Arrival Variability | Why Averages Hide Logistics Risk

Arrival variability is the degree to which actual logistics arrival times spread around the expected or typical arrival time.

In one line: the average tells you where the centre is; variability tells you how far reality wanders around it.

This is Article 20 in eduKateSG’s 100-article logistics authority build and completes Batch 05. The canonical parent is How Logistics Works. Article 17 mapped lead time, Article 18 isolated dwell, and Article 19 explained cut-off discontinuities. Arrival variability brings those clocks together into a risk question: how predictable is the result?

The Same Average Can Describe Two Different Worlds

Consider two routes whose average delivery time is three days.

• Route A: 2.8, 3.0, 3.1, 3.0, 3.1 days.
• Route B: 1.0, 1.5, 2.0, 4.5, 6.0 days.

The averages can look similar while the operating experience is entirely different.

Route A supports a narrow promise. Route B forces the receiver to protect against a much wider range of outcomes.

Variability Converts Time Into Inventory and Buffer

If replenishment always arrives close to three days, a receiver can plan tightly around that rhythm.

If arrival sometimes takes one day and sometimes six, the receiver needs more protection. That may appear as safety stock, longer order lead time, wider receiving windows, spare labour, contingency capacity or emergency transport options.

Variability therefore creates cost even when the average looks acceptable.

The Late Tail Matters More Than the Early Tail

Arriving much earlier than expected can create storage or scheduling problems. Arriving much later can stop production, miss a customer event, exhaust inventory or break a connection.

That asymmetry means logistics risk often concentrates in the right-hand tail of the arrival-time distribution: the small share of shipments that take much longer than normal.

Averages compress those shipments into the centre and can make the route appear healthier than the receiver experiences it.

Percentiles Can Be More Useful Than Averages

A percentile asks how much time is needed to cover a given share of shipments.

For example, a 95th-percentile arrival time describes a threshold within which roughly 95 per cent of observed shipments arrived.

That can be more useful for service design because it translates the distribution into a practical reliability question: how much time must we allow if we want almost all shipments to make the promise?

Median and Mean Tell Different Stories

The median is the middle observation. The mean is the arithmetic average.

A small number of very late shipments can pull the mean upward while the median remains stable. That gap is itself useful evidence that the distribution contains a long late tail.

Neither number is “the correct one” alone. They answer different questions.

Variability Is Built From Several Different Clocks

End-to-end arrival spread can come from many intervals:

• Warehouse release variability.
• Carrier pickup variability.
• Traffic and line-haul variation.
• Terminal dwell.
• Missed transfer connections.
• Border release time.
• Weather disruption.
• Final-mile route variation.
• Receiver availability.

The largest source is not necessarily the longest average interval. A short but highly variable handoff can dominate the uncertainty of the entire route.

Missed Cut-Offs Create Multi-Modal Distributions

Article 19 showed why missing a cut-off can add a whole service cycle.

That can produce clusters of arrivals rather than one smooth spread. Shipments that make the connection arrive around one time; shipments that miss it arrive around a much later time.

The distribution can therefore reveal hidden schedule structure. A second hump in arrival times may point to a missed-connection population rather than random noise.

Dwell Variability Often Drives Arrival Variability

A travel leg may be highly stable while terminal or border dwell varies widely.

The World Bank’s 2025 LPI 2.0 notes high unpredictability at ports, transshipment hubs and inland checkpoints. The WCO Time Release Study’s emphasis on time-interval analysis and outliers is valuable for the same reason: unstable waiting at nodes can dominate end-to-end arrival uncertainty.

This is why faster vehicles alone often fail to fix unreliable delivery.

Variability Propagates Across Handoffs

An inbound shipment that arrives thirty minutes late may still complete the route on time if the next connection has enough buffer.

The same thirty-minute delay can become a twelve-hour late arrival if it crosses a transfer cut-off.

Variability therefore interacts with schedule geometry. The same upstream spread creates very different downstream outcomes depending on buffers and connection frequency.

Buffers Absorb Variability Until They Do Not

A transfer buffer can absorb routine lateness. As long as inbound variation stays within the buffer, downstream departure remains stable.

Once variation exceeds that margin, the missed connection produces a step change in lead time.

This gives buffers an important role: they convert small upstream variability into stable downstream performance—at the cost of planned waiting.

Removing Buffers Can Improve the Average and Damage the Tail

Suppose a hub reduces scheduled transfer time from two hours to thirty minutes. Shipments that connect successfully now move faster.

