Cost Per Drop Under Pressure: How Peak-Season Density Changes Your Last-Mile Math in 2026

What cost per drop actually measures

Cost per drop is the full cost of a delivery route divided by the deliveries it completes successfully. The important word is successfully, because an attempt that fails still consumes driver time, vehicle cost and route capacity while adding a re-attempt and often a support contact.

Broken into components, the cost of any route is the fixed cost of putting a driver and vehicle on the road for a shift, plus the variable cost of travel between stops, plus the service cost at each stop. Knowing which of those three responds to density is what makes peak planning tractable, and it is where most operations reason incorrectly.

The stakes justify the arithmetic. Capgemini Research Institute puts last-mile delivery at 41% to 53% of total logistics and shipping cost, with dense operations at the upper end of that range.

Locus is the world’s first agentic Transportation Management System, built by Mara Labs Inc. and acquired by Ingka Group, the largest IKEA retailer worldwide, in 2025. Locus has supported 1.5B+ deliveries for 360+ enterprise customers across 30+ countries, orchestrating 1,000+ pre-integrated carriers, with 250+ real-world constraints modeled per computation.

Why density changes the math, and how far it can

Density affects the three cost components very differently.

Fixed route cost amortizes across more stops. If a driver completes more deliveries in the same shift, the cost of putting them on the road spreads further. This is the largest available gain.

Travel cost per drop falls. More orders in the same service area means shorter distances between consecutive stops, and often several deliveries reachable from a single parking position.

Service cost per drop does not fall at all. Time at the door is set by handover, building access, signature or age-verification requirements, and recipient availability. Doubling density does not make a controlled-entry apartment building faster to get into or a recipient quicker to answer.

That third point is the ceiling, and in dense North American metros it binds hard, because access rather than distance dominates the stop. Urban Freight Lab research instrumenting more than 1,800 deliveries in Seattle found urban commercial vehicles spend roughly 80% of daily operating time parked, with most of a driver’s time spent outside the vehicle walking the final stretch.

Congestion compounds it during the ramp. INRIX found US drivers lost 49 hours to congestion in 2025 at a cost of $85.8 billion in lost time, with congestion increasing in 88% of the 290 cities analyzed. Chicago drivers lost 112 hours and New York drivers 102.

The reciprocal case shows how powerful density is when it runs the other way. The US Postal Commission finds average cost per delivery in rural areas is approximately twice that in urban areas. That figure isolates density as the variable rather than anything about the goods or the operator, and it is the single most useful sourced anchor in this topic.

A worked example

The arithmetic below is an illustrative model with stated assumptions, not a benchmark or a claimed result. Replace the inputs with your own numbers; the mechanism is what matters.

Assumptions. One driver, one van, 480 productive minutes per shift. Fixed route cost expressed as 100 units, covering driver and vehicle. Baseline service time of 3 minutes per stop. Baseline travel time of 4 minutes between stops.

Scenario	Travel per stop	Service per stop	Minutes per drop	Stops per shift	Cost per successful drop
Baseline, normal season	4.0 min	3.0 min	7.0 min	68	1.47 units
Peak density, plan re-optimized	2.5 min	3.0 min	5.5 min	87	1.15 units
Peak density, fixed zones and static sequence	4.0 min	3.0 min	7.0 min	68, remainder overflows	1.47 units, plus overflow at marginal cost
Peak density, overloaded without re-optimization	4.0 min	2.5 min rushed	6.5 min	73 attempted, 65 completed	1.37 per attempt, 1.54 per success

Three conclusions follow, and they are the argument of this article.

Re-optimization captures roughly a fifth of cost per drop in this model. Cost falls from 1.47 to 1.15 units, around 22%, entirely because travel time per stop compressed while fixed cost spread across more stops. That is consistent in direction and magnitude with McKinsey’s estimate of 10% to 25% cost reduction for AI-driven, multi-constraint routing against a static daily plan, though the McKinsey range covers dynamic routing generally rather than peak density specifically.

Fixed zones forfeit the gain completely. In row three, density rose and nothing improved, because the sequence was not rebuilt to exploit the clustering. The extra volume became overflow handled by additional capacity at marginal cost, which is usually the most expensive capacity available.

Overloading inverts the outcome. In row four the driver attempts more stops by compressing service time, which is what drivers do under pressure. Eight attempts fail. Cost per attempt improves to 1.37, better than baseline, while cost per successful drop rises to 1.54, worse than baseline, and the failures generate re-attempts the following day. The productivity metric improved and the economics deteriorated.

Also Read: How AI Dispatch Reduces Cost-Per-Stop: A Benchmarking Guide for Last-Mile Operations

Peak density is uneven, and the average lies

The ramp through Black Friday, Cyber Monday and the December build does not raise density uniformly. It redistributes it.

Two figures frame the shape. ShipMatrix found parcel networks absorbing a 30% volume increase during peak compared with the rest of the year while holding 98% on-time performance. And US Census data shows Q4 ecommerce at 17.1% of total retail sales against a 14.7% average for the other three quarters, so the quarterly demand shift is narrower than the operational surge. Peak is concentrated rather than proportional.

