The Intelligent Factory: The Intelligent Factory Stack
From existing assets to adaptive performance.
The Intelligent Factory Stack is not an IT architecture checklist: it's a leadership framework. Seven layers show how a brownfield process plant turns the assets and systems it already has into better decisions, disciplined action, and adaptive learning.
The Framework
Seven layers, one direction of flow.
The intelligent factory is not created by replacing everything underneath it. It is created by connecting existing systems, adding context, applying intelligence, embedding that intelligence into workflows, and closing the loop through performance and learning. The stack does not replace the factory: it makes the existing factory readable, actionable, and adaptive. Each layer is explored below, drawn from Leading the Intelligent Factory.
Existing Assets & Process Equipment
The physical operation where value is created or lost, the mill you already own, not the one in a capital plan. The intelligent factory starts with the factory that already exists.
Existing Source Systems
DCS, historians, MES, CMMS, lab systems, shift notes, spreadsheets, tribal knowledge. Data and knowledge are not sitting in one clean place. Treat them as source systems, not obstacles to discard.
Integration Layer
Connects existing source systems so they can work together, without replacing the DCS, the historian, or anything else as a precondition for progress.
Operational Context Layer
The layer that turns data into meaning: connecting raw signals to assets, operating states, grades, maintenance events, quality results, human actions, and outcomes.
Intelligence Layer
Broader than analytics and AI: models, rules, statistical methods, repeatable code, diagnostic logic, and process knowledge that interpret patterns and support recommendations.
Workflow & Decision Layer
Where intelligence enters the work: runbooks, troubleshooting workflows, shift handoffs, exception management, and the six decisions that run every factory.
Performance & Learning Loop
Measurable outcomes: throughput, yield, reliability, energy, profit per hour, and the loop that feeds what was learned back into the system.
Layer 1
Existing Assets & Process Equipment
The intelligent factory starts with the factory that already exists.
The layer
This is the physical operation where value is created or lost: the actual equipment, process units, lines, and machines that make the product, along with the instrumentation and control systems that run them.
Why it matters
This layer reinforces the brownfield truth. The intelligent factory does not start with a greenfield fantasy. It starts with real assets, real constraints, real maintenance history, real control systems, real operators, and real economic pressure. Everything above this layer exists to make these assets perform better, not to replace them.
What It Includes
The mill you already own
- Process units, lines, and machines
- Pumps, valves, drives, and tanks
- Reactors, dryers, and boilers
- Utilities and packaging lines
- Instrumentation
- Control systems
Layer 2
Existing Source Systems
Treat existing systems as source systems, not obstacles to discard.
The layer
This layer makes the brownfield reality explicit: data and knowledge are not sitting in one clean place. They're spread across control systems, historians, maintenance systems, lab systems, operator logs, engineering spreadsheets, and the heads of experienced people.
Why it matters
Mills do not need a new factory, a new historian, or a new DCS as the starting point. They need to connect and activate what they already have. Every system on this list, including the informal ones like shift notes and tribal knowledge, holds a piece of how the mill actually runs. The stack treats them as assets to draw from, not legacy debt to write off.
What It Includes
Where the data and knowledge live
- DCS and PLC systems
- Historians
- Sensors and edge devices
- MES and production systems
- CMMS and maintenance systems
- Lab, quality, and LIMS systems
- ERP, scheduling, and planning systems
- Operator logs and shift notes
- Alarm and event records
- Engineering spreadsheets and calculations
- Manual reports, tribal knowledge, informal records
Layer 3
Integration Layer
The brownfield path is not rip-and-replace. It is connect-and-contextualize.
The layer
The integration layer connects existing source systems so they can work together. It does not replace the DCS. It does not replace the historian. It does not require the plant to rebuild everything before progress can begin.
Why it matters
This is the layer most transformation plans get backwards: treating integration as something that comes after modernization, when it's actually the alternative to it. The category need is clear: the mill needs an integrated layer that connects existing systems and makes them usable together for decisions, workflows, and learning. This is the practical starting point of Discipline 3: Retrofit the Information Layer.
What It Does and Doesn't
Connect, don't replace
- Connects historians, DCS/PLC, MES, CMMS, lab systems, and logs into one usable whole
- Makes source systems work together for decisions, workflows, and learning
- Adds targeted sensors or edge capture only where a decision needs missing information
- Does not replace the DCS or the historian
- Does not require a rebuild before progress can begin
- Does not hold intelligence hostage to a modernization program
Layer 4
Operational Context Layer
Connectivity tells you what happened. Context tells you what it means.
The layer
This is the layer that turns data into meaning, connecting raw signals and events to how the mill actually runs. It's the layer most mills are missing, and the reason a wall of historian tags stays a wall of tags.
