Consistency Is the Product: Why Spread Beats Average | HL Hunt
Consistency Is the Product: Why Spread Beats Average
Quality is discussed as a level and it is a distribution. A firm has an average and a spread, and customers experience one draw at a time with no distribution to place it in. Below, the firm that produced better work more often keeps less than half the customer value: $5,533 against $12,000, on a mean difference of 0.7. Retention is governed by the worst draw, not the average one, and firms optimize the average because the average is the part they can measure.
What you'll learn
The core thesis
Firms compete on quality, and quality is discussed as a level. It is a distribution. A firm has an average and a spread, and customers experience one draw from it at a time.
Our position is that spread matters more than average for anything a customer buys repeatedly, and that firms optimize the average because the average is what they can see.
Per our reference point analysis, a customer has one observation of your work and no distribution to place it in, which means a bad draw is indistinguishable from a bad firm. The firm knows it was unlucky. The customer does not and has no reason to assume it.
Two firms, scored out of ten
Take ten engagements each:
| Firm | Mean | Standard deviation | Worst result |
|---|---|---|---|
| A: 9, 9, 8, 9, 9, 8, 9, 9, 9, 8 | 8.7 | 0.46 | 8 |
| B: 10, 6, 10, 5, 10, 9, 4, 10, 10, 6 | 8.0 | 2.32 | 4 |
Firm B produces the best work of the two. It reached ten five times and A never did. Its average is lower and not by much.
Now set an acceptability threshold at 7, below which a customer is disappointed enough not to return:
- Firm A: 0 of 10 engagements below threshold, so 0% chance a given job disappoints
- Firm B: 4 of 10, so 40%
At $2,400 per engagement over five periods, with a disappointing job ending the relationship:
That is a 2.2 times difference produced entirely by spread. Per our variance analysis, the general principle is that a party exposed to variance pays for it, and here the customer is exposed to delivery variance and responds by leaving, which transfers the cost straight back.
And the asymmetry is sharper than the average suggests. Per our correlation analysis, the tens firm B produced bought it nothing durable, because a customer cannot bank a good experience against a future bad one. The fours ended relationships. Upside does not accumulate and downside does.
Why firms optimize the wrong moment
- The average is measurable and the spread is not. Satisfaction scores are reported as means, and per our reference point analysis a mean conceals exactly the information that predicts churn.
- Good work is visible and consistency is invisible. Nobody compliments a firm for a job being unremarkable, which is what consistency produces.
- The best work gets promoted. Case studies and references are drawn from the top of the distribution, which is the part that does not describe what a customer will receive.
- Peak capability is what firms hire and train for. Per our coordination analysis, the capability exists and whether it is reliably applied is a coordination question rather than a skill question.
- Variance is attributed to circumstances. Each bad job has a specific explanation, so nothing aggregates into a pattern.
Item five is the mechanism that keeps the problem invisible. A firm reviewing a poor outcome finds a cause: an unusual request, a supplier failure, a staff absence. All true, all individually non-recurring, and the aggregate is a 40% disappointment rate that nobody has ever stated as a number.
A falsifiable claim
Repeat purchase should correlate with the minimum of a customer's experience scores rather than with the mean. Any firm holding per-engagement satisfaction data and retention data can test it: regress repeat purchase on the mean score and on the minimum score. If the mean explains retention and the minimum adds nothing, quality level is what matters and this argument is wrong. We expect the minimum to dominate, which would mean a firm improving its average by raising its best work is improving the variable that does not drive retention while leaving untouched the one that does.
When the average is the right target
The argument has real limits and they are not edge cases.
- One-off purchases. If the customer buys once, there is no retention to protect and peak capability wins the sale.
- Winner-take-all selection. Where the work is awarded by competition on the best submission, the top of your distribution is what is being judged.
- Markets that pay for the ceiling. Some work is bought precisely because someone can occasionally do something nobody else can, and consistency is not the product.
