The bullwhip effect – why small ripples in a shop become crises in the factory

This is the script, written before recording. The text below is what gets said, and the lines in italics are production cues – they are not read aloud. Once the episode is recorded, this text is replaced by what was actually said.

Length: allow roughly 14–16 minutes.

This episode has one sentence carrying it: nobody did anything wrong. Come back to it three times – in the opening, after the arithmetic, and in the summary.

The opening

(Straight in. Numbers first, explanation after.)

A shop sells a hundred units a week.

Further up the chain, at the raw material supplier, an order comes in for a hundred and forty-six.

Nobody lied. Nobody miscalculated. Nobody did anything wrong at all.

And yet the number has grown by almost fifty per cent along the way.

It is called the bullwhip effect, and today we are going to look at why it happens, why it is so hard to get rid of – and what actually works against it.

The nappies

(Graphic: the P&G example. Hold on the nappy point – it is what makes the story good.)

The name arrived in the nineties, but the phenomenon was described as early as 1961 by Jay Forrester.

The breakthrough came with a paper by Hau Lee, V. Padmanabhan and Seungjin Whang in Management Science in 1997. And it opens with an example that has stuck.

Procter & Gamble found that orders for nappies from their distributors swung far more than consumption could explain.

(Here is the point. Give it room.)

And think for a moment about why nappies in particular make such a good example.

Nappy consumption is about the most predictable thing there is. The number of babies does not change from week to week. A baby uses roughly the same number of nappies every day. There is no season, no fashion, no impulse buying.

And still the orders swung sharply.

So the variation did not come from the market. It was created inside the supply chain itself.

The same paper describes the same thing at Hewlett-Packard: resellers’ orders to the printer division swung more than customer demand – and the orders onward to the circuit board division swung more still.

(This is the name of the effect, explained.)

And that last part is the real point. It is not just that the swing exists. It grows with every tier you go up. Like a whip: a small movement at the handle becomes a large movement at the end.

The four causes

(Graphic: the table of four. Go through them quickly – they are expanded on afterwards.)

The paper points to four sources.

Forecasting on a forecast. Each tier builds its own forecast from the orders below it – not from actual sales.

Ordering in batches. You order full pallets or full trucks, not what you sold.

Promotions. The price swings, and purchases shift in time.

The rationing game. When something is out of stock, everyone orders more than they need in order to get enough.

(The most important sentence in the episode. Say it slowly.)

And here is what surprises people: none of the four assumes that anyone is making a mistake.

All four are rational choices for the individual tier.

That is why the effect is so persistent. You cannot remove it by asking people to stop being stupid. Each tier behaves sensibly given what it can see.

The problem is that it cannot see enough.

Margin on margin

(Graphic: the staircase 100 → 110 → 121 → 133 → 146, built up one step at a time. Do not show the whole thing at once.)

Let us take the first cause, because it can be calculated.

Each tier adds a safety margin to avoid running out. That is sensible.

The problem is that the tier above does not see the sales. It sees the order – which already contains the margin. And then it adds its own.

(Build the numbers up one by one.)

The shop sells a hundred. That is actual consumption – the only real number in the whole chain.

The shop orders a hundred and ten. A little extra, in case things pick up.

The wholesaler sees a hundred and ten and adds ten per cent. A hundred and twenty-one.

The manufacturer sees a hundred and twenty-one and adds ten per cent. A hundred and thirty-three.

The raw material supplier receives a hundred and forty-six.

(Pause. Then the explanation.)

The arithmetic is 1.1 to the power of four. 1.46.

And here is why it gets so strong: the margins multiply. They do not add. Four tiers at ten per cent each is not forty per cent – it is forty-six.

Nobody did anything wrong. Every tier added a perfectly defensible margin.

(Note that the figure is an illustration.)

And for the record: the ten per cent is an illustration, not a figure from a particular chain. The point is the shape of the arithmetic, not the size.

One more thing that is easy to forget: if sales fall again next week, the whole chain goes into reverse with the same force.

The calculator

(Mention the tool midway rather than only at the end – it is strongest when it is used to show the zero margin.)

There is a calculator on the site where you can turn these dials yourself.

Two things are worth trying.

Set the margin to five per cent and the tiers to eight. A margin nobody would object to still produces nearly fifty per cent, because the chain is long.

And set the margin to zero. Then every single tier orders exactly what was sold.

(This is the bridge to the solution later in the episode.)

That last one is not an arithmetic trick. It is the whole point, and we will come back to it.

The other three causes

(Graphic: pallet ordering – steady sales versus what the supplier sees.)

Briefly on the other three.

Ordering in batches. A shop selling four packs a day does not order four packs. It orders a carton, or waits until it can fill a pallet, because the freight costs the same whether the truck is full or half empty.

From the supplier’s side it looks like: nothing, nothing, nothing, and then a large order. That is impossible to tell apart from real swings in demand.

And if many shops share an ordering day, all the batches land at once. Then the supplier sees a peak that does not exist in reality – it was made by the calendar, not by the customers.

Promotions. A promotional peak looks like demand, but it is usually purchases moved in time. People buy what they were going to buy anyway, just earlier and more at once. Afterwards comes a matching hole.

The rationing game. When something is out of stock and the supplier allocates what exists, it pays to order more than you need. Everyone does it. And then demand looks even larger than it is.

That pattern repeated itself with semiconductors: carmakers cut orders when the pandemic started, and when demand picked up again they were at the back of the queue – and then ordered more than they needed to secure supply.

What actually works

(Graphic: works poorly versus works, side by side. This is the episode’s conclusion.)

And now the most important part, which runs against what you might expect.

What does not work especially well: better forecasting models at each tier. Larger safety stock to absorb the swing. Better buyers.

The problem is that everyone is still guessing at the wrong number.

(Pause.)

What works is sharing the real number.

Point-of-sale data shared up the chain, so that everyone plans against actual sales rather than against orders. Smaller and more frequent deliveries instead of large batches. Steadier prices rather than big promotions. And allocation based on historical sales, so that over-ordering does not pay when things are scarce.

(Say this one clearly – it is the summary of the whole episode.)

Because bullwhip is not a forecasting problem. It is an information problem.

As long as each tier only sees the order from the tier below, they are planning against a number that is already distorted. And then it does not matter how good the model is.

Walmart is the best-known example of the opposite: sharing till data upward with suppliers, so that everyone sees the same thing.

The link to lean

(Short. This is a nuance, not a new section.)

Finally, a nuance worth having.

Lean removes inventory deliberately, so that problems become visible. And the bullwhip effect is exactly the kind of problem that becomes more painful once the buffers are gone: with no stock, a false peak propagates immediately through the whole chain.

That is not an argument against lean.

But it explains why lean and shared information have to arrive together. Remove the inventory without sharing the numbers, and you have removed the shock absorber and kept the bumps.

Summing up

(Graphic: the points.)

Six things to take with you.

That the bullwhip effect is orders swinging more than sales, and that the swing grows with every tier upward.

That Procter & Gamble found it on nappies – the most predictable product there is.

That the four causes are all rational, and that this is why the effect is so persistent.

That safety margins multiply rather than add.

That a promotional peak is purchases moved in time, not new demand.

And that the answer is not to guess better, but to share the real number.

The ending

(Back to the calculator, and to the opening sentence.)

If you want to see this with your own figures, the calculator is on sporretimen.no. Turn the margin, turn the number of tiers, and watch what happens.

And set it to zero once. The whole chain stands still at a hundred.

That is the difference between guessing better and knowing.

(End card.)