Delivery Promise Accuracy: The Last-Mile Efficiency Metric That Predicts Repeat Purchase

The two numbers that are both called on-time

A North American retailer reports 96% on-time delivery. The number is accurate, audited, and reported to the board every quarter.

A customer who ordered on a Tuesday was shown “arrives Thursday” at checkout, received a notification on Wednesday narrowing it to “Thursday between 2pm and 6pm,” and took delivery on Friday morning. Operations recorded that delivery as on time, because the carrier service level allowed three to five business days and Friday fell inside it.

Both facts are true. They describe different things, and only one of them is what the customer bought.

This is not a rounding difference or an edge case. It is a structural gap in how most enterprises measure last-mile delivery efficiency, and it exists because the promise shown at checkout and the window used to score performance are usually produced by two different systems that were never reconciled. The storefront generates a date from a lead-time rules table maintained by e-commerce. Operations scores against a planned window or a carrier service level. Nobody computes the difference, so nobody knows how large it is.

The number that predicts whether a customer orders again is the second one. The number on the quarterly report is the first.

Also Read: Last-Mile Delivery Efficiency: Cost Reduction Guide 2026

Why on-time rate is gameable, and which direction it gets gamed

Here is the part that should concern anyone using on-time rate as a customer experience metric.

On-time rate has two inputs: when the delivery arrived, and what it was measured against. Operations controls both. The second one is easier to change than the first.

Widening the promise raises the on-time rate immediately, at zero operational cost, with no deception involved. A five-day window is easier to hit than a two-day window. A date without a time window is easier to hit than a four-hour slot. Every buffer added to protect the metric makes the number go up.

It also makes the promise less useful to the person receiving it. A customer told “sometime next week” cannot plan around it, will not be home, and gains nothing from the delivery arriving inside a window so wide it carried no information. So the metric improves while the experience degrades, and the two movements are causally linked rather than coincidental.

This means a rising on-time rate is ambiguous evidence. It can indicate better execution. It can equally indicate that someone widened the windows after a bad quarter. Without tracking promise width alongside promise accuracy, those two situations are indistinguishable in the reporting, and only one of them is worth anything.

Why competing on speed breaks promises

The reliability-over-speed argument is well established and worth taking one step further, because the relationship is not merely that customers prefer reliability. It is that pursuing speed actively degrades promise accuracy.

A delivery promise is kept or broken depending on how much slack sits between the committed time and the operation’s actual capability. The fastest promise available is by definition the one with the least slack, because speed is achieved by removing buffer. So the same operation, promising next-day rather than three-day on identical orders, will break a materially higher share of its promises without anything about its execution having changed.

That produces an unattractive position for a brand competing primarily on speed. Each increment of promised speed increases the breakage rate, and each broken promise costs more than the slower promise would have. Failed deliveries alone are commonly benchmarked around $17.78 per failed attempt once reattempt labor, routing disruption, and support handling are counted, and that figure excludes the retention effect entirely.

The strategic implication for a CMO is uncomfortable and clear. Faster promises are a growth lever with a reliability cost attached, and most organizations have never measured the cost side. The right question is not how fast the promise can be, but how fast it can be while remaining keepable at the accuracy level your category requires.

Also Read: Top 10 Last-Mile Delivery Metrics to Track in 2026

Why a broken promise costs more than a slow one

The retention mechanism is worth stating explicitly, because it explains why this metric behaves differently from every other delivery measure and why it belongs to marketing.

A customer who is told five days and receives five days has had a slow experience. They planned around five days, the plan held, and nothing was disrupted. A customer told Thursday afternoon who receives Friday morning has had a faster delivery in absolute terms and a worse experience, because they arranged a day around a commitment that was withdrawn without warning.

The cost of the second case has two parts. The first is the disrupted arrangement, which is the obvious one and the smaller one. The second is what the customer now knows: that this brand’s stated dates are not dependable. That inference is not confined to delivery. It transfers to stock availability, to return timelines, to service commitments, and to the next promise the brand makes, which will be discounted on arrival.

This is why promise accuracy predicts repeat purchase better than speed does. Speed is a feature customers compare across brands. A kept promise is evidence about whether a brand’s claims can be believed, and that evidence compounds across every future interaction. It is also why recovery matters so much on the occasions when a promise does break, because a promise broken and handled well produces a different inference than one broken in silence.

The practical consequence for a CMO is that delivery promise accuracy is not an operational KPI that happens to touch marketing. It is a measure of brand credibility that happens to be produced by operations.

Also Read: Last Mile Delivery Optimization: Enterprise Strategies That Scale

The four properties of a promise that predicts repeat purchase

Promise accuracy on its own is insufficient, because a wide promise trivially achieves it. Four properties together describe whether a promise did any work.

