Making data operational

Across industrial operations, vast amounts of process data are generated every day, yet only a small fraction is used to support decision-making. This underutilisation represents a significant missed opportunity, particularly as processes become more complex and performance margins tighten. If digitalisation is to deliver meaningful impact, data must be converted into knowledge rather than stored in isolation. 

Realising value from process data requires more than dashboards or visualisation. Measurements only become useful when they are interpreted in the context of how the process behaves. Models provide this context by linking measured data to a representation of the process, enabling prediction, insight, and informed decision-making. 

The availability and quality of available data vary widely. Many existing plants operate with access to only essential measurements, shaped by historical design choices and instrumentation constraints. Conversely, newer plants increasingly rely on extensive sensor networks and automated data acquisition, generating large volumes of high-resolution data. In both cases, the challenge is to convert the available measurements into process knowledge. 

Hybrid modelling plays a key role in addressing this challenge. Data-driven models depend on large, high-quality datasets and struggle when extrapolating beyond observed conditions, whilst first-principles models are limited by incomplete understanding of underlying phenomena and practical implementation constraints. Hybrid models blend physics with data science, allowing meaningful insight to be extracted even when data are sparse, noisy, or incomplete. See our previous note for more information on hybrid modelling. 

For data-rich operations, models provide a structured framework to integrate multiple data streams, improve interpretability, and support confident, informed decisions. Models also act as a mechanism for knowledge capture and transfer. They support operator decision-making, accelerate the training of new engineers, and make process understanding accessible beyond a small group of specialists. 

For operations with limited data availability, hybrid models are able to help identify where understanding is lacking and which additional measurements would make the biggest impact. This enables targeted sampling campaigns and focused investment in instrumentation, rather than incremental data collection without clear purpose.  

In conclusion, process data alone does not create value. Value is created when data are embedded within a modelling framework that connects measurements to real-life behaviour, converting data into transferable knowledge. By turning data into a structured, interpretable representation of the process, modelling enables data to be used deliberately and effectively, regardless of whether data are scarce or abundant.