Delivery cost will not collapse into one price

A price on a screen is not a cost structure. Delivery cost sits in the vehicle, the energy, the people, the node, the insurance, the failed attempt, and the record. If a yen figure appears here, it is an assumption, shown so the division is visible. Do not call it a measured market price.

Seven layers land on different steps

The first layer is the vehicle or the airframe: capital recovery, the purchase spread over residual value and the years or hours you can use. It is close to fixed for a day or a year, and it falls per success as successes rise. A failed sortie still used the asset. The second layer is energy, varying with distance, weight, hover, and wind. Aside from storage losses while parked, it is nearly all variable. The third is labor: driving, remote piloting, loading, dispatch, and the people who answer questions. A quiet hour can still contain a person on a shift, so per-sortie cost and per-day cost are mixed. Crush these three into one "operating cost" and you can no longer see which lever you pulled.

The fourth layer is the node. Rent, a nest or a place to land vertically, chargers, and the forward warehouse that short range demands. It is fixed for the day. A 2018 Nature Communications analysis argued that the multicopters it modeled were short-legged enough to require more urban warehouses or waystations. A drone may add nodes near the customer rather than delete them, so do not assume the node line is a saving. The fifth layer is insurance, a price on hull, liability, and cargo that combines expected loss with uncertainty. A new operation with a thin history pays for ambiguity, not only for the average accident rate.

The sixth layer is failure and redelivery. A failed attempt consumes a sortie, often creates a second one, and adds a contact to support. Japan's ministry treats redelivery as a problem of cost, labor, and emissions because this layer is not a rounding error. The seventh layer is the system: dispatch software, fees for traffic management, communications, maintenance records, and compliance paper. High volume makes it small per package. Low volume makes it decisive. Linehaul carries vehicle, energy, crew, and insurance. The sort carries the node and handling labor. The last drop carries failure. The exception carries support. Read the layers against those steps, or the stack is only a list.

Fixed cost is divided by the successes

Cost per successful delivery has a simple shape. Divide the fixed cost of the period by the number of successes in that period. Add the variable cost of the attempts required to produce one success. A layer that arrives monthly, such as the node, sits in the first term. A layer that arrives with each attempt, such as electricity, sits in the second. More failure inflates the second term, because attempts multiply. Fewer successes inflate the first term. The same bad weather can worsen both at once. A day you cannot fly does not turn the rent off.

Utilization is the hinge of every fixed layer. Halve the hours flown in a year and the airframe's capital recovery nearly doubles per success. The word "nearly" is there because some maintenance and some residual value move with flying hours. It is not a market measurement. Energy behaves the other way: a day you do not fly, you mostly do not burn it. Labor sits in between. Hired as a shift, it is close to fixed. Paid by the piece, it is close to variable. Mis-classify a layer and a plan to fly more will stack variable cost instead of spreading fixed cost.

The World Bank's logistics indicators try to see time lost at transshipment and borders, not only time in motion. UNCTAD tracks ports and freight rates as separate subjects for the same reason. A large share of cost can sit in the connection rather than in the cruise. When a drone estimate is detailed only about cruise electricity, look first at whether the node, the failure, and the insurance are blank. A blank layer is not zero yen. It means you have not placed it yet, so the total is not something to believe.

If you write a yen figure, call it an assumption

Here is an assumption, not a measurement. Suppose one nest is allocated 800,000 yen a month for rent, power, and insurance. Suppose it completes 520 successes that month. This layer alone is about 1,540 yen per success, because 800,000 divided by 520 is about 1,538. If the same nest completes only 200 successes, the figure is 4,000 yen. The yen moved because the denominator moved, not because prices did. Do not set those two figures beside each other as if they were catalog prices. The only respectable comparison is the sensitivity of the denominator.

A second assumption, still not a measurement. Suppose one flight uses 150 yen of electricity. Suppose a success takes 1.25 attempts on average. Electricity is then about 190 yen per success, which is only 150 times 1.25. It is not a battery price list and not a utility statistic. The number exists to show that this layer's width is set by whether attempts per success sit near 1.0 or near 2. Do not add it to the nest assumption and write "delivery costs this much." The layers differ, and so does the way each assumption was posed.

Stating an assumption is the opposite of hiding one. In the same place, name the layer, say whether it is fixed or variable, define the success in the denominator, and say why the yen figure was chosen. Drop the second trip of a redelivery out of that definition and the cost falls on the page. That is not an improvement. It is an edit to the denominator. Keep assumption-yen in a different shelf from figures a ministry or a paper actually measured. The moment they mix, this guide stops being useful. Read my yen figures with that same partition.