But if routine inbound variation often exceeds thirty minutes, more shipments miss the onward service. The average may improve modestly while the late tail becomes much worse.

This is a classic hostile test for “leaner” logistics: did the faster plan remain robust to ordinary variance?

Variability Should Be Compared With the Receiver’s Tolerance

There is no universally acceptable amount of arrival spread.

A household parcel may tolerate a broad delivery window. A production line, surgery schedule or perishable product may not.

Performance therefore has to be measured against consequence. A two-hour standard deviation can be trivial in one flow and unacceptable in another.

ETA Accuracy Is a Forecasting Version of the Same Problem

An ETA system predicts where a particular shipment will fall within the distribution given current evidence.

As the shipment progresses, the estimate should update and uncertainty should ideally narrow.

A precise-looking ETA that repeatedly misses is less useful than a wider but well-calibrated window. Operational trust comes from whether predictions correspond to observed outcomes.

Variability Can Be Segment-Specific

A lane may look unreliable overall because one subset of shipments behaves differently.

• One day of week may have lower service frequency.
• One terminal may produce longer dwell.
• One product class may trigger inspections.
• One carrier or connection may generate the late tail.
• One receiving site may have narrow appointment windows.

Segmentation turns “the route is variable” into a more useful causal question: which population creates the spread?

Seasonality Can Change the Distribution

Peak periods, holidays, monsoons, promotions and major events can change capacity utilisation and service frequency.

A lane that is stable during ordinary weeks may become highly variable when terminals, warehouses and delivery fleets approach practical capacity.

Historical averages that blend peak and normal periods can hide both realities.

Averages Can Reward the Wrong Improvement

Imagine a change that makes already-fast shipments one hour faster but leaves late shipments unchanged.

The average improves. The receiver’s stockout risk may not.

Another change might leave the median unchanged but reduce the number of six-day outliers dramatically. The average improvement could be smaller while the service becomes much more dependable.

Choose metrics that match the receiver’s pain.

Arrival Variability at Three Zoom Levels

One shipment

Which event caused this shipment to arrive earlier or later than expected?

One lane

What do the median, spread, percentiles and late tail reveal across many shipments?

One network

Which shared nodes, cut-offs and service cycles create correlated lateness across several routes?

A Singapore Lens

Singapore’s role as a high-connectivity logistics hub makes variability especially consequential because many shipments depend on onward connections through port, airport, warehousing and regional distribution systems.

High schedule density can reduce the penalty of a missed connection when another service follows soon. At the same time, concentrated hub dependencies can make disruption at a critical node affect many flows together.

The transferable lesson is to measure both ordinary spread and shared tail risk.

Hostile Test: “Our Average Delivery Time Improved”

Ask four more questions.

• Did the median improve?
• Did the late percentiles improve?
• Did the worst tail become smaller or larger?
• Did receiver-facing service levels improve?

An average is a useful compression. It should not become a hiding place.

Arrival-Variability Audit

• What is the mean arrival time?
• What is the median?
• What are useful late percentiles?
• How wide is the typical spread?
• How severe is the late tail?
• Which dwell intervals contribute the most variation?
• Which cut-offs create separate late populations?
• How much transfer buffer exists?
• Which shipments exceed that buffer?
• Does variability change by day, season, carrier, node or product type?
• Are ETAs calibrated against actual outcomes?
• Which metric best represents the receiver’s consequence?
• Did a recent speed improvement reduce or increase tail risk?

Evidence and Further Reading

The World Bank’s Logistics Performance Indicators 2.0 — 2025 Report is the central evidence anchor for this article. Its redesigned shipment-level approach measures supply-chain connectivity, speed and reliability and highlights unpredictability concentrated at ports, transshipment hubs and inland checkpoints. The World Customs Organization’s Time Release Study Guide, Version 4 complements that view with structured time-interval, segmentation, trend and outlier analysis for border-release performance.

Return to the Logistics Hub

Arrival variability completes Batch 05’s time layer: lead time → dwell → cut-off → spread. Return to How Logistics Works | How the Right Thing Reaches the Right Place at the Right Time to reconnect time to inventory, warehousing, transport, handoffs and final receipt.

---

Final compression: an average can describe the centre of a logistics system while hiding the part that hurts. Reliable logistics pays attention to the spread, the missed connections and the late tail because receivers do not live inside averages; they live inside particular arrivals.