SQL / Python / Reconciliation

從資料核對，
走到庫存決策。

這個 Demo 不只是顯示表格：它以唯讀 SQL 合併銷售、退貨與庫存，先驗證報表總額，再標出庫存偏高或期間無需求的品項。所有資料皆為虛構。

JOINCTEWindow FunctionRead-only guard

NET SALES

$51,850

ON-HAND UNITS

160

REVIEW

NO DEMAND

DECISION OUTPUT

SQL 結果可以採取什麼行動？

把報表從「數字列表」轉成能被採購、庫存或營運人員討論的例外清單。

ACTION 01

DEMO-A100 庫存偏高

現有庫存 90，約為期間淨需求的 3.2 倍。建議先檢查採購節奏與安全庫存設定。

ACTION 02

DEMO-D400 期間無需求

仍有 40 件庫存，但分析期間沒有淨銷售。建議確認停產、替代品或促銷處理。

NET SALES BY PRODUCT

品項貢獻與異常訊號

銷售金額提供排序依據；顏色則指出需要檢視或期間沒有需求的品項。

DEMO-A100Atlas Valve

$33,600

DEMO-B200Beacon Mixer

$14,400

DEMO-C300Cobalt Panel

$3,850

DEMO-D400Drift Kit

TRACEABLE RESULT

可篩選的 SQL 明細

可依例外狀態或關鍵字縮小資料，並展開查看實際使用的唯讀 SQL。

商品代碼	名稱	分類	淨數量	淨銷售	退貨	現有庫存	訊號
DEMO-A100	Atlas Valve	Fixtures	28	$33,600	0	90	建議檢視
DEMO-B200	Beacon Mixer	Fixtures	9	$14,400	0	18	持續觀察
DEMO-C300	Cobalt Panel	Accessories	11	$3,850	2	12	持續觀察
DEMO-D400	Drift Kit	Accessories	0	$0	0	40	期間無需求

查看這份 Dashboard 使用的唯讀 SQL

WITH line_net AS (
  SELECT
    o.order_date,
    substr(o.order_date, 1, 7) AS sale_month,
    o.order_id,
    o.customer_id,
    l.product_id,
    CASE WHEN l.transaction_type = 'RETURN' THEN -l.quantity ELSE l.quantity END AS net_quantity,
    CASE WHEN l.transaction_type = 'RETURN' THEN -l.quantity * l.unit_price ELSE l.quantity * l.unit_price END AS net_amount,
    CASE WHEN l.transaction_type = 'RETURN' THEN l.quantity ELSE 0 END AS returned_quantity
  FROM sales_orders AS o
  INNER JOIN sales_lines AS l ON l.order_id = o.order_id
),
product_summary AS (
  SELECT
    product_id,
    SUM(net_quantity) AS net_quantity,
    ROUND(SUM(net_amount), 2) AS net_amount,
    SUM(returned_quantity) AS returned_quantity,
    COUNT(DISTINCT CASE WHEN net_quantity > 0 THEN order_id END) AS sales_orders,
    COUNT(DISTINCT sale_month) AS active_months
  FROM line_net
  GROUP BY product_id
),
inventory_latest AS (
  SELECT product_id, on_hand_quantity
  FROM inventory_snapshots
  WHERE snapshot_date = (SELECT MAX(snapshot_date) FROM inventory_snapshots)
),
monthly_rank AS (
  SELECT
    sale_month,
    product_id,
    SUM(net_amount) AS month_amount,
    ROW_NUMBER() OVER (PARTITION BY sale_month ORDER BY SUM(net_amount) DESC, product_id) AS revenue_rank
  FROM line_net
  GROUP BY sale_month, product_id
)
SELECT
  p.product_code,
  p.product_name,
  p.category,
  COALESCE(s.net_quantity, 0) AS net_quantity,
  COALESCE(s.net_amount, 0) AS net_amount,
  COALESCE(s.returned_quantity, 0) AS returned_quantity,
  COALESCE(s.sales_orders, 0) AS sales_orders,
  COALESCE(s.active_months, 0) AS active_months,
  COALESCE(i.on_hand_quantity, 0) AS on_hand_quantity,
  CASE
    WHEN COALESCE(s.net_quantity, 0) = 0 THEN 'No demand in period'
    WHEN COALESCE(i.on_hand_quantity, 0) >= COALESCE(s.net_quantity, 0) * 3 THEN 'Review inventory level'
    ELSE 'Monitor'
  END AS inventory_signal,
  (
    SELECT GROUP_CONCAT(m.sale_month || ' #' || m.revenue_rank, ', ')
    FROM monthly_rank AS m
    WHERE m.product_id = p.product_id AND m.revenue_rank <= 2
  ) AS top_months
FROM products AS p
LEFT JOIN product_summary AS s ON s.product_id = p.product_id
LEFT JOIN inventory_latest AS i ON i.product_id = p.product_id
ORDER BY net_amount DESC, p.product_code;