Why it matters
This layer is Discipline 2: Context Beats Connectivity made concrete. It also creates the foundation for repeatable intelligence: without context, every use case becomes custom; with context, patterns can repeat across assets, grades, lines, and mills. Two identical readings mean different things depending on grade, speed, operating state, and recent maintenance. This layer is what knows the difference.
What It Connects Signals To
The meaning behind the data
- Asset hierarchy and process relationships
- Operating states
- Products, grades, recipes, and batches
- Constraints and bottlenecks
- Maintenance events
- Quality events and lab results
- Alarm conditions
- Human actions and shift conditions
- Market and production context
- Decisions and outcomes
Layer 5
Intelligence Layer
AI is one ingredient in the intelligence layer, not the whole layer.
The layer
This layer is deliberately broader than “analytics & AI.” AI belongs here, but so do rules, heuristics, statistical methods, engineering calculations, repeatable code, diagnostic logic, and the process knowledge of experienced people: everything that interprets patterns and supports recommendations.
Why it matters
The intelligence layer is not the magic layer. It is useful only when the layers below it provide enough context and the layers above it convert outputs into action. Not every use case needs advanced machine learning: often a well-built calculation, a rule grounded in process knowledge, or a consistent diagnostic check creates more value than a model no one trusts.
What It Includes
More than models
- AI models and machine learning
- Rules and heuristics
- Statistical methods and calculations
- Repeatable code and diagnostic logic
- Process knowledge and pattern recognition
- Natural-language summaries
- Recommendations and knowledge retrieval
- Forecasts and risk scoring
Layer 6
Workflow & Decision Layer
Intelligence creates value when it enters the work.
The layer
Decision support is part of this layer, but the name is deliberately bigger: it includes the runbooks, routines, reviews, and action paths where intelligence actually gets used, and where the Five Decisions That Run Every Factory get made.
Why it matters
This layer keeps the stack from becoming a data-platform diagram. The purpose of the stack is not to move data upward, it is to improve decisions and workflows. This is Discipline 4, Put Intelligence in the Workflow, in structural form: an insight that doesn't enter the workflow is just information.
What It Includes
Where intelligence enters the work
- Intelligent runbooks
- Engineering troubleshooting workflows
- Operator response workflows
- Maintenance triage and reliability workflows
- Quality investigation workflows
- Shift handoff and production meeting summaries
- Leadership review workflows
- Exception management and escalation paths
- The Six Decisions That Run Every Factory
Layer 7
Performance & Learning Loop
Improvement is not complete until the factory learns from it.
The layer
This is the point of the whole stack: measurable outcomes such as throughput, yield, quality, downtime, energy, reliability, safety, cost, margin, and profit per hour, plus the harder-to-measure ones like decision quality, faster troubleshooting, better prioritization, and better knowledge retention.
Why it matters
But outcomes alone aren't the end of the layer. It also includes the loop back into the system: what action was taken, did it work, was the issue closed, and what did we learn? Should the runbook change? The threshold? The model? The maintenance plan? The operating standard? This loop is what makes the whole stack adaptive rather than just instrumented. It is Discipline 7, Build the Adaptive Factory, closing the system.
The Loop Back
Questions every outcome should answer
- Act: what action was taken?
- Verify: did it work?
- Close: was the issue resolved or intentionally deferred?
- Learn: what did we learn?
- Update: should the runbook, threshold, report, model, maintenance plan, or operating standard change?
The Full Picture
Not just a stack, a management system around it.
Two things wrap the seven layers. An outer learning loop feeds outcomes back down into updated context, intelligence, workflows, standards, and operating practice. And industrial discipline runs alongside the upper layers as a management band: ownership, triage, action, verification, closure, and learning, so signals become performance instead of noise. The stack moves data upward; the discipline and the loop are what turn it into results.
How the stack behaves at each maturity level
| Maturity Level | Stack Behavior |
|---|---|
| Reactive | Physical assets and source systems exist, but data is fragmented and learning is weak. |
| Connected | Source systems are connected, but context and workflow fit are limited. |
| Contextualized | Data is tied to assets, states, products, events, decisions, and outcomes. |
| Guided | Intelligence enters workflows and guides recurring decisions. |
| Adaptive | Performance outcomes feed back into updated workflows, models, standards, and learning loops. |
See where your mill stands on The Maturity Model.
Go Deeper
The stack, the disciplines, and the decisions fit together.
The stack shows what to build; The Seven Disciplines show how to lead it; and The Six Decisions show where the value shows up every shift. All three come from Leading the Intelligent Factory.