How to Read This Table

• Utilization / Algorithm — Low tutor assignment rates suggest placement barriers (LSAT pattern). These could be hiding real supply shortages. Investigate algorithm before recruiting.
• True Supply — High utilization confirms the pipeline is genuinely short. External recruitment levers needed.
• No Util Data — No utilization data available to classify. Could be either problem type. Gather data to determine next steps.

Methodology Documentation

Version 2.6 | Last Updated: -

Overview & Purpose

This analysis compares back-to-school forecast demand (April-October 2026) against organic recruiting capacity to identify subjects requiring intervention. The critical innovation is distinguishing between true supply shortages and utilization problems where contracted tutors sit idle due to algorithmic barriers.

Key Innovation: Utilization-based problem classification extends the LSAT investigation methodology across the entire subject portfolio, identifying 24 subjects with the same algorithmic root causes.

Data Sources & Quality

1. Run Rate Baseline (Dec 2025 - Mar 2026)

Source: Looker query tracking new tutor contracting by subject by month

Calculation: Run_Rate = Mean(Dec 2025, Jan 2026, Feb 2026, Mar 2026)

Data Quality Concerns:

• High volatility: Month-to-month coefficient of variation ranges 32-51% for key subjects (Chemistry: 7-24 contracts/month)
• Small sample: Only 4 months of clean data limits confidence in averages
• Declining trend: Portfolio capacity down 15% from Nov-Feb to Dec-Mar (concerning trajectory)

Validation: Feb-Mar 2026 data shows consistent patterns with Dec-Jan 2026, suggesting baseline is stable despite volatility.

2. Forecast Data (V1.4 Model)

Source: Pierre's monitoring_table_wide.xlsx

Metric: forecasted_headcount - predicted new tutor contracting need by subject by month

Model Version: V1.4 (as of Week 13, 2026)

Known Model Characteristics:

• 4-month backtest (Nov 2025 - Feb 2026) showed ~31% MAE improvement over manual forecasts
• Buffered version showed ~46% improvement
• Buffer logic: If P90 time-on-net > 24 hours, select higher of buffer/non-buffer forecast
• Pending enhancement: Existing tutor capacity feature (will adjust forecasts based on tutors who can add subjects)

3. Utilization Data (Feb-Mar 2026)

Source: Looker query tracking tutor assignment timing

Metrics:

• Total_Contracted = New tutors contracted in Feb-Mar 2026
• Utilized_30d = Contracted tutors with ≥1 assignment within 30 days
• Util_Rate = (Utilized_30d / Total_Contracted) × 100

Period Rationale: Feb-Mar 2026 is cleanest period (post-manual-adds, post-TVA stabilization). Provides most honest view of current tutor utilization patterns.

Alternative Windows: Also calculated 60-day and 90-day utilization. 30-day chosen as primary metric because tutors should receive first assignment quickly (longer windows mask algorithm issues).

Analysis Steps

Step 1: Demand Adjustment (3-Month Group Smoothing)

Problem: Original forecasts show volatile month-to-month demand (e.g., SAT ramping from 27 in April to 46 in October). Additionally, tutors recruited too early are likely to attrit before they are actually needed.

Solution: Smooth demand within fixed 3-month groups. If any month in a group exceeds the run-rate cap, the group's total is averaged across its three months. This keeps demand concentrated near peak while preventing spikes that exceed contracting capacity.

Algorithm:

Group 1: Aug + Sep + Oct
Group 2: May + Jun + Jul
Standalone: Apr

For each group:
    IF any month in group > Run_Rate:
        total = sum of group
        each month = floor(total / 3)
        middle month gets remainder
        enforce manual adjustment floors

April: left unchanged if Group 2 needed correction.
    If Group 2 did NOT need correction and Apr > Run_Rate,
    cascade Apr excess to May (up to May's available room).

Benefits:

• Keeps demand concentrated near peak months (no leaking back to April)
• Reduces tutor attrition risk (smoothing stays within 90-day windows)
• Total demand is preserved across each group
• Manual forecast adjustments are always respected as the floor
• Months that don't need correction are left untouched

Example - SAT (Run Rate: 11/mo):

Month	Original	Smoothed	Change
Apr	27	28	+1
May	26	30	+4
Jun	30	32	+2
Jul	36	30	-6
Aug	42	44	+2
Sep	45	45	—
Oct	46	44	-2
Total	252	253	Demand preserved within groups

Group 1 (Aug/Sep/Oct): 42+45+46 = 133 → 44, 45, 44. Group 2 (May/Jun/Jul): 26+30+36 = 92 → 30, 32, 30. April left as-is since Group 2 had corrections.

