demand-forecasting（需求预测）

Demand Forecasting

Overview

Demand forecasting uses historical sales data, seasonal patterns, and lead times to predict how much inventory you'll need and when to reorder. Chronic stockouts or overstock situations are usually a sign that reorder points are based on intuition rather than data. Purpose-built inventory planning tools handle this for most merchants — custom forecasting code is only necessary for unique operational requirements.

When to Use This Skill

• When chronic stockouts or overstock situations indicate that current reorder points are set incorrectly
• When building automated replenishment recommendations to reduce manual inventory review
• When planning inventory for seasonal peaks (Black Friday, back-to-school, holiday season)
• When you have 12+ months of sales history and want to extract meaningful demand patterns
• When integrating with supplier lead times and purchase order workflows for end-to-end replenishment

Core Instructions

Step 1: Determine your platform and choose the right forecasting tool

Step 2: Set up sales history data collection

Accurate forecasting requires clean historical data. Before running any forecast:

1. Ensure cancelled and refunded orders are excluded from your sales totals — most forecasting tools handle this automatically when connected to your platform
2. Tag promotional periods — flash sales, holiday spikes, and influencer-driven demand should be flagged as abnormal; they inflate baseline demand estimates if included uncritically
3. You need at least 6 months of history for basic seasonal pattern detection; 12+ months is required to see year-over-year trends

Shopify

Using Inventory Planner:

1. Install Inventory Planner from the Shopify App Store (14-day free trial, then $99+/month)
2. Inventory Planner pulls your full Shopify sales history automatically on connection
3. Connect your suppliers in Inventory Planner → Suppliers with their lead times (e.g., Supplier A = 14 days, Supplier B = 7 days)
4. Set your desired service level (e.g., 95% — meaning you want to have stock for 95% of demand scenarios) in Settings → Forecasting
5. Inventory Planner calculates reorder points and recommended order quantities per SKU, updated daily

Shopify Analytics (built-in, no app needed for basic trends):

1. Go to Analytics → Reports → Inventory
2. The "Days of inventory remaining" report shows how many days of stock you have at current sell-through rate
3. Go to Analytics → Reports → Sales over time → group by product to see monthly sales trends
4. Use these as inputs for manual reorder decisions if you don't want to pay for a forecasting app

WooCommerce

Using ATUM Inventory Management (free tier available):

1. Install ATUM Inventory Management for WooCommerce from WordPress.org (free) or purchase the premium version
2. ATUM adds a master inventory list view with real-time stock levels, daily sales rates, and low-stock alerts
3. In ATUM → Settings → Reorder Points, configure your reorder levels and safety stock per SKU
4. ATUM's premium Purchase Orders module generates POs automatically when stock hits the reorder point

Using Inventory Planner for WooCommerce:

1. Connect Inventory Planner to WooCommerce via their native API connector
2. Same workflow as Shopify — Inventory Planner analyzes your WooCommerce sales history and generates forecasts

BigCommerce

Using Inventory Planner:

1. Connect Inventory Planner via the BigCommerce API (Inventory Planner → Settings → Connect Store)
2. Inventory Planner pulls sales history from BigCommerce and generates replenishment recommendations

BigCommerce built-in low-stock alerts:

1. Go to Products → [Product] → Inventory
2. Set "Low stock level" for each product — BigCommerce emails you when stock drops below this threshold
3. This is a simple alert, not a forecast — use it as a backstop while you set up a proper forecasting tool

Step 3: Configure reorder points and safety stock

Reorder point = demand during lead time + safety stock buffer.

In Inventory Planner:

1. Inventory Planner calculates this automatically based on your sales history and supplier lead times
2. Review the recommendations in Inventory Planner → Replenishment — items are sorted by urgency (days of stock remaining vs. lead time)
3. Adjust recommendations manually before creating purchase orders (e.g., if you know a supplier has extra lead time for a specific product)
4. Export purchase orders directly from Inventory Planner to email to your suppliers

Manual calculation (if not using a forecasting tool):

• Average daily demand = total units sold in last 30 days / 30
• Safety stock = (maximum daily demand – average daily demand) × lead time in days
• Reorder point = (average daily demand × lead time in days) + safety stock
• Example: Average daily demand = 5 units, lead time = 14 days, max daily demand = 8 units

- Safety stock = (8–5) × 14 = 42 units - Reorder point = (5 × 14) + 42 = 112 units

Step 4: Plan for seasonal demand

Seasonality is the biggest cause of forecast errors. Plan for it explicitly.

