Demand Forecasting & Planning

🔄 Quick Recall: In the previous lesson, you optimized warehouse layout, slotting, and pick paths to increase throughput. Now you’ll tackle the planning side: forecasting what you need, when you need it, and how much safety stock to carry.

Demand forecasting is where logistics meets strategy. Every dollar of excess inventory ties up cash and warehouse space. Every stockout loses revenue and damages customer relationships. AI forecasting improves accuracy by 20-50% over traditional methods by integrating signals that spreadsheet models can’t process.

Multi-Signal Demand Forecasting

Traditional forecasting uses historical sales to predict future demand. AI adds external signals that capture demand drivers invisible to backward-looking models.

Signal categories for AI forecasting:

AI prompt for demand forecasting:

You are a demand planning analyst. Forecast next month’s demand for these products using the following data: (1) last 12 months of weekly sales by SKU [PROVIDE DATA], (2) known upcoming events: [LIST PROMOTIONS, HOLIDAYS, EVENTS], (3) weather outlook: [FORECASTED CONDITIONS], (4) market factors: [ANY KNOWN COMPETITOR CHANGES, ECONOMIC SHIFTS]. For each SKU, provide: point forecast (most likely demand), forecast range (80% confidence interval), key demand drivers, and recommended order quantities including safety stock. Flag any SKUs where demand is likely to deviate more than 20% from recent history.

✅ Quick Check: Your sunscreen sales average 200 units/week. Next week’s forecast shows record heat (105°F). A competitor just raised their prices 15%. Your marketing team is running a summer sale. How should these signals adjust your forecast? (Answer: Each signal pushes demand up: record heat (+25-40%), competitor price increase (+10-15%), and your promotion (+50-100%). Combined, you might see 400-600 units — 2-3x the historical average. Without multi-signal forecasting, you’d order 200 units and stock out by Wednesday. AI integrates these signals automatically, but the key is providing the external data for it to work with.)

Safety Stock Optimization

Safety stock is insurance against forecast errors and supply variability. Too little means stockouts. Too much means carrying costs eating your margins.

AI prompt for safety stock calculation:

Calculate optimal safety stock levels for my top 20 SKUs. For each SKU, I’ll provide: average weekly demand, demand standard deviation, lead time (days), lead time variability (days), target service level (95%/98%/99%), and holding cost per unit per month. Calculate: safety stock quantity, reorder point (lead time demand + safety stock), and annual carrying cost of the safety stock. Then identify SKUs where safety stock can be reduced without significant service level impact, and SKUs where current safety stock is insufficient for the target service level.

Service level vs. safety stock trade-offs:

Key insight: Service level targets should match item classification — 99% for A-items (can’t afford stockouts on top sellers), 95% for B-items, and 90% for C-items.

Promotional and Event Forecasting

Promotions create demand spikes that historical models handle poorly. AI builds promotion-specific forecasts by learning from past promotional patterns.

AI prompt for promotional forecasting:

I’m planning a promotion for [PRODUCT/CATEGORY] starting [DATE] for [DURATION]. Historical promotion data: [LIST PAST PROMOTIONS WITH DATES, DISCOUNT LEVELS, AND ACTUAL SALES LIFT]. Non-promotional baseline demand: [UNITS/WEEK]. This promotion involves: [DISCOUNT %, MARKETING CHANNELS, TIMING]. Forecast: promotional demand by day/week, recommended pre-build quantity, expected post-promotion demand dip (customers who bought early), and recommended inventory positioning to meet demand without overstocking after the promotion ends.

Promotional demand patterns:

✅ Quick Check: Your last 3 promotions showed a consistent pattern: 3x sales during the event, followed by a 30% dip below baseline for 2 weeks after. How should this affect your next promotion’s inventory plan? (Answer: Build enough pre-positioned stock for 3x demand during the promotion, but also reduce your replenishment orders starting at mid-promotion. The post-promotion dip means customers who bought during the sale won’t buy again for 2+ weeks. Over-replenishing during the promotion creates overstock that sits for weeks. AI models this “pull-forward” effect by analyzing how each promotion cannibalizes future demand.)

Key Takeaways

• AI demand forecasting integrates 7+ signal categories (history, weather, economics, competitors, promotions, events, social trends) to improve accuracy by 20-50% over historical-only methods
• Safety stock should match item importance: 99% service level for A-items, 95% for B-items, 90% for C-items — one formula doesn’t fit all SKUs
• Promotional forecasting must account for the full cycle: pre-promotion dip, promotional spike, and post-promotion cannibalization — AI learns from past promotions to predict each phase
• Every dollar of overstock costs 20-30% annually in carrying costs, and every stockout on a fast-mover loses revenue that may never come back — AI forecasting reduces both simultaneously
• Provide external signal data to AI (weather, events, competitor moves) — the more context it has, the better the forecast

Up Next

In the next lesson, you’ll optimize shipping and transportation — route planning, carrier selection, cost analysis, and last-mile delivery strategies powered by AI.