Demand Forecasting Collaboration, Inventory Optimization, and Logistics Coordination
This domain tests the quantitative core of planning: choose a forecast method, measure it honestly, size inventory against variability, and coordinate physical flow without destroying the numbers with the bullwhip effect.
Forecasting: method and measurement
- Qualitative (Delphi, market research, sales composite) for new products and long horizons; quantitative (naive, moving average, exponential smoothing with alpha, Holt-Winters for trend and seasonality, causal regression) for stable item histories.
- Measure with MAD, MAPE (undefined at zero demand), and bias (running sum of errors - a persistent sign means a systematically wrong forecast, not random noise). Tracking signal = running sum of errors ÷ MAD; trending outside control limits triggers a re-forecast.
- Forecast value added (FVA) compares each method against the naive forecast. A sophisticated model that loses to naive destroys value; stop using it.
- Collaboration: monthly S&OP (demand review → supply review → reconciliation → executive approval), CPFR, shared point-of-sale data, VMI.
Inventory: size it, then segment it
- Safety stock = Z × σ(demand per period) × √lead time for independent demand. Reorder point = average demand during lead time + safety stock.
- EOQ = √(2DS/H) balances order cost and carrying cost; it is a starting point, not a policy, when MOQs, shelf life and freight breaks intervene.
- ABC by value, XYZ by variability. Cycle count A items most frequently; count what matters, not what is easy.
- Bullwhip drivers: long lead times, batch ordering, promotions, and shortage gaming during rationing. Remedies: shorter lead times, smaller lots, shared POS data, allocation based on past sales rather than orders.
Logistics coordination
Landed cost includes freight, duty, insurance, handling and demurrage. Mode selection trades speed against cost and emissions; consolidation, milk runs and cross-docking cut cost without cutting service. Performance is measured with OTIF and perfect order index (on time, complete, damage-free, correct documentation) - not with internal fill rate alone.
Worked example
Weekly demand averages 250 units with σ = 60; lead time is 4 weeks at 95% service (Z = 1.65). Safety stock = 1.65 × 60 × √4 = 198 units. ROP = (250 × 4) + 198 = 1,198 units. Raising service to 99% (Z = 2.33) lifts safety stock to 280 units - an 82-unit increase for four points of service, which is why service level is a cost decision, not a default.
Traps that cost marks
- Applying √lead time to lead-time variability instead of demand variability.
- Using MAPE across items with zero-demand periods and reporting a falsely clean number.
- Assuming cost rises linearly with service level - it rises steeply at the tail.
- Confusing OTIF with internal fill rate, or plan attainment with forecast accuracy.