Training
Inferencing
Definition
Example use cases
Unique characteristics
Operational retraining & model recalibration
Model
development & experimentation
Generative
& interactive
Pattern
recognition & anomaly detection
Predictive
Real-time control & autonomous systems
Early-stage model development for domain-specific tasks
Training models on proprietary datasets
Fine-tuning and evaluation
Often supported by enterprise-owned HPC or local compute close to the data
Enterprises are likely to follow this path when increasing adoption of AI uses cases
AI used to update deployed models through repeatable, automated training cycles
Scheduled retraining as new data arrives
Fine-tuning models for updated environment on workflows
Model refresh cycles tied to compliance and risk management
Stable, scheduled workloads with predictable demand
Often anchored in a clear use case
Viable on enterprise-owned infrastructure
AI that creates content or interacts naturally with users through language or multimodal interfaces
Copilots
Chatbots and virtual assistants
Content generation
Inputs often small (prompts),
but retrieval context can be large
Possible multimodal inputs
AI that interprets images, video, audio, or sensor data to identify objects, patterns, anomalies
Image, video, and signal analysis
Quality, fault, and anomaly detection
Document and unstructured data analysis
Focus on what is happening now
Extracts meaning from raw unstructured data
Often streaming continuously
AI that analyses data to forecast or support decision making
Predictive maintenance and failure forecasting
Demand, capacity, and resource planning
Risk and scenario modelling
Forecasts or makes decisions based on structured data
Strong data integration rather than ingestion
AI that controls automated systems in real time with strict latency requirements
Robotics and automation control
Real-time optimisation and orchestration
Safety- or mission-critical system support
Requires low latency
Inputs often from sensors
(cameras, PLCs, LiDAR)
Requires reliable connectivity
AI training carried out in highly iterative environments to develop bespoke models