- Reducing variance can reduce the mean. Standardization that removes the fours frequently also removes the tens, and per our rationing analysis that is a genuine trade rather than a free improvement.
- Some variance is the customer. Per our legibility analysis, an unusual engagement produces an unusual result, and attributing all spread to the firm overstates what it controls.
The fourth bullet is the serious objection. A firm that tightens its distribution by standardizing has usually bought consistency by giving up its upside, and whether that is right depends on whether its market pays for the ceiling or for the floor. The claim here is not that tight is always better. It is that firms rarely know which they are selling and default to optimizing the ceiling.
Where the spread actually comes from
Variance in delivery is not randomly distributed across causes. It concentrates.
| Source | Share of variance it typically explains |
|---|---|
| Who did the work | Large, and measurable per our quality monitoring analysis |
| Whether the job was unusual | Large, and partly the customer |
| Load at the time | Large, per our capacity analysis, and entirely within your control |
| Handoffs involved | Moderate, per our coordination analysis |
| Whether anything was written down | Moderate, per our institutional memory analysis |
| Genuine randomness | Smaller than firms assume |
The third row is the one most firms could fix this quarter. Delivery quality falls when the firm is full, which means accepting the marginal job degrades every other job in progress, and per our capacity analysis that cost never appears in the decision to accept it. A firm with a 40% disappointment rate frequently has a 10% disappointment rate when not overloaded and a 70% rate when it is.
The last row deserves attention because it is where firms put most of the explanation. If genuine randomness were the main source, variance would be irreducible and nothing in this article would be actionable. The reason the first three rows matter is that they are the opposite: identifiable, attributable, and addressable without changing anyone's skill.
Scenarios and what we are watching
| Scenario | What we would expect |
|---|---|
| Review platforms become more granular | Spread becomes visible to customers for the first time, and tight firms gain against firms with higher averages |
| Firms start reporting minimums | The measure that predicts retention enters management reporting, which is the single change that would make this actionable |
| Capacity pressure rises | Spread widens industry-wide, and gets attributed to labour quality rather than to load |
| Nothing changes | Firms keep competing on their best work and keep losing customers on their worst, with no reporting line connecting the two |
The first row is the development that would change behaviour fastest. A rating system showing an average rewards the firm with the higher mean. One showing the distribution, or the worst recent outcomes, rewards the firm with the tighter spread, and per our measurement analysis firms optimize whatever is displayed. The current arrangement displays the mean, so firms manage the mean, which is a reasonable response to the measurement rather than a failure of judgment.
Frequently asked questions
Because retention follows the worst experience. Firm B did better work more often and kept $5,533 of five-period value against $12,000 for the tighter firm.
Because the mean is measured and the spread is not. Each bad job has its own explanation, so nothing aggregates into a rate anyone states.
No. Tightening the spread often costs the ceiling, which is wrong if your market pays for peak capability rather than reliability.
Load. Quality falls when the firm is full, so the marginal job degrades every job in progress, and that cost never enters the acceptance decision.
Key takeaways
- Quality is a distribution, and customers experience one draw from it with nothing to compare it against.
- A firm with a 0.46 spread kept $12,000 of five-period value against $5,533 for one with a 2.32 spread and a mean only 0.7 lower.
- Upside does not accumulate and downside does: a customer cannot bank a good experience against a future bad one.
- Firms optimize the average because satisfaction is reported as a mean, which conceals what predicts churn.
- Each bad outcome has an individually non-recurring explanation, so a 40% disappointment rate never gets stated as a number.
- Tightening the spread often costs the ceiling, which is the wrong trade in markets that pay for peak capability.
- Load is the largest controllable source of variance, so the marginal accepted job degrades every job in progress.
- Genuine randomness explains less of the spread than firms assume, which is what makes it addressable.
This analysis is educational and does not constitute business or management advice. All figures are stylized illustrations using assumed satisfaction scores, acceptability thresholds, engagement values, and retention behaviour, and do not describe any particular firm. The relationship between individual experiences and repeat purchase varies substantially by market, product, and switching cost.