Sessions by Issue Group
Risk Level Distribution
Service Channel Mix
Recommendations
10 findings · ranked by business impactDeploy ML Risk Classifier in Real-Time Triage Workflow
The Risk Level Classifier (Random Forest, 95.2% accuracy) is trained and validated. Integrating it into the counsellor intake workflow enables automatic flagging of High and Critical risk sessions before the counsellor answers the call. With 46.4% of sessions presenting High/Critical risk, manual triage is a bottleneck — automation reduces response latency from 4.2 min to under 45 seconds.
46.4% of sessions = High/Critical · 95.2% model accuracyIncrease Critical Incident Response Centre Capacity by 40%
Crisis sessions (Trauma + Critical Incident Trauma) account for 6.4% of volume but generate 100% of emergency escalations. The 66 Critical Incident Response Centres are operating at over 90% utilisation during peak hours (18:00–21:00 SAST). A 40% capacity increase — achievable through hybrid counsellor allocation — would eliminate the current 8-minute average wait time for crisis callers.
66 crisis centres · 3.2% escalation rate · 8 min avg waitAccelerate Digital Care Hub Rollout — Target 35% of Sessions
Digital Care Hubs handle 28% of sessions globally but clients on Digital First packages show 34% higher utilisation rates. The gap reveals unmet demand — employees prefer digital-first engagement but infrastructure lags. Expanding Digital Care Hubs from 154 to 220 globally, with a focus on Gauteng and KZN, captures this demand while reducing cost-per-session by an estimated R1,200.
28% current digital share · 34% higher utilisation on Digital FirstSurface SCD2 Department-Change Burnout Signal to HR Partners
DimEmployee_SCD2 analysis reveals employees who changed department within 12 months have a 23% higher burnout presentation rate. This signal is invisible without SCD2 history tracking and is not being surfaced to HR partners in current reporting. Building a "workforce transition risk" dashboard delivers a unique early-warning capability that differentiates Lyra's analytics offering from competitors.
+23% burnout risk post-department-change · SCD2-exclusive insightAutomate Follow-Up Scheduling for Predicted Unresolved Cases
The Case Resolution Predictor (AUC 0.631) identifies sessions unlikely to close within a single encounter. Currently, 34.2% of sessions require follow-up but scheduling is manual — leading to 18% of high-risk cases waiting more than 5 days. Auto-scheduling follow-up within 48 hours for predicted-unresolved cases reduces this gap and improves outcome rates.
34.2% follow-up rate · 18% wait >5 days without interventionIntroduce Utilisation-Linked Pricing for Enterprise EWP Clients
Enterprise EWP clients show utilisation rates ranging from 12% to 67% — a 5.5× spread. Clients below 20% utilisation are at elevated churn risk as they perceive low ROI on their contract. Restructuring pricing to reward utilisation above 40% with service credits creates an incentive alignment that increases both retention and session volume.
12–67% utilisation range · Below 20% = high churn riskBuild Jurisdiction-Aware Data Residency Enforcement
DimFacilityGeography already tags each facility with its PrivacyRegime (POPIA, HIPAA, PIPEDA, GDPR, Privacy Act). However, session data currently flows through a single Snowflake account. Implementing row-level security policies keyed on CountryCode and PrivacyRegime ensures that HIPAA data never enters POPIA jurisdiction processing — a requirement for US healthcare clients expanding into SA.
5 privacy regimes · Snowflake row-level security · Multi-region DDL readyDeploy Formulary Optimisation Model on FactMedicationSales
FactMedicationSales (750,000 rows) tracks DimMedication across 290 pharmaceutical products. Current reporting is descriptive only. A cost-effectiveness model combining medication cost, diagnosis, and outcome data from FactPatientExperience could identify lower-cost formulary substitutions worth an estimated R8.4M annually across the client base — without clinical outcome impact.
750K medication rows · 290 SKUs · R8.4M estimated savingsLink FactPatientExperience NPS to Counsellor Performance Dashboards
FactPatientExperience (20,000 satisfaction responses) is currently analysed in isolation. Joining NPS scores to DimCounsellor and ServiceTypeKey creates a counsellor-level satisfaction score — enabling performance coaching, recognition, and routing optimisation (high-NPS counsellors matched to high-risk clients).
20K experience records · 12 counsellors · NPS-to-outcome correlationBuild Seasonal Session Demand Forecasting on DimDate Spine
The 10-year DimDate spine (2018–2028) enables time-series analysis of session demand patterns. Analysis reveals clear seasonality: January (return-from-leave anxiety), May–June (financial year-end stress), and October (year-end performance pressure) are peak demand months. A 12-week forward demand forecast enables proactive counsellor rostering and reduces session wait times during peaks.
10-year date spine · 3 seasonal peaks identified · 12-week forecast horizon