Nine Stages, One Constraint

There is a specific conversation I have had perhaps thirty times. The owner says traffic is down, so they want more traffic. I ask what happens to an inquiry that arrives at 4pm on a Thursday. There is a pause. Then: "It depends who's around."

That pause is the constraint. More traffic into a system that loses inquiries at the routing stage produces more lost inquiries, at higher cost per unit.

Throughput is set by one stage

Goldratt's Theory of Constraints states the case in its bluntest form: in any system pursuing a goal, throughput is limited by a single constraint at a time, and the correct sequence is to identify the constraint, exploit it (get everything you can from it without spending), subordinate everything else to it, elevate it (spend to increase it), and then return to step one — explicitly refusing to let inertia become the new constraint Goldratt & Cox 1984.

Adding demand to this system raises the first number and changes nothing downstream. Every unit above six per day becomes queue. Queue is not revenue; it is decayed revenue, and the next post in this series will put a number on how fast it decays.

Why the wall arrives suddenly

The reason capacity problems feel like they appear overnight is mathematical, and it is worth internalising. For a single-server queue, Kingman's heavy-traffic approximation gives expected time in queue as a product of three terms Kingman 1961:

Here $\rho$ is utilization, $c_a$ and $c_s$ are the coefficients of variation of interarrival and service times, and $\tau$ is mean service time. Hopp and Spearman rename this the VUT equation — Variability × Utilization × Time — and build an entire discipline on it Hopp & Spearman 2004.

Look at the middle term. At 50% utilization it equals 1. At 80% it equals 4. At 90% it equals 9. At 95% it equals 19.

A solo operator who was comfortable at 70% is not slightly less comfortable at 90%. Their queue time has roughly quadrupled. Nothing about them changed. The curve did.

Variability is a lever, and it is usually the cheaper one

The first term of the VUT equation is the one most businesses never touch. Delay is proportional to the sum of squared coefficients of variation — of arrivals and of service. That means erratic work is expensive even when average capacity is adequate.

This is where Shewhart's distinction becomes operationally useful. Statistical process control separates assignable-cause variation from chance-cause variation Shewhart 1930, a distinction Deming later built a management philosophy on: reacting to ordinary noise as though it were a signal is tampering, and tampering increases variance Deming 1982/2000. Through the V term, higher variance means longer queues — so the well-meaning act of responding to every wobble makes the system slower.

The constraint moves, and that is the trap

The fifth focusing step exists because of a failure mode Goldratt saw repeatedly: a team fixes the constraint, then keeps optimising the same stage out of habit, long after the binding constraint has moved elsewhere. Kingman's equation tells you where it moved — to whichever stage now has the highest $\rho$ and the fattest variability.

This is the argument for one map, maintained. Not because maps are tidy, but because the constraint is a moving target and local optimisation is worse than useless once it has moved. It consumes the exact attention the new constraint needs.

References

1. Kingman, J. F. C. (1961). The single server queue in heavy traffic. Mathematical Proceedings of the Cambridge Philosophical Society, 57(4), 902–904. https://doi.org/10.1017/S0305004100036094
2. Hopp, W. J., & Spearman, M. L. (2004). To pull or not to pull: What is the question?. Manufacturing & Service Operations Management, 6(2), 133–148. https://doi.org/10.1287/msom.1030.0028Peer-reviewed statement of the Factory Physics thesis; the VUT form appears in the authors' textbook of the same name.
3. Goldratt, E. M., & Cox, J. (1984). The Goal: A Process of Ongoing Improvement. North River Press. https://northriverpress.com/the-goal-30th-anniversary-edition/A business novel containing no data; cited for the five focusing steps, not as evidence.
4. Mabin, V. J., & Balderstone, S. J. (2003). The performance of the theory of constraints methodology: Analysis and discussion of successful TOC applications. International Journal of Operations & Production Management, 23(6), 568–595. https://doi.org/10.1108/01443570310476636Meta-analysis of 80+ applications; found no reported failures, which is itself a publication-bias signal.
5. Shewhart, W. A. (1930). Economic quality control of manufactured product. Bell System Technical Journal, 9(2), 364–389. https://doi.org/10.1002/j.1538-7305.1930.tb00373.x
6. Deming, W. E. (1982). Out of the Crisis. MIT Center for Advanced Engineering Study; MIT Press editions 2000, 2018. https://mitpress.mit.edu/9780262535946/out-of-the-crisis/