1. The story: Listerine and the missing 13 percent
In the mid-1990s, Wal-Mart and Warner-Lambert, the maker of Listerine, tried something new. Instead of each company guessing separately, they sat down together and compared forecasts, promotion calendars and what was happening at the distribution center. The effort was co-led on the Wal-Mart side by its supply chain and information executives, with a Cambridge, Massachusetts software and strategy firm, Benchmarking Partners. [Sourced: Wikipedia on CPFR] It became known as CPFR: collaborative planning, forecasting and replenishment.
In a pilot reported in June 1997, Wal-Mart raised its in-stock rate on the Listerine line from 87 percent to 98 percent. [Sourced: Supermarket News, June 16, 1997]
In plain words: shelves that were empty 13 times in 100 were empty about 2 times in 100. Nobody built a bigger warehouse. The two companies simply stopped working from two different guesses.
2. The one idea
A forecast is a statistical guess. A demand plan is the one number the whole company agrees to buy, make and ship against.
Oliver Wight, the consultancy that helped originate sales and operations planning in the early 1980s, defines the demand plan as a formal request from sales and marketing to the supply chain to make materials and capacity available when needed. [Sourced: Oliver Wight; the 1980s origin from the same source] Notice the shift: a forecast describes; a plan commits.
3. The kitchen-table version
Your family has a forecast: "We usually eat out twice a week." That is a guess.
Now it is Sunday and everyone agrees: "This week, dinners out on Tuesday and Friday, Mom's visiting Thursday, so cook for six that night." That is a plan. Whoever shops for groceries uses the plan, not the guess.
A company has the same problem at scale. Sales says "we will sell a lot." Finance says "we cannot afford that much stock." Supply says "the factory can only make so much." Demand planning is how they end up with one answer.
4. The pieces of a demand plan
[General; the three-dimension view is also in your planning deck, reworded and with our own example]
The cube. Every number lives in three dimensions: product, place, time. Each can be zoomed in or out.
| Dimension | Zoomed out | In the middle | Zoomed in |
|---|---|---|---|
| Product | All products | Category: drinkware | SKU: bottle, blue, 20 oz |
| Place | The country | A region | One store |
| Time | A year | A month | A week |
Plans are made at the level that fits the decision. A buyer ordering from overseas thinks in "category, country, month". A store replenishes at "SKU, store, week". The art is moving between levels without the numbers contradicting each other.
Assortment. Breadth is how many different lines you carry; depth is how many SKUs within each line. The plan decides both. [General]
Targets. Sales, margin and inventory goals that the numbers must fit inside. [General]
5. Worked example: top-down meets bottom-up
Illustrative numbers. [Illustrative]
The company's category forecast for drinkware next week is 1,000 bottles. Based on past shares, a planner splits it into four SKUs:
| SKU | Share | Planned units |
|---|---|---|
| Blue 20 oz | 40% | 400 |
| Green 20 oz | 30% | 300 |
| Black 32 oz | 20% | 200 |
| White 32 oz | 10% | 100 |
| Total | 100% | 1,000 |
Meanwhile, the stores each send their own estimates, which add up bottom-up to 1,060. The two views disagree by 60 units, six percent.
The meeting is the plan: Are stores expecting a promotion the category forecast missed? Is the category number stale? Is anyone padding for safety? Whichever answer wins, write it down, with who decided and why. That written reason is what the later modules (inventory, transport, cash) rely on.
6. A short history of planning together
| Year | What happened | Why it mattered |
|---|---|---|
| Early 1980s | Sales and operations planning (S&OP) originated with Oliver Wight [Sourced] | Sales and supply meet monthly and agree one plan |
| 1996 | Wal-Mart and Warner-Lambert launch a joint forecasting pilot on 12 products [Sourced: Supermarket News, Nov 11, 1996] | A retailer and a supplier plan together |
| 1997 | Pilot reports in-stock up from 87% to 98% on Listerine [Sourced] | First public proof point |
| 2000s | Industry standards body formalises CPFR; the VICS Association publishes the model [Sourced: NC State tutorial] | Shared language across companies |
| 2020s | AI proposes the plan; people decide exceptions. Gartner expects only 5% of organisations to make 10%+ of planning decisions autonomously by 2030 [Sourced] | Assistant, not autopilot, for now |
7. Do human changes help? A live argument
Many companies track Forecast Value Added (FVA): did a person's manual edit make the forecast better or worse than the computer's number? One reviewed paper in a forecasting journal examines what prompts these adjustments and what they do to accuracy and bias. [Sourced: International Journal of Forecasting 2024, abstract] A 2024 critique argues FVA cannot tell whether an edit was right for the reason the planner believed, so "positive" and "negative" value can mislead. [Sourced: Doherty, Lokad PDF, excerpt]
Practical takeaway for beginners: keep the computer's number, the human's number and the result side by side. After a few months you will see where your people really add value, often on promotions and new items, and where they just add optimism. [Our view]
8. What AI changes
- Splitting and re-splitting numbers across thousands of SKUs and stores, quickly. [General]
- Spotting stale or odd plans, such as a store that always over-forecasts. [General]
- Drafting the first plan and explaining it in words, so the meeting is about decisions. [Our view]
- Not changing: someone must own the number. An unowned plan is a forecast.
Where this is headed (a labelled [Our view]): as software gets better at running the routine part, the human role shifts to negotiating trade-offs and handling surprises. That is a direction, not a finished fact.
9. Try it yourself (no code, 20 minutes)
- Write a week-ahead food plan for your household and a separate list of what each person says they will eat.
- Add up the totals both ways. Do they match? What changed when you talked it through?
- Note one thing a computer could have predicted and one only a person knew.
- Next week, check which list was right, and by how much.
10. Self-check
- What is the difference between a forecast and a demand plan? A forecast is a guess; a plan is the agreed commitment.
- Name the three dimensions of the planning cube. Product, place, time.
- A category forecast of 1,000 is split 40/30/20/10. What does the 30% SKU get? 300.
- What happened in the Wal-Mart and Warner-Lambert pilot? In-stock on the Listerine line rose from 87% to 98%.
- Why keep both the computer's and the human's number? To learn where human edits help and where they hurt.
Sources
- Supermarket News, Nov 11 1996: https://www.supermarketnews.com/meat/wal-mart-warner-lambert-test-interactive-forecasting ; CPFR background (Wikipedia, which dates the origin to 1995, a date the trade press does not support): https://en.wikipedia.org/wiki/Collaborative_planning,_forecasting,_and_replenishment
- Supermarket News, June 16 1997: https://www.supermarketnews.com/mergers-acquisitions/wal-mart-cpfr-pilot-charts-course-for-in-stock-growth
- Oliver Wight, demand plan definition: https://oliverwight-eame.com/effective-demand-planning/ ; S&OP origin: https://oliverwight-eame.com/transitioning-from-sales-and-operations-planning-to-integrated-business-planning/
- VICS CPFR tutorial, NC State: https://scm.ncsu.edu/scm-articles/article/cprf-model-collaborative-planning-forecasting-and-replenishment-cpfr-a-tuto
- FVA: https://www.sciencedirect.com/science/article/pii/S0169207024000736 ; critique: https://www.lokad.com/pdf/doherty-critical-evaluation-forecast-value-added-2024.pdf
- Gartner, Sept 24 2026: https://www.gartner.com/en/newsroom/press-releases/2026-09-24-gartner-predicts-only-5-percent-of-organizations-will-make-at-least-10-percent-of-supply-chain-planning-decisions-autonomously-by-2030