afrexai-demand-forecasting
// Build accurate demand forecasts using multiple methodologies. Combines statistical models with market intelligence for actionable predictions.
$ git log --oneline --stat
stars:1,933
forks:367
updated:March 4, 2026
SKILL.mdreadonly
Demand Forecasting Framework
Build accurate demand forecasts using multiple methodologies. Combines statistical models with market intelligence for actionable predictions.
When to Use
- Quarterly/annual demand planning
- New product launch forecasting
- Inventory optimization
- Capacity planning decisions
- Budget cycle preparation
Forecasting Methodologies
1. Time Series Analysis
Best for: Established products with 24+ months of history.
Decompose into: Trend + Seasonality + Cyclical + Residual
Moving Average (3-month):
Forecast = (Month_n + Month_n-1 + Month_n-2) / 3
Weighted Moving Average:
Forecast = (0.5 × Month_n) + (0.3 × Month_n-1) + (0.2 × Month_n-2)
Exponential Smoothing (α = 0.3):
Forecast_t+1 = α × Actual_t + (1-α) × Forecast_t
2. Causal / Regression Models
Best for: Products where external factors drive demand.
Key drivers to model:
- Price elasticity: % demand change per 1% price change
- Marketing spend: Lag effect (typically 2-6 weeks)
- Seasonality index: Monthly coefficient vs annual average
- Economic indicators: GDP growth, consumer confidence, industry PMI
- Competitor actions: New entrants, price changes, promotions
Demand = β₀ + β₁(Price) + β₂(Marketing) + β₃(Season) + β₄(Economic) + ε
3. Judgmental / Qualitative
Best for: New products, market disruptions, limited data.
Methods:
- Delphi method: 3+ expert rounds, anonymous, converging estimates
- Sales force composite: Bottom-up from territory reps (apply 15-20% optimism correction)
- Market research: Survey-based purchase intent (apply 30-40% intent-to-purchase conversion)
- Analogous forecasting: Map to similar product launch curves
4. Blended Forecast (Recommended)
Combine methods using confidence-weighted average:
| Method | Weight (Mature Product) | Weight (New Product) |
|---|---|---|
| Time Series | 50% | 10% |
| Causal | 30% | 20% |
| Judgmental | 20% | 70% |
Forecast Accuracy Metrics
| Metric | Formula | Target |
|---|---|---|
| MAPE | Avg( | Actual - Forecast |
| Bias | Σ(Forecast - Actual) / n | Near 0 |
| Tracking Signal | Cumulative Error / MAD | -4 to +4 |
| Weighted MAPE | Revenue-weighted MAPE | <10% for top SKUs |
Demand Planning Process
Monthly Cycle
- Week 1: Statistical forecast generation (auto-run models)
- Week 2: Market intelligence overlay (sales input, competitor intel)
- Week 3: Consensus meeting — align Sales, Marketing, Ops, Finance
- Week 4: Finalize, communicate to supply chain, track vs prior forecast
Demand Segmentation (ABC-XYZ)
| Segment | Volume | Variability | Approach |
|---|---|---|---|
| AX | High | Low | Auto-replenish, tight safety stock |
| AY | High | Medium | Statistical + review quarterly |
| AZ | High | High | Collaborative planning, buffer stock |
| BX | Medium | Low | Statistical, periodic review |
| BY | Medium | Medium | Hybrid model |
| BZ | Medium | High | Judgmental + safety stock |
| CX | Low | Low | Min/max rules |
| CY | Low | Medium | Periodic review |
| CZ | Low | High | Make-to-order where possible |
Safety Stock Calculation
Safety Stock = Z × σ_demand × √(Lead Time)
Where:
Z = Service level factor (95% = 1.65, 98% = 2.05, 99% = 2.33)
σ_demand = Standard deviation of demand
Lead Time = In same units as demand period
Scenario Planning
For each forecast, generate three scenarios:
| Scenario | Probability | Assumptions |
|---|---|---|
| Bear | 20% | -15% to -25% vs base. Recession, market contraction, competitor disruption |
| Base | 60% | Historical trends + known pipeline. Most likely outcome |
| Bull | 20% | +15% to +25% vs base. Market expansion, product virality, competitor exit |
Red Flags in Your Forecast
- MAPE consistently >20% — model needs retraining
- Persistent positive bias — sales team sandbagging
- Persistent negative bias — over-optimism, check incentive structure
- Tracking signal outside ±4 — systematic error, investigate root cause
- Forecast never changes — "spreadsheet copy-paste" problem
- No external inputs — pure statistical = blind to market shifts
Industry Benchmarks
| Industry | Typical MAPE | Forecast Horizon | Key Driver |
|---|---|---|---|
| CPG/FMCG | 20-30% | 3-6 months | Promotions, seasonality |
| Retail | 15-25% | 1-3 months | Trends, weather, events |
| Manufacturing | 10-20% | 6-12 months | Orders, lead times |
| SaaS | 10-15% | 12 months | Pipeline, churn, expansion |
| Healthcare | 15-25% | 3-6 months | Regulation, demographics |
| Construction | 20-35% | 12-24 months | Permits, economic cycle |
ROI of Better Forecasting
For a company doing $10M revenue:
- 5% MAPE improvement → $200K-$500K inventory savings
- Reduced stockouts → 2-5% revenue recovery ($200K-$500K)
- Lower expediting costs → $50K-$150K savings
- Better capacity utilization → 3-8% OpEx reduction
Total impact: $450K-$1.15M annually from a 5-point MAPE improvement.
Full Industry Context Packs
These frameworks scratch the surface. For complete, deployment-ready agent configurations tailored to your industry:
AfrexAI Context Packs — $47 each
- 🏗️ Construction | 🏥 Healthcare | ⚖️ Legal | 💰 Fintech
- 🛒 Ecommerce | 💻 SaaS | 🏠 Real Estate | 👥 Recruitment
- 🏭 Manufacturing | 📋 Professional Services
AI Revenue Calculator — Find your automation ROI in 2 minutes
Agent Setup Wizard — Configure your AI agent stack
Bundles
- Pick 3 — $97 (save 31%)
- All 10 — $197 (save 58%)
- Everything Bundle — $247 (all packs + playbook + wizard)