Hello TOC

TOC in Distribution · Lesson 4 of 8

From push to pull

Better forecasting won’t end the swings. A stock buffer that absorbs uncertainty, refilled with exactly what was sold, will.

12 min · Not completed yet

Challenge the assumptions

TOC holds that there are no conflicts in reality, only wrong assumptions that create them. To challenge an assumption, ask two questions: “Is it so?” and “Is it really so?” Take: I must keep high inventory, because the forecast is inaccurate. Is the forecast inaccurate? Yes, almost always, to some degree. Is it really so, in every situation? Or try inversion: look for a case where the forecast is accurate. Hold that thought.

Can’t we just forecast better?

When a manager says “we need more stock because the forecast was wrong,” the usual answer from above is “then improve the forecast.” But what exactly would we improve?

Ask yourself how many chapatis you eat at lunch. On average, three. But some days four, some days two, and some days none at all. That variation is the uncertainty in demand. Improving the forecast is an attempt to remove it: dig deeper into the last few weeks’ data, add more variables (school hours, play time, body weight), or simply ask the expert, your mother, to guess. None of it changes how hungry you’ll be tomorrow.

Two forces act on the availability of every product at every warehouse:

  • Unreliable supply: different products arriving late at different times.
  • Uncertain demand: sales that jump around from day to day and week to week.

Companies have spent fortunes trying to beat the second one with better forecasts. The two usual methods both disappoint:

  1. Use past sales data. But past sales are distorted: by stock-outs (you can’t sell what you didn’t have), by one-off events, by growth, by schemes and price changes.
  2. Ask the sales team. But people bring their own biases: memories of past shortages and late deliveries, the pressure of targets, the urge to over-order “just in case”.

Pause and think: a company invests in the best forecasting software on the market. What happens to the forecast for one product at one small distributor next month?

  • It becomes accurate
  • It stays wrong, because small numbers at a single location swing by nature
  • It no longer matters
  • It becomes worse than before

Better methods help a little at the aggregate level. But for a single product at a single location over a single month, the noise is in reality itself, not in the method. No software can predict whether three or eight customers walk in next Tuesday.

The scientific approach: absorb the uncertainty

Accept what can’t be changed: uncertainty in demand and unreliability in supply can never be fully eliminated. They are the noise in the system, and sometimes they amplify each other.

Instead of trying to predict the noise, absorb it. Put a stock buffer at each stocking point: a target level of stock that acts as a shock absorber. Consumption takes stock out of it, unevenly, day by day. Supply refills it.

Replenish only what was sold

Now change how the buffer is refilled. Stock-outs and excesses can largely be avoided if each link is supplied what it actually sold or dispatched, not what a forecast says it will need.

  • The distributor reports what it sold.
  • The warehouse sends that, and no more.
  • The warehouse in turn is refilled by the plant with what it dispatched.

The order is a simple calculation:

Order = buffer target − stock on hand − stock in transit

No forecast enters this decision. If a product sold fast this week, more is on its way. If it sold nothing, nothing is sent, and no cash gets locked into it at the distributor.

Pause and think: a distributor’s buffer target for a product is 120 units. It has 45 on the shelf and 30 on a truck. How many should be ordered today?

  • 45
  • 75
  • 120
  • Whatever the monthly forecast says

120 − 45 − 30 = 45. That replaces exactly what was consumed since the last order.

One warning

Pull only works if the starting buffer sizes are right. Start too low and you will have stock-outs; too high and you will hold excess. So the next question is: what decides how big a buffer should be?

Check your understanding

Answer all five. Get four right to complete the lesson.

1. Why does past sales data make a poor forecast?

  • It is always lost
  • It is too accurate
  • It is distorted by stock-outs, one-off events, growth and schemes
  • It is only kept for a month

You never see the sales you lost when you were out of stock.

2. TOC’s approach to demand uncertainty is to…

  • Forecast more often
  • Ignore it
  • Eliminate it with promotions
  • Absorb it with a stock buffer

The buffer is a shock absorber for the noise you can’t remove.

3. The replenishment order equals…

  • Buffer target − stock on hand − stock in transit
  • Next month’s forecast
  • Last month’s sales × 1.2
  • Whatever fills the truck

It replaces what was consumed, nothing more.

4. In a pull system, what is each link supplied?

  • Its forecast for the month
  • What it actually sold or dispatched
  • Whatever the plant has extra
  • The same quantity every week

Supply follows real consumption, link by link.

5. What is the risk when switching to pull?

  • Forecasts become more accurate
  • Trucks become empty
  • If starting buffers are sized wrongly, you get stock-outs or excess
  • Distributors stop reporting sales

Sizing the buffer is the subject of the next lesson.