In most cases a vendor should consume an existing P&ID graph rather than rebuild it, since a P&ID graph that the company has already built and maintains is a valuable source of truth, and rebuilding it separately creates a second version that will drift. Building a graph of how equipment, lines, instruments and valves connect across a plant takes significant engineering effort, and a company that has done that work has usually also set up processes for keeping it current as the plant changes.
Consuming the existing graph keeps one model of the plant that everyone uses. When a vendor builds its own version, the two models start out close but diverge with every modification, management of change record and drawing revision, and after a year or two people find that different tools give different answers about what is connected to what, which is a real problem when that information feeds isolation planning or procedures.
The practical question is whether the existing graph has the detail and quality the vendor needs, and how updates will flow when it changes. A graph built for one purpose, such as engineering design, may not hold everything needed for another, like isolation points, operating states or links to procedures, so it is worth reviewing a sample section together early to see what is there and what is missing before committing to an approach.
If gaps exist, the better approach is usually to fill them in the existing graph rather than build a separate one, so the company keeps ownership of its own data. That might mean adding attributes, correcting connections or extending coverage to areas that were never modelled, with the vendor contributing to the company’s graph rather than holding its own copy that the company cannot easily use elsewhere.
Updates are where integrations most often go wrong, so it helps to agree how the vendor will learn about changes, whether through regular syncs, notifications when a revision is approved or direct access to the current version. It is also worth being clear about what happens when the vendor finds an error in the graph, since a simple route for reporting it back keeps the source of truth improving.
