Moving from Systems of Record to Systems of Understanding

For the last several decades, the manufacturing industry has prioritized the "system of record." Companies have invested billions in ERP and MES platforms designed to do one thing: document transactions. These systems excel at capturing what happened — a part was completed, a batch was cleared, or a shipment was sent.

However, as we move into 2026, many operations leaders are realizing that while they have more data than ever, they lack clarity. The traditional manufacturing stack is passive. It requires humans to manually input data, manually interpret reports, and manually bridge the gap between a raw sensor reading and a corrective action.

The industry is reaching a tipping point where the transactional model of data is no longer sufficient to keep pace with operational complexity. To move forward, the focus must shift from merely recording history to building a system of understanding.

The push for digital transformation was supposed to free up engineering and leadership capacity. In many cases, it has done the opposite. According to research from CoLab Software, engineers now spend between 23% and 25% of their time on non-value-added work, such as searching for files and managing version control.

On the shop floor, the friction is even more pronounced. Research from ScreenCloud and Unily indicates that frontline workers lose approximately 22% of their shift — over 90 minutes per day — searching for or waiting on information. When a system is purely a system of record, the data sits dormant until a human interacts with it. This creates a culture where high-skilled employees spend their shifts feeding the software rather than optimizing the process.

To solve this friction, we have to look at the fundamental way industrial data is structured, often visualized through the DIKW pyramid (Data, Information, Knowledge, Wisdom).

In a traditional ISA-95 stack, the software usually handles the base of the pyramid. It collects raw Data (e.g., a temperature of 100°C) and organizes it into Information (e.g., a line graph of temperatures over time). From there, the system stops. A human, usually a supervisor or engineer, must apply their own Knowledge to analyze the trend and finally reach Wisdom, which is the decision to stop the machine or adjust the cooling flow.

A system of understanding changes this dynamic by collapsing these layers. By integrating artificial intelligence and a more flexible data architecture, the system can move from raw data to wisdom almost instantly. It doesn't just show a temperature spike; it understands that "100°C is too hot for this specific product on this specific machine" and pushes an immediate recommendation to the operator.

One of the primary reasons traditional MES and ERP systems struggle to provide a system of understanding is their inherent reliance on structured data. These systems are built for tables, numbers, and rigid fields, types of data that fit neatly into a database. Yet, the vast majority of what actually happens on a factory floor is unstructured.

Valuable operational context is hidden in sources that traditional systems treat as noise:

• Voice notes from an outgoing shift lead explaining a tribal-knowledge workaround.

• Photos of a defect captured on a smartphone that show the nuance of a quality issue.

• The sound of a motor that an experienced technician recognizes as a precursor to failure.

• Handwritten notes on whiteboards regarding temporary process adjustments.

According to McKinsey, more than 90% of organizational data is unstructured, comprising these types of images, recordings, and raw text. Because traditional ERP and MES systems were built for rigid databases, this windfall of information has historically been invisible to the enterprise.

The result is a massive volume of "dark data", or data that has been captured but never used for analysis or strategic decisions. Splunk reports that 55% of an organization's data is considered dark, untapped, hidden, or unknown. A system of understanding is designed to finally bridge this gap, using multimodal AI and computer vision to turn these dormant, physical signals into actionable wisdom.

The primary drain on manufacturing productivity in 2026 is not the speed of the machines, but the coordination tax, the hidden cost of manual scheduling, status checks, and data movement. When a system is merely a record of history, humans must act as the connective tissue between silos.

The shift toward a system of understanding is not about replacing human judgment; it is about elevating it. When the system handles the routine interpretation of data, such as identifying a defect pattern or verifying a setup, the human is no longer a middleware component. Instead, they are empowered to manage exceptions rather than standards.

This architectural shift addresses a critical structural reality: the talent gap. As systems automate the routine, frontline roles shift from executing repetitive steps to making smart, cross-functional calls. For an operations leader, the value is found in decision velocity. By removing the administrative friction of waiting for information, you allow your most skilled employees to return to the work they were hired to do: solve problems and optimize production.

A system of understanding cannot exist as a monolithic, closed tool. It requires a composable architecture because understanding is a multi-signal problem. Traditional systems of record are designed for linear, milestone-driven changes and work well for long planning cycles and stable processes. But they struggle in an environment where context is continuous and fragmented.

Composability enables this transition in three key ways:

• Decoupling data from logic: Manufacturers can ingest messy, unstructured signals without risking the integrity of the core ERP by separating the data layer from the application layer.

• Incremental evolution: Modular applications allow operations teams to introduce intelligence, such as computer vision, to a single station or cell incrementally, rather than waiting for a site-wide rollout.

• Interoperability: Only a flexible architecture can stitch together disparate signals, such as vision snapshots, machine logs, and operator voice notes, into a shared operational picture that flows across roles.

By building a composable engagement layer on top of existing systems, teams can finally resolve exceptions at the speed of the shop floor rather than at the speed of an IT release cycle.

As operations leaders look toward the next year of technology investment, the goal should not be to collect more data, but to increase decision velocity.

To evaluate if your current stack is moving toward a system of understanding, consider these questions:

1. The Waiting Test: How much time do our frontline operators spend waiting for a supervisor to interpret a system alert or provide the next set of instructions?

2. The Shadow Data Test: How much of our most valuable operational knowledge lives in physical notebooks, whiteboards, or unrecorded conversations?

3. The Admin Test: Are our engineers spending more time solving production problems or cleaning data for the ERP?

The move from record-keeping to understanding is not a single software purchase. It is a shift in how we value data, moving away from historical documentation and toward real-time operational wisdom.