Zone type	What happens to density during peak	Where the cost lever is	What to watch
Dense urban core	Rises sharply, often the largest increase	Clustering multiple stops per parking position, re-sequencing	Access and dwell becoming the binding constraint
Suburban single-family	Rises moderately with high absolute volume	Route compaction and stop sequencing	Longer walk distances at larger lot sizes
Mixed metro commercial and residential	Rises with time-window conflicts between segments	Separating commercial receiving hours from residential availability	Plans assuming uniform service time across both
Exurban and rural	Volume rises but density often does not	Consolidation, service frequency, pickup points	Attempting daily service where alternate-day is economic

A network-level cost-per-drop figure averages all four and describes none of them. Reporting it weekly during the ramp will show a modest improvement while concealing that two zone types got materially cheaper and two got worse.

The practical instruction is to segment cost per successful drop by density tier before the season starts, so the comparison exists. Without a segmented baseline, post-peak analysis cannot attribute anything. Geography makes this unavoidable: the US Census Bureau reports 80% of the US population living in urban areas at an average 2,553 people per square mile, with the remaining 20% spread across the rest of the country.

Dense urban cores: what actually works

Four tactics, in order of effect.

Cluster so one parking position serves several stops. Where 80% of operating time is spent parked, reducing parking events matters more than reducing distance. This requires the plan to know where parking is feasible, which is constraint data rather than routing cleverness.

Re-sequence during the shift, not only at build time. Every dense plan is wrong by mid-morning, and during peak it is wrong sooner because same-day and next-day volume enters after the plan was built. The metric to hold a vendor to is re-decisioning latency, the interval between a signal and a revised dispatched sequence.

Capture access knowledge once and reuse it. Controlled-entry procedures, package room locations, concierge hours, locker availability and buzzer codes are resolved by a driver on the first attempt and then usually lost. Storing them against the address is one of the highest-return, lowest-cost interventions available before a peak season.

Protect service time rather than compressing it. The worked example shows why. Compressed service time buys attempts and loses completions.

Exurban and rural routes: different math entirely

Density tactics do not transfer, because there is no density to exploit.

Three levers apply instead. Service frequency rather than daily coverage, where alternate-day or scheduled service to sparse clusters beats daily attempts at low stop counts. Consolidation into fewer, fuller routes, accepting longer runs in exchange for amortizing fixed cost across more stops. And pickup points, lockers or retail counters where the alternative is a long detour for a single low-value delivery.

The rural cost multiple is why these need separate treatment rather than a metro playbook at lower volume. If cost per delivery in rural areas runs at roughly twice urban levels, a 20% routing improvement on a rural route still leaves it more expensive than an unoptimized urban one, which makes the decision about service design rather than routing.

Also Read: The Hidden Cost Categories of Failed First Attempts in US Last-Mile Operations

Exploiting density without overloading drivers

The tension in peak planning is that the same density gain can be taken as lower cost or as higher throughput, and taking too much of the second destroys the first.

Three controls keep it on the right side.

Model realistic service times. A plan built on optimistic service times creates overload by construction, then records drivers as underperforming. McKinsey finds static planning models can leave as much as 60% of operating hours either understaffed or overstaffed, and understaffed hours are absorbed by whoever is on shift.

Distribute route difficulty. During peak the hardest territories get harder. If the same drivers repeatedly receive them, first-attempt performance falls in exactly the zones where volume is highest, and attrition follows at the worst possible moment of the year.

Re-sequence rather than extend. When a stop runs long, the correct response is re-planning the remainder, not adding time to the end of the driver’s day. The first protects cost per successful drop; the second protects the plan on paper.

Unproductive waiting is worth accounting for separately, because it is largely outside the driver’s control. ATRI found drivers detained at 39.3% of all stops in 2023, losing between 117 and 209 hours per year depending on sector. Any productivity metric that does not exclude that time is measuring facility behavior and attributing it to the driver.

Also Read: The Three-Workforce Fleet Reality: How Owned, 3PL, and Gig Drivers Actually Operate at Most Enterprises

What cannot be benchmarked here

This topic attracts more unsourced numbers than almost any other in logistics, so it is worth stating plainly what does not exist.

No research firm, consultancy, government statistics body or peer-reviewed source publishes credible figures for absolute cost per drop or cost per stop, first-attempt delivery failure rates, deliveries or stops per driver hour by sector, fleet utilization by vertical, or per-lever cost reductions attributable to specific routing techniques. Every circulating version traces to software vendors or aggregator content, including the widely quoted claims that a re-attempt costs two or three times a first delivery and the frequently cited dollar figure for a failed delivery attempt.

What is sourceable is the structure: the last-mile share of total cost, the rural-to-urban cost multiple, the share of operating time spent parked, the peak volume surge, and the range for dynamic versus static planning. Use those to frame the problem, then build your own numbers.

