Summary
The Work Knowledge Graph gives recurring process knowledge a shared structure. Tasks describe independently assignable responsibilities; directed edges describe execution handoffs; a workflow selects a connected subgraph for its particular context. Supplier onboarding and a change to supplier details can therefore reuse a banking-check task when its requirements match, while retaining differences that require separate treatment.
Evidence has to support the connections as well as the tasks. Knowing how to validate banking details and how to authorise payment does not establish that one may immediately follow the other. Generated candidates cite evidence for the workflow, each task and each execution edge. The graph also records observed handoff frequency, represented by shorter distances for common transitions. Frequency helps recover established paths; it does not establish their cost, speed or suitability for a particular organisation.
Context is organised across industry, sub-industry, geography, BPMN level and language. Graph gardeners use gaps in that coverage, missing handoffs and duplicate or sparse tasks to choose bounded improvements. They gather evidence and prepare proposals outside the approved graph. Grounding, structural validation and entity resolution precede admission, with human review required for new connections between workflows even where routine additions can be approved automatically.
Approved changes enter immutable snapshots with their evidence, approvals and parent version. Concurrent proposals are checked for overlap and revalidated when their starting state is outdated. Search uses approved graph versions, and imported workflows retain their origin, workspace and version so that an organisation’s existing process remains distinguishable from a generated best-practice proposal. Shared knowledge can then be examined, corrected and reused without losing the history that explains it.
Enterprises can extend the common foundation in private deployments, keeping their workflows, evidence and gardening activity private. Promotion into the master graph requires explicit client authorisation. The intended benefit is that understanding one process helps the next compatible process start from established knowledge. Passing the graph’s admission checks still does not establish that every workflow assembled from it will execute successfully.