Choosing Delivery Points for the Addresses They Add

A delivery point is valuable partly because of the addresses it makes reachable. That sounds straightforward until two attractive locations reach many of the same addresses. Ranking each location on its own can then produce a different answer from comparing complete pairs. The important distinction is between the reach of a site and the additional reach it contributes to the particular network being considered.

On 3 April 2026, Interfax reported that EFKO and HSE were developing robotic delivery in Russia. The proposed system included robots, servicing points and parcel lockers; a prototype was undergoing tests.

That proposal provides a reason to examine the geography of a delivery network, but it supplies no address-level location comparison. The example below is an original planning exercise, not a reconstruction of the companies' project. Its numbers, candidate locations and eligibility rules are invented. It asks a deliberately narrow question: if two sites can be selected, which pair includes the greatest number of distinct eligible addresses?

The illustrations show generic street and parcel-handling layouts, not actual project facilities. Their drawn blocks and markers do not encode the numerical example; the tables below define that example completely.

Choose the unit before counting the reach

An address is not an order. One eligible address might generate no purchases during a particular week, while another generates several. Nor is an address necessarily one person or household: a building may contain many customers. A comparison must decide what its entries represent before adding them. Here, each entry is simply one distinct delivery address, counted once whenever at least one selected site can serve it under the stipulated rules.

Eligibility is also narrower than a guarantee of service. It means the address passes the location test used in this exercise. The example assumes that all other requirements are satisfied and that capacity does not constrain the count. It does not estimate how many robots are required or how many parcels can be handled. Those questions would need another model and additional observations, rather than a new label attached to the same total.

These boundaries let the comparison isolate one decision. All candidate sites have the same assumed cost, and exactly two will be chosen. Customers covered by both sites remain one customer address in the network total. No extra weight is awarded for being reachable from two directions. This is a unique-address objective, not a resilience score, a revenue forecast or a measure of equipment utilisation.

A small network with three choices

Imagine six mutually exclusive zones. Zones one through four each contain ten eligible addresses; zones five and six each contain fifteen. Together they contain seventy addresses. A zone is only a convenient grouping of entries that share the same eligibility pattern. Its number gives no information about its size, distance from a site, population density or place on a real city map.

Three possible sites are available. Site A can serve zones one, two, three and four. Site B can serve zones one, two and five. Site C can serve zones three, four and six. The table states the complete input, so none of the results depends on an unseen assumption about travel speed or street layout. Actual robots and parcel lockers may have very different constraints; this exercise does not assign them these capabilities.

Evaluated individually, A looks strongest. Its forty addresses exceed the thirty-five available to either rival. If only one site could be chosen under these rules, A would be the answer. Nothing in the later comparison makes that single-site result incorrect. The decision changes when the organisation has permission and resources to choose two sites together.

Compare every possible pair

With A and B, the network reaches zones one through five. Zones one and two appear in both site lists, so their twenty addresses must not be counted twice. Adding the individual figures produces seventy-five, but removing the duplicated twenty leaves fifty-five distinct addresses. Zone six remains outside the selected pair. The total describes a union of address lists, not the sum of two independent customer populations.

A and C also reach fifty-five addresses. Their common zones are three and four, and their combined list leaves out zone five. B and C have a different relationship: none of their eligible zones overlaps. Their combined thirty-five plus thirty-five therefore reaches all seventy addresses. These are the only three possible pairs, and the complete comparison is consequently enough to identify the best pair for this particular exercise.

The best pair contains neither the highest-ranked standalone site nor an improvement to that site. It consists of the other two candidates. The conclusion is not that large sites are bad, or that overlap should always be eliminated. It is that the ranking of whole networks cannot in general be recovered by taking a ranking of individual sites and selecting its first two entries.

The next site's contribution depends on what is already chosen

Suppose A has already been selected. Adding B brings only fifteen new addresses: the ones in zone five. B's other twenty addresses were already reachable through A. Adding C instead also brings fifteen, from zone six. Neither addition is mismeasured by calling it fifteen. That figure answers the conditional question of what the site adds after A, whereas thirty-five describes what the same site reaches by itself.

Now begin with B. Adding C brings thirty-five new addresses, while adding A brings twenty. C has not changed location or capability between the two comparisons. The starting network has changed. This is why an expansion proposal should name its baseline. A statement that a site adds thirty-five addresses is incomplete unless it identifies the existing sites against which the addition was calculated.

Selecting the strongest standalone site first and then the largest remaining addition yields fifty-five in this example. Selecting the pair jointly yields seventy. That is a counterexample to treating the first procedure as automatically optimal; it is not proof that the procedure fails in every network. Some candidate lists would produce the same answer under both methods. For three sites, checking all pairs is simpler and more informative than assuming either outcome.

