Ten months of stock on the shelf. Two months is enough.

Executive summary

The category sold 1,495 units in the last twelve months and holds 1,311 units today — 316 days of supply, turning 1.15 times a year. Demand is lumpy at SKU level but stable at family level; a statistically defensible buffer at 95–98% service needs roughly 60 days, not 316.

Demand profile and segmentation

Demand = sales-order and cash-sale line quantities by month (order date, not ship date), subsidiaries 1–3. September 2026 is excluded as an incomplete month. Three product families map one-to-one to the three furniture suppliers.

Patterns

SKU segmentation — top 12 by annual value

ABC on 23-month COGS value (A ≤ 70% cumulative, B ≤ 90%). XYZ on CV of monthly demand (X < 0.5, Y < 1.0, Z ≥ 1.0). ADI = average demand interval in months. Days = 365 × on hand ÷ trailing-12-month units. Full 29-SKU table in Appendix A.

Forecast models

Five methods were fitted on months 1–17 (Oct 2024 – Feb 2026) and scored on a 6-month holdout (Mar – Aug 2026). Smoothing parameters were grid-searched on the training window only. Selection rule: lowest WAPE, overridden only when a model within 5 WAPE points cuts absolute bias by more than 10 points.

SKU-level results and why the forecast is built at family level

The same five methods were run on all 29 SKUs (Croston added for intermittent series). Winners: Holt 13, Croston 9, MA3 3, SES 3, seasonal 1. But SKU-level WAPE ranged 52–100% (median ≈ 70%) — lumpy unit demand of 2–6 per month cannot be forecast precisely at that grain. The operating model is therefore: forecast each family monthly, allocate to SKU by trailing-12-month unit share, and size safety stock per SKU from that SKU's own demand variability. This is standard practice for intermittent A-class items and is what the safety-stock formula in Section 05 assumes.

What makes the model adaptive

Scenario forecasts

Monthly forecast table (units)

* Disruption supply = units the two constrained suppliers can deliver if the event starts 1 Oct: 60% capacity for 10 weeks, then a 6-week recovery lag. Level forecasts are flat because the chosen models have no trend or seasonal term; the small drift in the tables family (28.0 → 28.3) is rounding of a residual Holt-free SES level.

Safety stock and reorder points

Safety stock is computed per SKU and summed to family, using the combined demand and lead-time variability formula, then converted to days of supply. Cycle stock is half the monthly order quantity (monthly replenishment observed on all three suppliers).

Worked examples — 95% service

Working for SER_Box Spring baseline: σ_LTD = √(0.70 × 8.90² + 6.62² × 0.23²) = √(55.45 + 2.32) = 7.60; SS = 1.6449 × 7.60 = 12.5; ROP = 6.62 × 0.70 + 12.5 = 17.1. Disruption: √(1.40 × 79.2 + 43.8 × 0.221) = √120.6 = 10.98; SS = 18.1 — a 45% increase, of which 8 points come from lead-time variance alone.

Family buffers by scenario — pooled across sites

Service level vs cost — baseline, category total, pooled

E[short] = σ_LTD × L(Z) × 12 cycles, summed over SKUs (standard normal loss function). Lost margin = E[short] × family unit margin (Mattress $188, Estes Park $165, tables $102). "Gap" = 123 units the constrained suppliers cannot deliver during the 10-week disruption regardless of safety-stock policy; covering it requires a pre-built reserve of ≈ $30.5K at cost (Section 07).

Working capital impact

Current position: $322.3K on hand against $371.9K trailing-12-month COGS — 1.15 turns, 316 days. Every scenario policy at every service level releases capital; the question is how much and how fast.

Recommendations

Buffer strategy by scenario and class

Scenario-switching triggers

Prioritised actions

Assumptions, data gaps and next steps

Next steps

1. Approve actions #1–#4 and #7. Sonar can draft the inventoryitemlocations ROP / preferred-stock updates with a dry-run diff for approval, and a saved search for the tracking-signal and lead-time triggers.
2. Run the first monthly forecast review at the October close with real September data; the mattress tracking signal is the item to watch.
3. Re-run this analysis in Q1 2027 once one quarter of measured lead times exists; replace the assumed LT and σ_LT and re-size the buffers.

