Stage 06 · Sales: Variance Is the Tax You Never Budgeted For

The sixth stage on Ariadne's map is "a repeatable process" — a pipeline someone other than the founder can run. Owners often hear that as bureaucracy. It is not. It is the cheapest available lever on speed, and the reason is a term in an equation.

The V in VUT

Recall Kingman's heavy-traffic approximation for queue time Kingman 1961, renamed the VUT equation in the Factory Physics tradition Hopp & Spearman 2004:

The previous post in this series concentrated on $U$. Here I want the first term, because it is the one you can actually move without spending money.

Note that it is squared coefficients of variation. A process whose handling time has a coefficient of variation of 1.5 contributes $1.5^2 = 2.25$; halving that CV to 0.75 contributes 0.5625. The V term falls by a factor of four. Delay falls with it — with no change in capacity, headcount, or utilisation.

Same people, same average speed, same workload. Five times the wait.

Where sales variance comes from

In an owner-led firm, handling-time variance has predictable sources, and none of them are about talent:

Undefined qualification. Some inquiries get a 5-minute triage; others get a 40-minute exploratory call before anyone establishes there is no budget. That single ambiguity can dominate $c_s^2$.

Improvised proposals. A proposal written from a template takes an hour. A proposal written from scratch takes a day, sometimes three, and its arrival date is unpredictable to the customer as well as to you.

Founder-dependent steps. Any step only one person can perform inherits that person's calendar variance. This is the single most common structural defect I find, and it is invisible on an org chart.

Batch behaviour. Doing all follow-ups on Friday raises $c_a^2$ downstream. Batching converts a smooth arrival stream into a spiky one, and spikiness is exactly what the V term punishes.

Which variation deserves a response

Reducing variance requires knowing which variation means anything. This is Shewhart's contribution: separate assignable-cause variation, which has a specific locatable source, from chance-cause variation, which is inherent to the system as designed Shewhart 1930. The control chart exists to tell them apart.

Deming built a management philosophy on the consequence Deming 1982/2000. Common-cause variation is the responsibility of the system and therefore of management; special-cause variation is local and assignable. Responding to common-cause noise as though it were a special cause is tampering — and tampering demonstrably increases variance.

Concretely: your close rate this month was 22% against a long-run average of 31%. Is that a signal? With ten deals, almost certainly not — that difference is well inside what chance produces. Rebuilding the pitch in response is tampering. The correct response to common-cause variation is to change the system deliberately or leave it alone, not to react to the last data point.

That second figure is worth dwelling on. In a full year of real call-centre data, service-time standard deviation (248 s) exceeded the mean (201 s) Brown et al. 2005. Service work is intrinsically high-variance. You will not eliminate it — you are trying to move from unmanaged to managed.

What standardising actually means

Not scripts. Not removing judgement. Four specific things:

1. Fixed decision points. The stages a deal can be in, and the written criterion for moving between them. Ambiguity about state is the largest single source of handling-time variance.
2. Templates for artifacts. Proposal, scope, follow-up. The thinking stays bespoke; the assembly does not.
3. Time-boxed steps. A triage call is 15 minutes. A discovery call is 45. When a step reliably overruns, that is assignable-cause variation and it deserves investigation.
4. One owner per step. Not for accountability theatre — because unowned steps have unbounded duration.

The test of whether you have a process is simple and slightly uncomfortable: can someone who is not you run a deal from inquiry to signature using only what is written down? If not, your $c_s^2$ is a function of your calendar, and every queueing result in this series says you will pay for that in delay.

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.1030.0028
3. 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
4. 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/
5. Brown, L. D., Gans, N., Mandelbaum, A., Sakov, A., Shen, H., Zeltyn, S., & Zhao, L. (2005). Statistical analysis of a telephone call center: A queueing-science perspective. Journal of the American Statistical Association, 100(469), 36–50. https://doi.org/10.1198/016214504000001808