Infrastructure for Deliverable Recommendation Engine

Deliverable recommendation requires documented standards for what constitutes a valid, reusable deliverable — template standards, quality criteria, and methodology alignment. These exist at L2 in the ps-km baseline: deliverable standards and knowledge sharing processes are defined. However, the gap between documented methodology and actual practice means the system recommends 'official' templates that practitioners consider outdated or impractical, eroding trust in recommendations and driving workarounds.

The recommendation engine requires systematic capture of three data types: project context (industry, phase, service line), document usage patterns (which deliverables were accessed, adapted, and rated as helpful), and outcome linkage (which recommendations led to successful project completion). Template-driven project setup in PSA and CRM captures context fields. Usage analytics from repository access logs capture behavioral signals. Without systematic capture of both, recommendations are based on similarity alone, not demonstrated usefulness.

Recommendations must be scoped to project context — suggesting an assessment-phase deliverable to a consultant in the implementation phase is low-value noise. Consistent schema mapping Project → Industry → Service Line → Phase → Deliverable Type enables the engine to filter candidates before ranking by similarity. The ps-km taxonomy provides this structure. Without it, the recommendation engine would surface every deliverable related to 'digital transformation' regardless of phase, overwhelming consultants with irrelevant results.

The recommendation engine must query document repositories for candidate deliverables, access project metadata from PSA or CRM to understand current context, and retrieve usage analytics. Modern repository APIs and CRM APIs provide this access. The engine can assemble a context package (current project attributes) and retrieve candidates (documents with matching attributes) programmatically. The gap is that proposal and deliverable content in binary formats requires parsing pipelines to generate semantic embeddings for similarity ranking.

Deliverable recommendations that surface outdated templates — last modified in 2020, referencing deprecated methodologies — actively harm project quality. Event-triggered maintenance ensures that when a methodology is updated or a template is retired, the recommendation engine's candidate pool reflects current standards. The ps-km baseline confirms that stale content accumulates; at L3, methodology changes trigger document review and index updates, keeping recommendations current enough for the engine to be trustworthy.

The deliverable recommendation engine needs to know current project context to scope recommendations — ideally pulling this from PSA (current project phase, client industry) and CRM (account sector, service line). However, knowledge repositories are standalone from these systems in the ps-km baseline. The engine works with point integration to document repositories for candidate retrieval and manual or minimal context input from users specifying their project type. Richer PSA integration would improve recommendations but is not available.