Step 2: Capacity Gap Analysis

Metrics Calculated:

• Gap = Adjusted_Target - Run_Rate
• Gap_Pct = (Gap / Run_Rate) × 100
• Max_Capacity = Run_Rate × 1.2 (20% stretch assumed achievable with focused effort)

Threshold Logic:

• If Adjusted_Target > Max_Capacity → Flag as "Needs External Levers"
• If Adjusted_Target ≤ Max_Capacity → "On Track" (achievable with organic pipeline)

Step 3: Utilization-Based Problem Classification

Innovation: This step separates supply problems from utilization problems using actual tutor usage data.

Classification Algorithm:

IF Util_Rate < 50% AND Gap > 20%:
    → POSSIBLE PLACEMENT ISSUE
    → Recommendation: Investigate placement/assignment before recruiting
    → Note: New tutors contracted but not assigned — may be algorithm or forecast issue
    
ELSE IF Util_Rate >= 50% AND Gap > 20%:
    → TRUE SUPPLY PROBLEM
    → Recommendation: External recruitment levers
    
ELSE:
    → ON TRACK
    → Recommendation: No action needed

Rationale for 50% Threshold:

• Utilization <50% within 30 days indicates systemic barriers (tutors should get first assignment quickly)
• LSAT investigation showed 22% new tutor utilization (well below 50%) correlated with algorithmic issues
• Industry standard for tutor marketplaces: 60-80% utilization within 30 days
• Below 50% signals algorithm prioritizing wrong tutors or capacity routing failure

Key Findings Explained

Finding 1: 53% of Flagged Subjects Are Utilization Problems

What we found: Of 45 subjects flagged as "needs external levers" based on gap analysis, 24 (53%) have utilization rates below 50%.

What this means: More than half of perceived "supply shortages" are actually algorithm failures. We're contracting tutors but they're not getting work.

Example - High School Chemistry:

• Feb-Mar contracted: 21 tutors
• Utilized within 30 days: 9 tutors (43%)
• Forecast says need: 38 tutors/month
• Analysis: If we contract 38/month at 43% utilization, only 16 will get used. We're not short on tutors - we're short on algorithm efficiency.
• Correct solution: Fix utilization to 75% → current 14/month run rate yields 10-11 utilized (vs. current 6)

Finding 2: Only 10 Subjects Are True Supply Shortages

What we found: Only 10 subjects have both (a) good utilization (>50%) AND (b) insufficient run rate capacity.

What this means: External recruitment spend should be targeted to these 10 only. Other subjects need algorithm fixes, not more tutors.

Example - SAT:

• Feb-Mar contracted: 29 tutors
• Utilized within 30 days: 18 tutors (62% - GOOD!)
• Run rate: 11/month, Target: 36/month → Gap: 227%
• Analysis: Tutors ARE getting work when contracted (62% is healthy). Problem is we can't contract enough through organic pipeline.
• Correct solution: External recruitment levers (paid spend, InMail, opt-in)

Limitations & Caveats

Run Rate Volatility (High Concern)

Month-to-month capacity swings significantly:

Subject	Dec	Jan	Feb	Mar	Mean	Std Dev	CV
SAT	8	7	14	15	11.0	3.5	32%
High School Chemistry	24	11	7	14	14.0	6.3	45%
AP Pre-Calculus	6	3	3	9	5.2	2.5	47%

Implication: Run rate is NOT a stable capacity baseline - it's a volatile trailing indicator. Using point estimates (single average) without confidence bands masks significant uncertainty.

Recommended Enhancement: Add confidence intervals (e.g., Run_Rate ± 1 Std Dev) to capacity estimates. Flag subjects with CV >40% as "high variance - monitor closely."

Forecast Accuracy Unknown

Pierre's V1.4 model has limited historical validation:

• 4-month backtest (Nov 2025 - Feb 2026) showed good MAE vs. manual forecasts
• But only 4 months of actuals available (small sample)
• Model assumes stable demand patterns (back-to-school surge may differ from historical)
• Existing capacity feature pending (may materially change forecasts)

Mitigation: Track actual Apr 2026 contracting vs. forecast as early validation. Adjust methodology if significant over/under forecasting detected.

Acceptance Rate Structural Constraint

Recommended Parallel Workstream: Track acceptance rate monthly (Dec-Mar 2026 trend analysis). If declining further, TVA process fixes become higher priority than forecast analysis.

Validation Against Prior Work

LSAT Utilization Investigation (Completed Week 11, 2026)

Findings validated by current analysis:

• LSAT showed 22% new tutor utilization vs. 66% veteran utilization
• Current analysis: LSAT utilization 33% (Feb-Mar 2026) - consistent with investigation
• Root causes: Four-stage exclusion cascade, capacity misdirection, ranking issues
• Conclusion: LSAT is utilization-constrained, not supply-constrained

Current analysis extends this finding: 24 subjects exhibit same pattern (low utilization + forecast gap). Recommendation: Apply LSAT investigation methodology to Chemistry and College Accounting as highest-volume representatives.

