Infrastructure for Onboarding Time Prediction & Optimization

Onboarding time prediction requires documented definitions of what 'full productivity' means per role and industry, ramp-up milestones, and the factors that accelerate or delay consultant onboarding. These must be current and findable — not in a staffing manager's head. Without L3, the ML model has no ground truth for what it's predicting: if 'productive' is undefined, ramp-up timelines are unmeasurable.

Predicting onboarding time requires systematic capture of historical productivity curves, training completion, project complexity at time of assignment, and individual background factors. Template-driven capture through the resource management system ensures these fields are populated consistently per assignment. Without structured capture, the ML model lacks the training data needed to identify which consultant and project factors actually drive faster ramp-up.

The prediction model requires consistent schema linking Consultant → Role → Industry → Project → RampUpDuration with standardized fields for experience level, certifications, and onboarding support. Skill taxonomy must map consultant backgrounds to project requirements so the model can identify 'industry familiarity' as a feature. This is the defined-schema level — all records share these fields, enabling feature engineering for ML.

The onboarding prediction system must query the resource management platform for historical assignment data, pull training and certification records, and access project complexity metadata. API access to the primary resource system and HRIS is sufficient for this capability — the model doesn't require real-time streaming but does need to query multiple systems to assemble the feature set for each prediction.

Onboarding prediction models can tolerate periodic refresh rather than event-triggered updates. Ramp-up patterns shift with firm strategy, client complexity, and workforce composition — but these change on a quarterly or annual cycle, not daily. Scheduled periodic review of the training data and model recalibration aligns with the pace of change in onboarding norms, making L2 sufficient without requiring event-triggered updates.

Onboarding time prediction requires point-to-point connections between the resource management system, HRIS for consultant background, and the LMS for training completion. These are the core data sources for ramp-up prediction. Full mesh integration isn't required — the model can consume periodic syncs from these systems. The baseline already has resource-to-HRIS and resource-to-PSA connections that cover the primary data flows.