Analytics Maturity Model
A four-level framework for manufacturing analytics
Manufacturing analytics is not a single capability — it is a progression. Each level builds on the one before it, and trying to skip levels almost always fails.
This framework describes four levels of analytics maturity, each answering a progressively more valuable question. Most manufacturers are somewhere in Level 1 or 2. The goal is not necessarily to reach Level 4 — it is to reach the level that delivers the most value for your operation.
Level 1: Descriptive
Manufacturing Intelligence — "What is happening?"
Collecting, organising, and visualising operational data so you can see what is happening on the shop floor. This is the foundation — without reliable descriptive analytics, nothing else works.
Capabilities
- Real-time production dashboards (OEE, uptime, output rates)
- Equipment performance monitoring and trending
- Quality tracking (FPY, defect rates, scrap and rework)
- Cost visibility (cost per unit, downtime cost, scrap cost)
- Delivery performance metrics (OTD, order fulfilment, lead time)
Data Requirements
- Machine connectivity (PLCs, sensors, IoT gateways)
- ERP integration for orders, costs, and scheduling
- MES or manual data capture for quality and production logging
Outcome: You can answer 'how did we perform today?' with data rather than anecdote.
Level 2: Diagnostic
Root Cause Analytics — "Why did it happen?"
Working backwards from metric outcomes to causes. When OEE drops, you don't just see that it dropped — you understand which losses drove the decline and why.
Capabilities
- Root cause analysis (5 Whys, fishbone diagrams, fault trees)
- Pareto analysis of downtime, defects, and losses
- Loss waterfall breakdowns (Six Big Losses)
- Stop-event drill-down by equipment, shift, and product
- Correlation analysis between process variables and outcomes
Data Requirements
- At least 30 days of reliable Level 1 data
- Reason codes for downtime, scrap, and quality events
- Contextual data (shift, operator, product, batch)
Outcome: You move from 'what happened' to 'why it happened' — enabling targeted improvement rather than guesswork.
Level 3: Predictive
Predictive Analytics — "What will happen?"
Using statistical and model-based techniques to anticipate problems before they occur. This is where data moves from reporting the past to informing the future.
Capabilities
- Statistical Process Control (control charts, Cp/Cpk, Western Electric rules)
- Predictive maintenance (vibration analysis, current signatures, thermal monitoring)
- Anomaly detection for early warning of equipment degradation
- Quality prediction models based on process parameters
- Model validation and drift monitoring
Data Requirements
- At least 90 days of clean Level 1 data for baseline model training
- High-frequency sensor data for predictive maintenance
- Statistical expertise or data science capability
Outcome: You can intervene before failures occur — preventing downtime, defects, and waste rather than reacting to them.
Level 4: Prescriptive
Optimisation and Simulation — "What should we do?"
Recommending optimal actions based on data, models, and constraints. This is the most advanced level — where analytics doesn't just inform decisions but actively guides them.
Capabilities
- Golden batch analysis (identifying optimal process parameters)
- Digital twin simulation for what-if scenarios
- Scheduling optimisation under real-world constraints
- Closed-loop control recommendations
- Automated parameter adjustment based on model outputs
Data Requirements
- Stable Levels 1–3 data and processes
- ERP/MES integration for scheduling and constraint data
- Specialist tooling (simulation platforms, optimisation solvers)
- Significant data infrastructure and engineering capability
Outcome: The system recommends the best course of action — you shift from 'we predicted a problem' to 'here is what to do about it'.
The Journey Is Sequential
Each level depends on the one before it. Predictive maintenance models need months of clean sensor data from Level 1. Diagnostic analysis needs reliable metrics to diagnose. Prescriptive optimisation needs validated predictive models to prescribe against.
The most common mistake is trying to implement predictive or prescriptive analytics without having reliable descriptive data. The models may be technically sophisticated, but if the underlying data is inconsistent or incomplete, the outputs will be unreliable.
Start with Level 1. Get it right. Then build upward.