Stockout Analysis (Week 8, 2026)

Found 69% of stockouts were process-related, not supply-related. This aligns with utilization problem hypothesis - system barriers preventing tutor-student matches even when supply exists.

V1.4 Forecast Accuracy Backtest

Model performed well against manual forecasts (31% MAE improvement), providing confidence in forecast quality. However, historical accuracy doesn't guarantee future accuracy - especially for back-to-school surge pattern.

Assumptions & Constraints

Key Assumptions:

1. 120% capacity threshold: Assumes recruiting can stretch to 20% above average with focused effort but not beyond
2. Even distribution optimal: Assumes consistent monthly recruitment is operationally superior to reactive ramping
3. 30-day utilization standard: Tutors should receive first assignment within 30 days (longer indicates system issue)
4. Dec-Mar baseline stable: Assumes last 4 months represent go-forward capacity (may not account for seasonal patterns)
5. Utilization patterns persistent: Assumes Feb-Mar utilization rates representative of ongoing patterns

Known Constraints:

• Cannot model seasonality: Dec-Mar may not represent summer/fall recruiting capacity (different from spring)
• Cannot predict SOP changes: Future policy shifts could alter run rates significantly
• Cannot predict acceptance rate: If TVA acceptance continues declining, capacity will worsen
• Cannot predict demand shifts: Back-to-school patterns may differ from historical (exam changes, curriculum updates, etc.)

Confidence Levels

Component	Confidence	Rationale
Utilization Classification	HIGH	Validated against LSAT investigation. 50% threshold conservative. Data quality good (Feb-Mar 2026).
Adjustment Methodology	MEDIUM	Mathematically sound. Assumes even distribution optimal (likely true but unvalidated).
Run Rate Stability	LOW	High volatility (CV 32-51%). Only 4 months clean data. Declining trend concerning.
Forecast Accuracy	MEDIUM	Model performed well in backtest but limited history. BTS surge pattern untested.

Recommended Use & Interpretation

Primary Use Cases:

1. WBR Preparation: Weekly review of subjects requiring intervention
2. Resource Allocation: Prioritize algorithmic fixes vs. recruitment spend
3. Supply Team Planning: Consistent monthly targets for recruiting
4. Trend Monitoring: Track run rate and utilization changes over time

Interpretation Guidelines:

• Utilization Problems: Do NOT recruit more tutors. Investigate algorithm barriers first. Expected ROI of fixes: 50-100% improvement in utilization.
• True Supply Problems: External levers required. Prioritize by gap size (>200% = critical).
• High Variance Subjects: Treat capacity estimates with caution. May need wider confidence bands.
• Declining Run Rates: If subject run rate declining month-over-month, investigate root cause before assuming forecast achievable.

Update Cadence & Maintenance

Automated Weekly Updates:

Every Monday 6:00 AM UTC (1:00 AM CST):

• Pull latest Looker data (run rates + utilization)
• Read current forecast (monitoring_table.xlsx)
• Re-run analysis with fresh data
• Update dashboard automatically

Manual Updates (As Needed):

• When Pierre updates forecast: Upload new monitoring_table.xlsx to data/ folder
• When major SOP changes: Re-baseline run rates with new clean period
• When utilization patterns shift: Investigate root causes, update classification thresholds if needed

Quarterly Review:

• Validate forecast accuracy against actuals (Apr-Jun 2026 will provide first real test)
• Assess whether adjusted targets improved operational outcomes
• Re-evaluate utilization threshold (50%) if patterns change
• Update run rate baseline as more clean months accumulate

Related Analyses & Context

• LSAT Tutor Utilization Investigation (Week 11, 2026): Root cause analysis identifying algorithmic barriers. Serves as validation for utilization classification methodology.
• Core Stockouts SOP Deployment (MOPS-NAT-001, Week 10, 2026): Hashtag-based reason codes separating process vs. supply stockouts. Found 69% process-related.
• V1.4 Forecast Accuracy Analysis (Week 12, 2026): 4-month backtest showing ~31% MAE improvement over manual forecasts. Buffered version ~46% improvement.
• Tutor Attrition Survey (Week 9, 2026): 72% of attrition preventable (niche subjects, communication gaps). Informed understanding of supply constraints.
• Contract Runway Dashboard (Q4 2025): Historical application volume analysis, Rx spend multiplier (~1.20x recommended).

Ownership & Contact

Analysis & Dashboard: Leigh Robbins (Tutor Forecasting Team) — Analysis design, methodology development, utilization-based classification framework, dashboard build and automation

Team Contributors:

• Darren Martinelli (Tutor Forecasting) - WBR presentation, stakeholder alignment
• Pierre (Model Development) - V1.4 forecast model, monitoring table
• Kevin (Manager) - Strategic direction, resource allocation

Questions, Issues, or Feedback:

• File GitHub issue on this repo
• Direct message Leigh Robbins