Feedback Delay Analysis

Thinking Tools4.1 Root Causes and the Case for Going Back

From the system dynamics of Jay W. Forrester (Industrial Dynamics, 1961), shown in John D. Sterman's beer distribution game paper (Modeling Managerial Behavior, 1989), and set out by Donella H. Meadows (Thinking in Systems, 2008).

Finds the time lags in a feedback loop that make sensible responses overshoot and swing, and points to fixes such as shorter delays, gentler responses, or acting on leading signals.

The same ordering rule settles with a one-week delay and swings with a four-week one

The store orders to refill its shelf target of 400 units after a rush emptied half of it. At load deliveries take four weeks and the store ignores what is already on order: Play, and stock overshoots, then falls short. Set the delay to one week, or count the stock on order, and replay: the same store settles on target. Illustrative.

Select a loop to trace it, or a box to read it.B1, Refill the shelf. A shortfall prompts orders, which arrive after the delay and close the shortfall. The longer the delay, the more the store has over-ordered by the time the first delivery lands. One negative (−) link: an odd count, so the loop balances.B2, Count what is coming. Stock already on order reduces the next order. This loop works only if the store counts it; at load it does not. One negative (−) link: an odd count, so the loop balances.Shelf target. 400 units: four weeks of sales at 100 a week.Shortfall. Target minus what is on the shelf. The store orders half the shortfall each week, on top of the week's sales.Orders placed. This week's order to the distributor.Stock on order. Ordered but not yet delivered. Deliveries arrive spread over the delay.Stock on the shelf. 200 units in week 0, after a rush. Sales take 100 a week; deliveries add to it.
4
0%

Stock on the shelf, units

0: 2001: 2002: 2003: 2254: 2695: 3236: 3817: 4338: 4759: 50310: 51411: 50912: 49113: 46514: 43315: 40116: 37317: 35218: 34019: 33620: 3410200400600012345678910111213141516171819200: 200: 2002: 200: 2254: 269: 3236: 381: 4338: 475: 50310: 514: 50912: 491: 46514: 433: 40116: 373: 35218: 340: 33620: 341020040060002468101214161820

Orders placed, units

0: 1001: 1002: 2003: 2004: 1885: 1666: 1387: 1108: 839: 6210: 4911: 4312: 4613: 5414: 6815: 8416: 9917: 11318: 12419: 13020: 132050100150200012345678910111213141516171819200: 100: 1002: 200: 2004: 188: 1666: 138: 1108: 83: 6210: 49: 4312: 46: 5414: 68: 8416: 99: 11318: 124: 13020: 13205010015020002468101214161820

Week 20 of 20

Stock on the shelf, units
341
Stock on order, units
447
Orders placed, units
132

B1, Refill the shelf
A shortfall prompts orders, which arrive after the delay and close the shortfall. The longer the delay, the more the store has over-ordered by the time the first delivery lands. One negative (−) link: an odd count, so the loop balances.
B2, Count what is coming
Stock already on order reduces the next order. This loop works only if the store counts it; at load it does not. One negative (−) link: an odd count, so the loop balances.
Shelf target
400 units: four weeks of sales at 100 a week.
Shortfall
Target minus what is on the shelf. The store orders half the shortfall each week, on top of the week's sales.
Orders placed
This week's order to the distributor.
Stock on order
Ordered but not yet delivered. Deliveries arrive spread over the delay.
Stock on the shelf
200 units in week 0, after a rush. Sales take 100 a week; deliveries add to it.

An example

A chain of hardware stores kept swinging between empty shelves and overflowing back rooms on its fast-moving lines. Each store manager followed the same sensible rule: order the week's expected sales plus half of whatever the shelf was short of its target. The regional director suspected the managers and asked for tighter ordering discipline.

