C4, Lesson 1: The Supply Chain Map, and Why Small Wobbles Become Big Swings
Tags: Sourced / General / Our view / Illustrative. About 12 minutes.
1. The story: the diapers that steadily sold, and the orders that did not
In an article published April 15, 1997, in MIT Sloan Management Review, Hau Lee, V. Padmanabhan and Seungjin Whang (Stanford) described how logistics executives at Procter & Gamble examined order patterns for Pampers, one of their best-selling products. Sales at retail stores fluctuated, but not by an alarming amount. Distributors' orders varied more. P&G's own orders of materials to suppliers such as 3M "were even greater." As the authors put it, babies consumed diapers at a steady rate, yet order variability was "amplified as they moved up the supply chain." P&G called this the "bullwhip" effect. [Sourced: Lee, Padmanabhan and Whang, MIT Sloan Management Review, Apr 15, 1997] The article gave a second example: Hewlett-Packard found that orders from a reseller for a printer swung much more than the reseller's sales, and orders from the printer division to the integrated-circuit division swung more still. [Sourced: same article] The authors' scholarly paper, published in Management Science in April 1997, analyzed four sources of the effect: demand signal processing, the rationing game, order batching and price variations. [Sourced: Management Science abstract, Apr 1997] The point for this course: most supply-chain trouble is information trouble. Each company orders from the next one, and the order is a distorted copy of what customers really bought. [Our view]
2. The one idea
A supply chain is a line of companies, each guessing demand from the orders of the one before it. Small swings get bigger at each step. Software, including AI, can help each company guess better, but it cannot fix a distorted signal it is fed. The strongest fix is often to share real sales data across the chain. [General]
3. The kitchen-table version
You sell the $12 bottle to a shop that orders from you every week. Last week the shop sold 100 and ordered 100. This week it sold 110, so it orders 130 to be safe. You see an order for 130, think demand is booming, and order extra caps from your supplier. Customers wanted ten more bottles. Your supplier got an order for far more. [General]
4. The map: five steps from factory to shelf
| Step | Question | Where AI shows up | Covered in |
|---|---|---|---|
| Plan | How much will we sell? | Forecasting | Lesson 2 |
| Source | Who supplies, and what if they fail? | Supplier risk | Lesson 6 |
| Make / hold | How much stock, where? | Inventory, placement | Lessons 3-4 |
| Move | Which load, which route? | Planning and routing | Lesson 5 (and C5) |
| Sell | What is really selling? | Store and online data | C1, C2 |
| [General] |
5. Worked example: how a wobble grows
All numbers [Illustrative], from a script with random demand; a different seed gives different numbers but the same shape. Consumer demand averages 100 a week with a standard deviation of 10. Each of four stages (retailer, distributor, manufacturer, supplier) orders using a simple rule: forecast demand from the last 5 weeks of orders it received, add a safety cushion, and order what it needs to restore that level.
| Stage | Std. dev. of its orders | Compared with consumer demand |
|---|---|---|
| Retailer | 13.9 | 1.4x |
| Distributor | 23.5 | 2.3x |
| Manufacturer | 43.4 | 4.2x |
| Supplier | 77.6 | 7.6x |
| The consumer signal did not change; the ordering rules amplified it. This is the same pattern the P&G and HP executives saw. The cost is excess stock at some steps and shortages at others, plus expediting and overtime (the article lists these symptoms). [Sourced for the symptoms: Lee et al., 1997] | ||
| Notice what AI can and cannot do. A better forecaster at the manufacturer helps only if it can see consumer sales, not just the distributor's orders. If it sees only orders, it will chase distorted data faster. [Our view] |
6. What reduces the whip
- Share point-of-sale data upstream, so every stage plans from real demand. The HP example in the article turned on not seeing sales at the channel stage. [Sourced: Lee et al., 1997]
- Smaller, more frequent orders instead of big batches.
- Steadier prices rather than deep promotions that cause stock-up (links to C3 Lesson 5).
- Honest allocation when supply is short, so customers do not inflate their orders. [General; the four causes are named in the paper: demand signal processing, the rationing game, order batching, and price variations]
7. What it costs and what it breaks
- Trust and contracts: sharing data needs agreements on who sees what.
- Systems: data from many companies is messy and late.
- Over-reaction to AI: a faster forecast on bad data amplifies faster.
- Hidden cost: you pay for the whip in warehouse space, markdowns, rush freight and missed sales.
8. Monday checklist
- Chart your weekly sales and your weekly orders to suppliers. Is the orders line wigglier?
- Compute the ratio of the two standard deviations (Section 5 shows the idea).
- Ask your biggest customer for their weekly sell-through data.
- List your last three promotions and the order spikes that followed.
- Write your five-step map with the owner of each step.
9. Try it yourself (no code, 30 minutes)
Put 12 weeks of your sales and your purchase orders side by side in a spreadsheet. Compute the standard deviation of each (STDEV). Divide orders by sales. A number above 1 is your own bullwhip.
10. Self-check
- What product did P&G examine, and what did they find? Pampers; order swings grew from retail to distributors to suppliers.
- What are the four sources named in the 1997 paper? Demand signal processing, rationing game, order batching, price variations.
- In the example, how much did the supplier's orders vary compared with consumer demand? About 7.6 times.
- Why can a better forecast make things worse? It may chase distorted order data.
- Name one fix. Share real point-of-sale data upstream.
11. Where this is going
More data sharing and faster planning tools may flatten some swings, but only if companies agree to share information. [Our view]
Sources
- Lee, Padmanabhan and Whang, MIT Sloan Management Review, Apr 15, 1997: https://sloanreview.mit.edu/article/the-bullwhip-effect-in-supply-chains/
- Lee, Padmanabhan and Whang, Management Science 43(4):546-558, 1997: https://pubsonline.informs.org/doi/10.1287/mnsc.43.4.546 ; Sloan Management Review article (Apr 15, 1997): https://sloanreview.mit.edu/wp-content/uploads/1997/04/633ecdb037.pdf ; Stanford GSB listing: https://www.gsb.stanford.edu/faculty-research/publications/information-distortion-supply-chain-bullwhip-effect
Review flags
- RESOLVED (earlier verification, not repeated in this pass): the four causes were confirmed by the verifier from the full text of the Management Science paper (43(4):546-558), and both the paper and the Sloan article are cited.
- The 1990s examples are old; they show a pattern, not current figures.
- The simulation is a simple order-up-to rule I wrote; it is a teaching example, not a model of any company. Values are seed-dependent.
- Price promotion cross-link depends on C3 Lesson 5 (draft).
Claim ledger (C4 Lesson 1)
| # | Claim | Source | Verified | Status | Re-check |
|---|---|---|---|---|---|
| 1 | P&G Pampers example; "amplified as they moved up"; 3M | MIT SMR Apr 15, 1997 | 2026-10-06 | Read | none |
| 2 | HP printer example | Same | 2026-10-06 | Read | none |
| 3 | Four sources of bullwhip | Management Science paper 43(4):546-558 (full text, per verifier) | 2026-10-06 | Confirmed | none |
| 4 | Symptoms: excess inventory, poor service, expediting, overtime | MIT SMR | 2026-10-06 | Read | none |
| 5 | Order-variance simulation | Our script | 2026-10-06 | Illustrative, seed-dependent | none |