Source queries and method

All figures derive from the SuiteQL below, run against the production account on 16 September 2026 and reduced in a sandboxed worker. Item scope: itemid LIKE any of INV_Estes%, INV_Ascend%, INV_Baja%, INV_Leather Square%, INV_Contour%, INV_Patriarch%, SER_Box%.

SELECT <line CASE> AS line, COUNT(DISTINCT tl.item) AS skus,
       COUNT(DISTINCT TO_CHAR(t.trandate,'YYYY-MM')) AS months,
       MIN(t.trandate) AS first_sale, MAX(t.trandate) AS last_sale,
       SUM(ABS(tl.quantity)) AS units, ROUND(SUM(ABS(tl.netamount)),0) AS revenue
FROM transaction t
JOIN transactionline tl ON tl.transaction = t.id AND tl.mainline='F' AND tl.taxline='F'
JOIN item i ON i.id = tl.item
WHERE t.type IN ('CustInvc','CashSale') AND tl.subsidiary <> 4
  AND i.itemtype IN ('InvtPart','Assembly','Kit')
GROUP BY <line CASE>

SELECT tl.item, i.itemid, TO_CHAR(t.trandate,'YYYY-MM') AS ym, t.type,
       SUM(ABS(tl.quantity)) AS qty, ROUND(SUM(ABS(tl.netamount)),2) AS amt
FROM transaction t
JOIN transactionline tl ON tl.transaction = t.id AND tl.mainline='F' AND tl.taxline='F'
JOIN item i ON i.id = tl.item
WHERE t.type IN ('SalesOrd','CashSale') AND tl.subsidiary <> 4
  AND (<item scope>)
GROUP BY tl.item, i.itemid, TO_CHAR(t.trandate,'YYYY-MM'), t.type

SELECT i.id AS item, i.itemid, i.averagecost, i.parent,
       SUM(il.quantityonhand) AS onhand, SUM(il.quantityonorder) AS onorder,
       SUM(il.onhandvaluemli) AS onhandvalue,
       SUM(CASE WHEN il.location = 5  THEN il.quantityonhand ELSE 0 END) AS oh_la,
       SUM(CASE WHEN il.location = 12 THEN il.quantityonhand ELSE 0 END) AS oh_mia,
       SUM(CASE WHEN il.location = 1  THEN il.quantityonhand ELSE 0 END) AS oh_sf,
       SUM(CASE WHEN il.location = 3  THEN il.quantityonhand ELSE 0 END) AS oh_ny
FROM item i
LEFT JOIN inventoryitemlocations il ON il.item = i.id
WHERE i.itemtype = 'InvtPart' AND i.isinactive = 'F' AND (<item scope>)
GROUP BY i.id, i.itemid, i.averagecost, i.parent

SELECT COUNT(*) AS so_lines,
       SUM(CASE WHEN tl.quantitybackordered > 0 THEN 1 ELSE 0 END) AS bo_lines,
       SUM(COALESCE(tl.quantitybackordered,0)) AS bo_qty,
       SUM(CASE WHEN t.status = 'G' AND ABS(tl.quantityshiprecv) < ABS(tl.quantity) THEN 1 ELSE 0 END) AS short_closed,
       SUM(CASE WHEN t.status IN ('A','B','D','E') THEN 1 ELSE 0 END) AS open_lines
FROM transaction t
JOIN transactionline tl ON tl.transaction = t.id AND tl.mainline='F' AND tl.taxline='F'
JOIN item i ON i.id = tl.item
WHERE t.type = 'SalesOrd' AND (<item scope>)
-- → 653 lines · 2 back-ordered (8 units) · 0 short-closed · 26 open

Method notes

Key derived figures