The supply chain analyst drew the loop instead and listed the delays in it. The store learned of a shortfall at the weekly count, ordered the next day, and the distributor's deliveries arrived over the following four weeks. The rule took no account of the four weeks of stock already on its way, so after any rush each store kept ordering against a shortfall that was already filled, and every delivery that landed in week five was surplus. The chain changed two things: stores counted stock on order as if it were on the shelf, and the distributor moved the fast lines to a one-week delivery. The swings on those lines all but stopped within a quarter, with the same managers following almost the same rule.

Why it works

A balancing loop corrects a gap by acting on it, and that works when the result of the action shows up quickly. When the result arrives late, the person correcting the gap keeps seeing it open and keeps acting, so by the time the first correction lands there are several more behind it. The system overshoots, the correction reverses, and it overshoots the other way. The swing comes from the structure, not from anyone's poor judgment.

Comparing the total delay with how fast the system moves shows whether to expect trouble. The fixes follow from the structure: shorten the delay, respond more gently to each gap, or act on information that leads instead of lags, such as stock already on order, orders in the pipeline, or offers already accepted.

The thinking behind it

Jay W. Forrester Jay W. Forrester, an MIT engineer and management theorist, founded system dynamics: the modeling of organizations as stocks, flows, and feedback loops with delays. Industrial Dynamics (1961) drew on a study of Sprague Electric and gave the first account of what is now called the bullwhip effect, in which small changes in customer demand grow into wide swings in orders and inventory as they pass back up a supply chain through its delays.

John D. Sterman John D. Sterman, director of the MIT System Dynamics Group, showed with the beer distribution game that managers misread feedback and delays, ordering in ways that create wild swings in inventory. He reported the experiment in Management Science in 1989, in a paper on how managers misperceive feedback in dynamic decisions. His Business Dynamics (2000), the standard textbook of the field, covers the stocks, flows, feedback loops, and delays behind such swings.

Donella H. Meadows Donella H. Meadows, an environmental and systems scientist and lead author of The Limits to Growth (1972), explained stocks, flows, feedback loops, and delays for general readers in Thinking in Systems (2008), edited by Diana Wright after her death. Her ranking of places to intervene in a system puts the length of delays, relative to how fast the system changes, among them.

How to run it

  1. Name the swing. Inventory whipsaw, hiring booms and freezes, prices that overshoot: one pattern, with its period.
  2. Map the loop. From the gap, to the response, to the result that closes the gap.
  3. List every delay in it. Seeing the gap, deciding, acting, and the system responding, each with its length.
  4. Compare the delay with the system's pace. A delay as long as the time the gap takes to change is enough to cause swings.
  5. Find what the responder cannot see. Usually the corrections already made and not yet landed, such as stock on order or people already hired.
  6. Pick a fix. Shorten the delay, respond to each gap more gently, or count what is already in the pipeline.
  7. Test it on paper first. A small spreadsheet of the loop shows whether the fix damps the swing before anyone changes a rule.

Where it goes wrong

The swing gets blamed on people. Managers are told to be more careful, and the same structure produces the same swings with more anxious managers. Draw the loop and its delays before looking for a culprit.

The fix is a harder response. Leaders see the shortfall and tell everyone to react faster and more strongly, which, with the delay still in place, makes the overshoot worse. Slow and gentle responses damp a delayed loop; harsh ones feed it.

Only one delay gets counted. The delivery lag is measured and the counting and approval delays are not, so the fix shortens the wrong one. List every step between the gap appearing and the result landing.

When to use something else

  • Stock-and-Flow Models. When the job is building the quantitative model of the stocks and flows, not just finding the delays.
  • Causal Loop Diagram. When the loops themselves are not yet clear and need drawing before any delay can be placed on them.
  • Control Chart. When the question is whether a swing is a real signal or ordinary variation that needs no response.

Where it sits in SOLVE eX

Go to the source

  • Industrial Dynamics, 1961
  • Modeling Managerial Behavior: Misperceptions of Feedback in a Dynamic Decision Making Experiment, Management Science 35, 1989
  • Thinking in Systems, 2008

Back to the catalog