An AI vendor pitched the same predictive maintenance solution to two clients last month. One was a pharmaceutical manufacturer (process industry). The other was an automotive parts supplier (discrete manufacturing). The vendor’s approach was identical for both.

That’s a problem. Process and discrete manufacturing have fundamentally different characteristics that affect how AI should be applied.

The fundamental differences

Before diving into AI implications, let’s clarify what actually distinguishes these manufacturing types.

Process manufacturing

Continuous or batch production of products through chemical, physical, or biological transformations. Products are typically measured by weight, volume, or concentration rather than counted as units.

Examples: chemicals, pharmaceuticals, food and beverage, oil refining, paper, cement.

Characteristics:

• Material flows through equipment rather than moving between stations
• Product quality depends on process parameters (temperature, pressure, flow rate, etc.)
• Operations are often 24/7 continuous
• Equipment is highly interconnected—problems cascade
• Batch-based production involves complex recipes
• Products may be difficult to inspect after production

Discrete manufacturing

Production of distinct, countable items, often through assembly of components. Products maintain identity throughout production.

Examples: automotive, aerospace, electronics, machinery, consumer goods.

Characteristics:

• Products move between work stations
• Quality can be inspected at various stages
• More flexibility in scheduling and sequencing
• Equipment failures typically affect one station, not the whole system
• Products are often configurable or customised

Many real factories have elements of both. A food company might have continuous processing upstream (mixing, cooking) and discrete packaging downstream.

How these differences affect AI applications

Predictive maintenance

In process industries: Equipment failures often cascade through interconnected systems. A failing pump doesn’t just affect that pump—it affects the whole process train. Predictive maintenance is critical because of the domino effects.

But process equipment often has simpler failure modes (pumps, valves, heat exchangers) with well-understood signatures. Vibration analysis and process data correlation work well.

The challenge: you can’t easily take equipment offline for inspection. Predictions need to be timed with maintenance windows that might be months apart.

In discrete manufacturing: Equipment failures affect the specific station, often with more options for workarounds. A broken robot on one station might allow manual operation or rerouting.

But discrete equipment is more varied—CNC machines, robots, presses, assembly equipment all have different failure modes. Generic solutions work less well; you need models tailored to each equipment type.

Quality control

In process industries: Product quality often can’t be fully assessed until production is complete. You can’t inspect the middle of a batch the way you can inspect a partly assembled product.

AI focuses on process parameter optimisation—predicting quality from the conditions during production. Soft sensors that estimate quality in real-time based on process data are valuable.

The challenge: cause and effect relationships can be delayed. A temperature variation in step 3 might only show up as a quality problem in step 7. Models need to capture these time-lagged relationships.

In discrete manufacturing: Products can be inspected at multiple stages. Computer vision for defect detection is directly applicable—you can see the product as it’s being made.

AI applications include visual inspection, dimensional measurement, and defect classification. The relationship between cause (a setting, a material) and effect (a defect) is often more immediate and traceable.

Scheduling and optimisation

In process industries: Scheduling is heavily constrained by chemistry and physics. You can’t speed up a reaction time. Sequences are often fixed by recipe requirements.

AI adds value in optimising within these constraints: minimising changeover times, reducing energy consumption, maximising throughput without violating process limits.

Batch scheduling in particular is a complex optimisation problem where AI can help—balancing equipment availability, clean-out requirements, quality constraints, and customer demands.

In discrete manufacturing: More flexibility in sequencing. Products can often be made in different orders, on different equipment, with trade-offs between efficiency and responsiveness.

AI scheduling can optimise for multiple objectives: minimising WIP, reducing lead times, balancing lines, responding to demand changes. The solution space is larger, and AI-based optimisation can find better solutions than rules-based approaches.

Data characteristics

Process industries typically have:

• High-frequency sensor data (temperatures, pressures, flows every second or faster)
• Long historical records from distributed control systems (DCS)
• Time-series data with complex temporal relationships
• Strong correlation between process variables

Discrete manufacturing typically has:

• Event-based data (cycle times, counts, defect occurrences)
• More heterogeneous data from different equipment types
• Less temporal continuity (products are discrete events)
• Structured data from MES and ERP systems

These data characteristics affect what AI techniques work best. Time-series models and signal processing are central to process AI. Classification and computer vision are more prominent in discrete.

Implications for AI strategy

If you’re a process manufacturer:

• Prioritise process analytics and soft sensors
• Focus predictive maintenance on cascading failure risks
• Invest in time-series data management
• Look at batch optimisation opportunities
• Consider vendors with process industry expertise (not just manufacturing generally)

If you’re a discrete manufacturer:

• Prioritise computer vision and inspection
• Focus predictive maintenance on bottleneck equipment
• Invest in integrating data from heterogeneous systems
• Look at scheduling optimisation opportunities
• Consider whether your production variability warrants AI or rules-based approaches

The vendor problem

Many AI vendors sell to “manufacturing” generically without distinguishing process from discrete. Their case studies might all be from one type, applied to the other without adaptation.

Questions to ask vendors:

• What experience do you have in our specific type of manufacturing?
• How does your approach differ for process vs discrete?
• Can you show case studies from similar operations to ours?
• What data types does your system work best with?

A vendor who can’t answer these questions probably doesn’t understand the differences.

Hybrid situations

Many facilities combine process and discrete elements. A food plant might have continuous cooking (process) feeding discrete packaging (discrete). A metal fabricator might have continuous coating (process) on discrete parts.

In these cases, you often need different AI approaches for different parts of the operation. The integration points—where process meets discrete—can be particularly valuable to optimise.

Don’t try to force one approach across fundamentally different operations. Match the AI solution to the nature of each area.

Understanding whether you’re fundamentally a process or discrete operation—or some hybrid—is step one in developing an AI strategy that actually fits. Generic approaches from vendors who don’t understand these differences waste time and money.