Three internal measurements replace the missing benchmarks. Cost per successful drop segmented by density tier. First-attempt completion segmented by address or building type. And stops per productive driver hour, excluding waiting time outside the driver’s control. All three are computable from records you already hold, and all three are what a post-peak review needs.

Three generations of last-mile planning

Static planning. Zones and sequences built in advance and executed. Density gains are forfeited because the plan cannot see them.

Analytics-assisted planning. Historical performance informs the next plan. Density gains are captured a cycle late.

Continuous orchestration. The plan re-optimizes against live conditions, so a density change is exploited within the day. Locus operates here through its SDEL architecture, Sense-Decide-Execute-Learn.

The generation determines whether peak density is an opportunity or a problem. Gartner finds 95% of supply chains must react quickly to change while only 7% can execute decisions in real time, which is the industry-level measure of how many operations can act on a density shift while it is happening.

How Locus handles density economics

Locus operates as the decisioning layer above the existing estate, with eight agents sharing one constraint model.

The Dispatch Agent plans and re-sequences continuously against live traffic, hub readiness and incoming orders across 250+ modeled constraints, which is what allows clustering to be exploited rather than merely observed. Constraint coverage is the mechanism: parking feasibility, access restrictions, service duration by location type, vehicle class limits and time windows all have to be representable for a dense plan to hold. The Capacity Agent forecasts demand by zone and time band and right-sizes the fleet, which is how overload is prevented rather than discovered. The Carrier Agent allocates across owned and contracted capacity where overflow is genuinely needed, so the marginal stop goes to the cheapest eligible option rather than the nearest available one. The Hub Agent coordinates outbound readiness so drivers are not waiting on loads during the hours when density is highest. The Customer Agent tracks each order against its promise and captures structured failure reasons at the point of failure, which is what makes first-attempt improvement possible.

Six governance mechanisms, Explainability, Traceability, Evaluation, Autonomy Levels, Execution Sandbox and Human-in-the-Loop, allow autonomy to be raised for the peak window by decision category and audited afterwards.

Deployment evidence

Capacity you already own but cannot see: a Fortune 50 parcel and logistics provider. This operator moves 1M+ freight shipments a year with a 4,500-strong driver pool split across roughly 1,500 captive and 3,000 third-party drivers. Captive shifts ran zone-based routing while third-party carriers needed tendering and on-demand assignment, and no single tool unified the pool.

Orchestrator and Dispatch agents took over pickup, transit and delivery decisioning against 250+ operational constraints, with Capacity and Carrier agents governing the full driver pool under one policy. Weekly execution rate climbed from 75% to 92% across 51 active service-center locations, and a single-site capacity analysis surfaced $565K in unused capacity that scaled to $14M+ annualized across 25 sites, at 99.99% platform uptime. Detail in the Fortune 50 parcel centralized dispatch case study.

Read the capacity figure through the lens of the worked example. That was not a saving created by optimization. It was capacity the operation already had and could not see against demand, including premium-tier service being given away on cheaper classes, which is precisely what fixed zones do to a density gain.

Fixed routing patterns replaced: a leading North American retailer. This retailer supplies a multi-hundred-store footprint through several distribution centers and a network of hubs, with a private fleet of several hundred trucks moving tens of thousands of deliveries a year alongside 3PL capacity. Routing followed fixed patterns while loads, appointments and freight bills were handled manually. Planning ran leg by leg rather than as one system, so trailers went out underfilled while return legs ran empty, with no way to match backhaul or maximize trailer and dock utilization.

On Locus, Dispatch agents run routing across DC, hub and last mile against 250+ operational constraints, while Capacity and Carrier agents plan loads and match backhaul. Results: 95%+ route compliance, 80%+ reduction in manual dispatch, 99%+ on-time store delivery with exceptions resolved in under two hours, and $1M+ in savings with break-even inside the first year. Detail in the multimodal logistics automation case study.

The underfilled outbound and empty return legs are the density argument at freight scale. Neither was a routing failure. Both were planning-in-isolation failures, visible only once one system held both legs.

Also Read: The Morning Plan Problem: Why US Last-Mile Networks Need Dynamic Resequencing

Analyst validation

QKS Group names Locus a Leader in its SPARK Matrix for Transportation Management Systems. G2 ranks Locus #1 for Route Planning software. Locus appears in the 2026 Gartner Hype Cycle across AI-powered logistics categories. ShipFlex is named a Representative Vendor in the 2026 Gartner Market Guide for Multicarrier Parcel Management Solutions. Gartner has recognized Locus for seven consecutive years. The full set is at Locus analyst recognition.

What to do before the ramp

Four actions, all completable from existing records.

• Segment your cost-per-successful-drop baseline by density tier now. Without it, nothing after the season is attributable.
• Audit whether your zones are fixed. If zone boundaries have not been redrawn in the last two quarters, density gains will be forfeited by construction.
• Capture access knowledge against addresses. Controlled-entry procedures, package room locations and concierge hours resolved by drivers should be stored and reused rather than rediscovered each attempt.
• Set realistic service times per location type. Optimistic service times manufacture overload, and overload converts density into failures.