An existing commitment creates a different decision

A business with no committed location can compare all three pairs. A business that must retain A cannot choose B and C as its complete two-site network. Under that restriction, its feasible answers are A with B or A with C, both reaching fifty-five. Describing the latter business as making a mathematical mistake would confuse the decision it actually controls with a different decision imagined by an observer.

Retaining A may reflect a contract, an operational requirement or an earlier decision that is outside the present exercise. None of those circumstances is assumed to exist in the reported project. They illustrate why a planning brief needs to distinguish fixed sites from candidates. A comparison that silently treats a fixed location as removable may recommend an unavailable option, however accurately its address totals are calculated.

If replacement is genuinely available, it becomes another alternative with its own consequences. The seventy-address pair does not by itself establish that closing A is worthwhile. Moving, disruption and any changed obligations are outside this model. The useful first step is to show the coverage difference clearly, then assess the actual choices without pretending that a coverage table has already answered a financial or contractual question.

Overlap can serve a purpose that this score does not reward

The twenty addresses shared by A and B are duplicates only for the unique-address count. They could matter for another objective. If either site were unavailable, some addresses might remain reachable through the other. Whether that possibility constitutes dependable backup would require information about capacity, failure conditions and the ability to transfer work. Merely drawing two connections does not establish a tested continuity arrangement.

The example can nevertheless distinguish the underlying sets without inventing a reliability estimate. B and C maximise distinct eligible addresses, but none of those addresses has two eligible sites within the selected pair. A and B reach fewer distinct addresses, while twenty have two eligible sites. These are different descriptions of the same inputs. One is not an error that should be removed from the report to make the preferred option look stronger.

A decision-maker who values both reach and redundancy must state how those objectives will be compared. This article supplies no universal exchange rate between them. It also assigns no probability to a site outage. A transparent comparison can present unique coverage and duplicated eligibility separately, leaving the organisation to specify the additional requirements that a real service must meet.

Do not turn the address total into an operating forecast

Seventy eligible addresses do not imply seventy simultaneous deliveries. If each site has a limited number of available machines, the assignments may compete for capacity. A locker may also have compartments of different sizes or be unavailable at the desired time. Those practical issues can change the feasibility of a service plan without changing which addresses fall inside a static eligibility list.

Likewise, two sites that cover separate addresses may face different levels of actual demand. Giving every address one unit of weight is an explicit choice in the exercise, not a claim that all customers behave identically. A demand-weighted comparison would need a defined period and evidence for its weights. It should not quietly substitute hoped-for orders for the address counts while continuing to present the result as the same calculation.

Customer choice creates another distinction. An address eligible for two sites may prefer one of them, or may not use the service at all. Counting the address once prevents double counting; it does not predict which site receives an order. A location study can establish an available alternative while leaving adoption uncertain. Keeping those statements separate makes the result more useful than attaching an unsupported sales number to a geographical total.

Build the eligibility lists from consistent evidence

For a real comparison, each candidate needs an address list constructed under the same rules. Mixing one site's practical service area with another site's broad radius estimate would undermine the result before any arithmetic begins. The relevant rules depend on the proposed service and must be established by its operators. This article does not specify a safe robot route, an authorised operating area or a regulatory threshold.

Address identity deserves equal care. The same destination can appear with different spellings, building references or account details. Conversely, two distinct destinations can share a similar label. The comparison needs a consistent identifier for whatever delivery unit has been chosen. Otherwise, apparent additional coverage can arise from inconsistent records rather than from a location that genuinely reaches somewhere new.

• State the delivery unit that one row represents.
• Use the same eligibility conditions for every candidate.
• Identify existing sites that must remain in the network.
• Count each eligible destination once in the combined total.
• Report shared eligibility separately if redundancy matters.
• Keep demand, capacity and service commitments distinct from geographical reach.

The input should also have a common reference date. If one candidate is assessed before a route change and another afterwards, their lists may not describe simultaneous choices. Updating a constraint means reconsidering the affected sets, not simply increasing or decreasing a grand total. Retaining the address-level input makes it possible to explain which destinations entered or left the comparison and why.

Change an assumption openly rather than stretching the conclusion

Equal site cost makes the example a clean comparison of pairs. If B costs more than A, the seventy-address result remains an address count but no longer settles which pair fits a particular budget. A new comparison must use the actual permitted combinations. Similarly, allowing three sites would make all seventy addresses eligible, but A would add no new addresses after B and C under the present matrix.

Changing the coverage objective can also change the answer. An organisation might require a particular zone to be included, insist on two eligible sites for selected destinations, or assess an expansion from a fixed base. Each requirement narrows or reshapes the choice. It should appear in the question before alternatives are ranked, so that the eventual recommendation answers the organisation's decision rather than the easiest metric to maximise.

The reported robotic-delivery proposal is a system idea, not evidence that this invented pair of sites should be built. Its broader lesson for location analysis is precise: count the addresses a candidate adds to a named network, and compare complete feasible combinations when the decision concerns a combination. A prominent standalone total can be informative without being the right basis for choosing the whole network.