ETL and Capturing the Digital Twin of the Organisation
Capturing the Digital Twin of an Organisation’s information system requires continuous monitoring of the system’s telemetry. The captured telemetry is aggregated into a real-time digital representation of the executing activity. This representation unambiguously exposes all sequential, concurrent and parallel activity within a single, unified representation of the implementation.
Overview
ETL process mining is only able to infer a representation of the system’s activity for two reasons:
- Source data is the system’s event logs. Event logs are designed to provide post- fact system administration information, not to provide visibility of the system’s telemetry.
- ETL assumes sequential system behaviour. ETL was not designed to capture concurrent (overlapping temporal intervals) and parallel (identical temporal data) activity. When such activity is evident ETL either re-interprets this behaviour as serialised or fails to interpret it at all.
As a result all ETL output requires “fitment”. This is the manipulation of the output to align – or “fit” it – with observed system behaviour. This is inevitably imperfect, with typically a 75%- 80% “close enough” result to the actual system behaviour being generated.
Attempts to improve ETL system representations focus on migrating from only modelling “case-specific” system activity – for example tracking a single transaction ID – to the examination of interconnected case-objects. This requires converting the conventional event logs into “Object-Centric Event Logs” (OCEL). By representing events as interacting “objects” the lifecycles of multiple objects may be analysed simultaneously.
This enables elements of non-serialised system behaviour to be identified. This, however, is an extremely complex, multi-step process. The procedure for constructing an OCEL database and extracting a representation of non-serialised system behaviour is examined below.