Failure raises the cost of every success

Failure is not a moral deduction. Absence, misdelivery, a weather cancellation, damage, and the complaint that follows raise the cost of one success. The rise is division by the success rate. Failures that arrive together cost more than failures that arrive apart. Do not mix a measured rate with an arithmetic illustration.

Absence gives birth to another attempt

Absence means the parcel arrived and nobody received it, so a second movement is born. The ministry's explanatory page, citing its sample surveys, puts redelivery at about 10.4 percent in the April 2024 round and about 10.2 percent in October 2024. It equates a share on the order of one tenth with about 60,000 drivers' labor for a year, and it cites about 254,000 tonnes of carbon dioxide as the ministry's estimate for fiscal 2020. A 2022 ministry survey on that page found that about one fifth of the reasons were that the receiver did not know a delivery was coming. Read all of these with their dates attached.

The rate moved. A ministry release dated 10 July 2026 reports that, in the April 2026 sample of major carriers, the redelivery rate was about 7.6 percent, 0.7 points under the October 2025 sample. Because the series moves, do not freeze an older rate as the rate now. Rescaling the 60,000-driver figure by 7.6 percent yourself, and calling the result a new official headcount, is equally careless. Leave that labor conversion inside the page that assumed a share near one tenth. Cite the newer rate as the newer rate.

Put the other failures in the same column. Misdelivery creates a retrieval, a correct second attempt, and a complaint. Damage creates repacking, a reshipment, and sometimes a consequential loss. Insurance prices the pattern. A complaint consumes the time of someone whose job is to repair trust, not to move the box. For an aircraft, a day you cannot fly is not a cheap pause. The node and the shift are still paid. Just as a van's absence creates another run, a cancellation cuts the day's denominator of successes in one stroke. The next section separates that cut as a matter of rates.

Divide by the success rate and the attempt grows dear

Let c be the cost of one attempt and p the probability that the attempt succeeds. If failures are retried and the trials are independent, the expected number of attempts per completed delivery is 1 divided by p. Attempt-cost per success is c divided by p. This is arithmetic close to a definition. It is not a market measurement. As an illustration, if p is 0.95 then attempts per success are about 1.05. If p is 0.70 they are about 1.43. Attempt-cost rises by about 36 percent, before claims, refunds, and customers who leave. That 36 percent is only the ratio of 1.43 to 1.05, minus one.

Not retrying is also a choice. You save the second sortie and you swallow the refund and the lost sale. The minimum is not "always retry." It is the smaller of the retry and the cost of giving up. The cost of giving up includes reputation you will not see on a single invoice. An assumption that sets reputation at zero has to be labeled as an assumption. Setting it at zero without saying so is concealment. A business whose customer leaves after one failure, and a business whose customer returns next week, face different costs of giving up at the same p.

The design this arithmetic suggests is that raising p can be cheaper than raising cruise speed. A confirmed receiver, a landing site that exists, a weather gate that cancels early instead of launching into an abort: these are less glamorous than speed, and they shrink 1 divided by p directly. If speed lowers p, cost per success can rise. A fast failure can cost more than a slow success. That is not a slogan. It is the direction of the division. Judge it only after you have set your own c and p down as assumptions.

Failures that arrive together strand the fixed cost

Independent failures, one bad address or one aircraft with a fault, can be partly averaged across a fleet. Failures that arrive together cannot. A windy afternoon, a closed airspace, or a dark depot stops many sorties at once. Fixed cost is stranded. The next flyable window inherits the backlog, which brings overtime, penalties for lateness, and a second wave of absence because the people who waited are no longer there. Crush independent and correlated failure into one word, "failure rate," and you will stock the reserve in the wrong place.

The sample surveys are a picture of absence on the ground. One household is close to independent. A commuter block empty at the same hour is close to correlated. Aircraft add a correlation vans feel less, because a van can often still creep when a nest is grounded. Do not import a van's completion rate into an aircraft plan without asking what stops the whole nest. What you may import is the structure, that absence creates a second attempt. You may not import the percentage itself.

The 2018 Nature Communications paper is a reminder that the small multicopters it modeled were short-range, so a missed drop is rarely a small detour on a long tour. It is often the whole sortie. Set that beside 1 divided by p and a failed drop is expensive relative to the cruise you thought you were buying. Dropping the sample survey's 7.6 percent or 10 percent straight into that p is careless too. Those are van redelivery rates, not your aircraft's success rate. Set p aside as its own assumption. Then ask whether, on a correlated day, the denominator dies all at once.