Oversell, split parcels and broken delivery promises across three nodes
An online home-goods retailer ships from two owned DCs (New Jersey and Nevada) plus a 3PL in Texas for oversized items. Inventory is synced to the storefront every 45 minutes. Last quarter 1.8% of orders were cancelled for oversell, concentrated in 40 high-velocity SKUs; 22% of orders ship in more than one parcel because allocation splits between DCs; and delivery dates shown at checkout were missed 14% of the time. Free shipping applies above $75 and average order value is $92. A parcel costs $9.40 on average, and the second parcel in a split order costs $8.10 with no extra revenue. Customer service sees a spike in 'where is my order' contacts after each missed-date batch. The new fulfilment lead wants to consolidate all inventory into the Nevada DC.
1. Why are 40 high-velocity SKUs overselling, and what should change?
Read the worked answer
The mechanism. Inventory is decremented on a 45-minute batched feed across three nodes. On fast movers, two shoppers can both see the last unit and both check out - the storefront is advertising availability that is already consumed. Oversell concentrated in 40 SKUs is the signature: those are the only ones turning faster than 45 minutes.
Check the aggravating factors
- Node-level rather than pooled availability publishing, so a re-allocated last unit can be sold twice
- Returns added back to sellable before physical receipt and grading
- The Texas 3PL's feed latency being worse than the owned DCs - common, and worth isolating
Fixes in order
- Near-real-time reservation at checkout for the top ~200 velocity SKUs; keep batching for the long tail where 45 minutes is harmless.
- Safety stock held back from sellable by velocity class (2–3 units on A-movers).
- Returns counted only once received and graded.
- Daily alerting on negative on-hand by node instead of a quarterly report.
Do not consolidate into Nevada to solve this. It lengthens transit for a large share of the country, likely increases splits in the East, and does nothing about the reservation race.
2. Is the 22% split rate worth fixing, and what decision rule would you use?
Read the worked answer
Size it. At 100,000 orders a year, 22% split = 22,000 orders × $8.10 = $178,200 a year in extra parcel cost, plus incremental packing labour.
The alternative has a cost too. Consolidating to one node adds roughly 1.5–3 days transit for the far half of the country. A later promise date depresses checkout conversion, and later delivery raises WISMO contacts and can erode repeat rate.
Decision rule: split only when the incremental speed is worth more than $8.10 to that order.
- Consolidate with a ≤1-day delay for low-value orders on standard shipping.
- Split when consolidating would push delivery past the displayed promise, or the order is expedited or high-value.
- Never split for a $9 accessory; always split when the alternative breaks a paid delivery commitment.
Validate it: compare repeat purchase rate and contact rate for split vs consolidated orders, matched on value and destination. If split orders show no retention or contact-rate benefit, force consolidation and take the savings.
3. How would you get the 14% missed-date rate down without simply quoting slower delivery everywhere?
Read the worked answer
Separate the causes before fixing
- Carrier transit variance by lane (the usual dominant term)
- Advertised warehouse cut-off later than the real pack cut-off
- 3PL handling time configured shorter than observed
- Orders accepted on a node that has insufficient inventory for the line, so the ship day slips after allocation
Fixes
- Compute the promise per destination zone × node × service from observed transit distributions (90th percentile), not carrier marketing SLAs. A date that holds 90% of the time is the honest promise.
- Add a cut-off buffer: after the real pack cut-off, add a day rather than pretending.
- Allocate before promising - do not quote a date from a node that cannot ship the inventory.
- Show a range ('Arrives Tue–Thu') on high-variance lanes; customers punish broken promises harder than honest ranges.
Business case: each missed-date batch drives a WISMO surge, several dollars per contact plus retention damage. Moving from 14% to ~5% on 8,000 orders a month is worth more than the parcel savings that created the problem.