In Inventory Planner:

1. Go to Inventory Planner → Settings → Seasonality
2. Inventory Planner detects seasonal patterns automatically from your sales history
3. For first-year merchants (no prior year data): manually set seasonal multipliers — e.g., December = 3x normal demand for holiday products

Building a seasonal calendar manually:

1. Export your sales by month for the past 2+ years from Shopify Reports / WooCommerce / BigCommerce
2. Calculate the ratio of each month's sales to the annual average (December / average month = seasonal index)
3. Apply the seasonal index to your daily demand forecast when placing orders for the upcoming peak season

Promotional calendar:

• Flag planned promotions (flash sales, influencer campaigns) in your forecasting tool
• Most forecasting tools allow manual demand overrides for specific date ranges
• For Black Friday/Cyber Monday: increase your forecast by your historical BFCM lift percentage (typically 3–8x for e-commerce)

Step 5: Generate and approve replenishment recommendations

In Inventory Planner:

1. Go to Inventory Planner → Replenishment → Review recommendations
2. Filter by "Critical" (stock-out in fewer days than lead time) and "Warning" (stock-out within 2x lead time)
3. Adjust quantities if you have market intelligence the model doesn't know (planned sales, anticipated supply issues)
4. Click "Create Purchase Order" — Inventory Planner generates a PO to send to your supplier
5. Track the PO status in Inventory Planner → Purchase Orders; when received, update actual receipt dates to improve future lead time estimates

Step 6: Custom / Headless — demand forecasting from sales data

// Compute average daily demand from your order database
async function computeAverageDailyDemand(
  productId: string,
  lookbackDays: number = 30
): Promise<number> {
  const since = new Date();
  since.setDate(since.getDate() - lookbackDays);

  const result = await db.raw(`
    SELECT COALESCE(SUM(ol.quantity), 0) AS total_units
    FROM order_lines ol
    JOIN orders o ON o.id = ol.order_id
    WHERE ol.product_id = ?
      AND o.status NOT IN ('cancelled', 'refunded')
      AND o.created_at >= ?
  `, [productId, since]);

  return result.rows[0].total_units / lookbackDays;
}

// Calculate reorder point with safety stock
function calculateReorderPoint(params: {
  avgDailyDemand: number;
  maxDailyDemand: number;   // observed peak daily demand
  leadTimeDays: number;
}): number {
  const safetyStock = (params.maxDailyDemand - params.avgDailyDemand) * params.leadTimeDays;
  return Math.ceil(params.avgDailyDemand * params.leadTimeDays + safetyStock);
}

// Generate replenishment recommendations for all active products
async function generateReplenishmentReport(): Promise<{
  productId: string;
  sku: string;
  currentStock: number;
  reorderPoint: number;
  daysOfSupply: number;
  urgency: 'critical' | 'warning' | 'ok';
}[]> {
  const products = await db.products.findAll({ is_active: true, track_inventory: true });
  const recommendations = [];

  for (const product of products) {
    const inventory = await db.inventory.findByProductId(product.id);
    const avgDemand = await computeAverageDailyDemand(product.id, 30);
    const maxDemand = await computeAverageDailyDemand(product.id, 7); // shorter window = more volatile
    const leadTimeDays = product.supplier_lead_time_days ?? 14;

    const reorderPoint = calculateReorderPoint({ avgDailyDemand: avgDemand, maxDailyDemand: maxDemand, leadTimeDays });
    const daysOfSupply = avgDemand > 0 ? Math.floor(inventory.quantity_on_hand / avgDemand) : 999;
    const urgency = daysOfSupply < leadTimeDays ? 'critical' : daysOfSupply < leadTimeDays * 2 ? 'warning' : 'ok';

    if (urgency !== 'ok') {
      recommendations.push({ productId: product.id, sku: product.sku, currentStock: inventory.quantity_on_hand, reorderPoint, daysOfSupply, urgency });
    }
  }

  return recommendations.sort((a, b) => a.daysOfSupply - b.daysOfSupply);
}

Best Practices

• Start with a tool like Inventory Planner before building custom forecasting — $100–$300/month is far cheaper than engineering time, and the models are more accurate than a hand-rolled moving average
• Set a minimum of 6 months of sales history before trusting any forecast; for new products, use a category-average demand rate as a proxy
• Adjust forecasts for known events — planned promotions, seasonal campaigns, and supply chain disruptions should be entered as manual overrides in your forecasting tool
• Track forecast accuracy monthly — compare forecast units to actual units sold; if you're off by more than 25% consistently, recalibrate your assumptions or look for a structural change in demand patterns
• Account for pending purchase orders — subtract quantity on order from your recommended replenishment qty before placing a new PO; double-ordering is a common and expensive mistake

Common Pitfalls

Related Skills

• @order-management-system
• @vendor-management
• @multi-channel-selling