Chapter 4 of 4

Last-Mile Service, Routing, and Performance

3 sections · ~6 min read · 3 practice questions · part of Logistics & Transportation

Learning objectives

  • Explain why last-mile is expensive in absolute and relative terms
  • Estimate route time, route cost, and cost per stop from a sequenced route, and identify what actually drives cost per stop
  • Choose a last-mile operating model that fits demand density and service promise

Why last-mile is structurally expensive

Last-mile delivery is the final leg from a depot or local facility to the end customer, and it is the most expensive per unit distance in the entire chain for two structural reasons. First, utilization is low: each vehicle typically carries only a handful of orders per route, so the fixed cost of the driver, vehicle, and insurance is spread over few stops. Second, density is uneven: urban routes have many stops per mile but suffer from congestion and access constraints, while suburban and rural routes have long distances between stops and high drive time. The result is that last-mile cost per stop ranges from a few dollars in dense urban routes to twenty or more dollars in sparse rural routes. The implication is that any model that promises flat-rate next-day delivery everywhere must be subsidized somewhere, usually by the dense urban customers.

Routing heuristics and their limits

The Traveling Salesman Problem (find the shortest round trip visiting each stop once) is NP-hard. Exact solution time depends on the instance and solver; size alone does not give a fixed cutoff for when heuristics are needed. In practice, planners use heuristics: nearest neighbor (always go to the closest unvisited stop), sweep (fan out from a depot in angular order), Clarke-Wright savings (merge two routes whenever the saving in distance exceeds a threshold), and local search (2-opt and 3-opt swaps). In practice a savings-style construction followed by local-search improvement lands close enough to the optimum that the remaining gap is smaller than the error in the travel-time estimates feeding the model, and it does so fast enough to re-run whenever the order book changes. That last point matters more than the gap: a route plan that can be rebuilt in seconds when a large order lands is worth more than a slightly shorter plan computed overnight. Time windows, vehicle capacity, driver hours, and access constraints add complexity that commercial routing software handles. The planner's job is to set the right inputs (travel time matrix, service time per stop, vehicle capacity) and to review exceptions rather than to build routes by hand.

Choosing a last-mile operating model

Four operating models dominate. In-house fleet works for high-volume, dense operations where the firm has the scale to keep utilization high. Contracted third-party works for medium density where the carrier can aggregate volume across shippers to keep trucks full. Crowdsourced delivery (gig drivers) works for same-day in urban areas but has limited capacity and quality variability. Pickup points and locker networks work for low-urgency, high-density residential delivery and dramatically cut last-mile cost per stop at the price of customer convenience. The right model is often a mix: in-house for next-day business deliveries, third-party for standard residential, lockers for low-priority items, and crowdsourced for surge. Each model has a different cost-per-stop signature, and the choice should be revisited when order profile or density changes.

Worked example

Problem

A delivery van leaves a depot at 8:00 AM with 8 stops. A routing pass has already sequenced the stops, and on the resulting route the drive time between consecutive stops averages 8 minutes while service time at each stop averages 6 minutes. Ignoring the return to depot, compute total route time and route cost at $1.50 per minute of loaded operating cost (driver, vehicle, fuel, insurance). Then compute the same figures for a 12-stop route in the same territory, first assuming average drive time stays at 8 minutes, and then assuming it rises to 9.5 minutes because the extra stops are drawn from a wider area.

Step by step

  1. Count the drive segments: depot to the first stop, then 7 legs between the 8 stops, so 8 segments in total.
  2. Route time = drive + service = (8 x 8) + (8 x 6) = 64 + 48 = 112 minutes.
  3. Route cost = 112 x $1.50 = $168.00. Cost per stop = $168.00 / 8 = $21.00.
  4. 12-stop route holding drive time at 8 minutes: 12 drive segments and 12 service stops give (12 x 8) + (12 x 6) = 96 + 72 = 168 minutes. Cost = 168 x $1.50 = $252.00, again $21.00 per stop. Under a constant average drive time, cost per stop is flat by construction, which is exactly why this version of the model proves nothing.
  5. 12-stop route with drive time at 9.5 minutes: (12 x 9.5) + 72 = 114 + 72 = 186 minutes. Cost = 186 x $1.50 = $279.00. Cost per stop = $279.00 / 12 = $23.25.
  6. The comparison isolates the real driver. Adding stops does not raise cost per stop on its own; it raises cost per stop only to the extent that reaching those stops lengthens the average leg. Note the damping: an 18.75 percent increase in average drive time (8 to 9.5 minutes) produced only a 10.7 percent increase in cost per stop, because the 6 minutes of service time per stop is unaffected and dilutes the change.

Answer. The 8-stop route takes 112 minutes and costs $168.00, or $21.00 per stop. Extending to 12 stops at unchanged density costs $252.00, still $21.00 per stop. Extending to 12 stops where the average leg stretches to 9.5 minutes costs $279.00, or $23.25 per stop, an 11 percent penalty. The lesson for a planner is that route density, measured as drive time per stop, is the variable to manage; stop count on its own tells you almost nothing about efficiency.

Practice

Work each question before opening the solution.

  1. Why does a flat-rate next-day shipping promise usually require cross-subsidization across customer segments?

    Show solution for question 1

    Per-stop cost varies enormously by density and distance. Charging all customers the same flat rate means dense urban customers (cheap to serve) overpay and subsidize rural customers (expensive to serve). The subsidy is sustainable while urban volume is large; if rural volume grows, the carrier must either raise the flat rate or tier the service.

  2. A firm considers replacing 30 percent of residential stops with locker drop-off. Each locker stop saves 4 minutes of drive and service time versus door delivery. Estimate the cost saving on the 8-stop example above if 30 percent of stops (2.4 stops) move to lockers.

    Show solution for question 2

    Time saved = 2.4 stops x 4 min = 9.6 minutes. New time = 112 - 9.6 = 102.4 minutes. New cost = 102.4 x $1.50 = $153.60, savings of $14.40 or 8.6 percent. Real savings would also include reduced failed-delivery attempts and lower driver fatigue.

  3. When is a crowdsourced delivery model likely to fail in quality terms?

    Show solution for question 3

    When the goods are high-value, fragile, or time-critical and the gig driver has no accountability beyond the platform rating. Quality control, exception handling, and customer communication are weaker than with a contracted fleet, so the model works best for low-value, low-urgency, high-density orders.