.

The stop event already knows which SKU was running, which order it belongs to, and which component it involves, because those facts were never in separate systems to begin with. What the enterprise reassembles for millions, a unified platform captures natively. Native capture also does not mean replacing everything: the platform coexists with the corporate ERP and existing enterprise systems, drawing order and material context from them rather than competing with them.

Neither road is free. The second one still requires disciplined groundwork: a clean asset hierarchy, aligned clocks, honest downtime reason codes. We will be concrete about that groundwork in Part 2, because it is where contextualization is won or lost in practice.

The question that matters

So the useful question for a plant director in 2026 is not whether you have data. You almost certainly do. The question is: when something goes wrong on your most important line, how many systems does the answer have to cross before someone can act on it? Every system boundary the answer crosses is delay, translation loss, and another chance for the thread to break.

In Part 2 of this series we take one contextualized minute apart, join by join, and show exactly what shop floor data needs to know to become a decision. In Part 3 we describe what sits on top of context: the decision layer, where recommendations become approved actions and measured outcomes.

Fabrico is a manufacturing operations platform combining OEE monitoring with computer vision, a full CMMS, MES capabilities and production planning in one data model. fabrico.io