Specificity. How narrow the window was at the moment of purchase. A four-hour slot is a commitment a customer can arrange their day around. A five-day range is a disclaimer.

Stability. How many times the promise changed between checkout and delivery. Each revision is a small withdrawal of the original commitment, and revisions that narrow a window read differently from revisions that move a date.

Accuracy against the last version shown. This is the definitional point most measurement gets wrong. Customers remember the most recent and most specific thing they were told, not the original checkout estimate. If a notification narrowed delivery to Thursday afternoon, Thursday afternoon is the promise. Scoring against the original three-to-five-day service level measures a commitment the customer stopped holding days ago.

Computability. Whether the promise was derived from actual network state, meaning capacity, carrier serviceability, and the operation’s real performance on that lane, or read from a static lead-time table. A promise from a rules table is a guess presented with the confidence of a commitment, and its accuracy is whatever the table’s assumptions happen to be worth this week.

The fourth property determines the other three. An operation that cannot compute a date cannot narrow it safely, cannot keep it stable, and cannot improve its accuracy except by widening it.

Three ways enterprises measure delivery performance

The row that matters is who can change the benchmark. A metric whose target can be adjusted by the team being measured is a management tool, not a customer measure. Promise accuracy is the only one of the three where the benchmark is fixed the moment the customer sees it, which is precisely what makes it uncomfortable to adopt and worth adopting.

The final row is the practical obstacle. Measuring promise accuracy requires retaining every promise shown to a customer and when it was shown, and most commerce stacks retain only the current expected date. That is a data retention decision rather than an analytics problem, and it has to be made before the measurement is possible.

Also Read: Last Mile Delivery Analytics: Key Metrics & Benefits in 2026

What to measure

Promise accuracy against the last promise shown. Deliveries arriving inside the most recent window communicated to the customer, at the granularity communicated. No credit for the wider original.

Median promise width at checkout, trended. The buffer-inflation detector. Track it on the same chart as on-time rate. If both are rising, execution is not improving.

Promise revisions per order. How often the commitment changed, split by revisions that narrowed and revisions that moved.

The promise-to-on-time gap. The difference between your reported on-time rate and your promise accuracy. This single number quantifies how much of your reported performance is measurement artifact.

Repeat purchase rate segmented by promise kept. The retention link, and the reason this belongs to marketing rather than operations. Most enterprises can compute this from existing order history within a quarter, and it converts an operational metric into a revenue argument that a CFO will act on.

Also Read: How to Reduce WISMO Calls in Retail

How Locus makes the checkout promise an operational commitment

Locus, the world’s first Decision-Intelligent, Agentic TMS, treats the delivery promise as an output of the operation rather than an input to it. Its Dispatch, Capacity, and Carrier agents, coordinated by an Orchestrator, run a continuous Sense-Decide-Execute-Learn loop against a model of more than 250 real-world constraints, which is what allows a delivery date to be computed from live capacity, carrier serviceability, and actual lane performance instead of read from a lead-time table.

Two capabilities carry the argument in this piece. Carrier status normalization resolves every carrier’s proprietary event codes into one standard set, which is the prerequisite for knowing whether a promise was kept when the parcel moved through a third party. And because each decision retains the inputs and plan version behind it, the promise history is preserved, which is what makes accuracy against the last promise shown measurable rather than theoretical.

Locus is recognized by Gartner for seven consecutive years, featured in the 2026 Hype Cycle for Supply Chain Execution and Logistics Technologies, named a Leader in TMS by QKS Group (SPARK Matrix), and ranked #1 in Route Planning on G2’s 2026 Best Software Awards. In October 2025, Ingka Investments, the investment arm of Ingka Group, the world’s largest IKEA retailer, acquired Locus. Locus continues to operate independently. Further analyst recognition is published in full.

Two deployments show what changes when the promise becomes computable.

A leading ASEAN apparel retailer running a large store network alongside a global e-commerce business could not compute a delivery date at all, because last-mile moved almost entirely through carriers with separate systems, rates, and service areas reporting in their own status codes. The storefront therefore displayed a rough lead time, and the gap between that lead time and reality produced hundreds of thousands of delivery and returns complaints in a single half-year. The sequence of the fix is the instructive part: carrier statuses were harmonized into one standard set first, a network-aware delivery date was computed second, and every shipment was then tracked against its promise. Reported outcomes were a 40%+ reduction in WISMO and returns queries and delivery SLA above 99%. The complaints fell because the promise became keepable, not because the tracking page improved.

A grocery brand delivering across 30+ North American cities through contracted 3PLs moved fresh and perishable orders where a broken promise means spoiled product rather than an inconvenience. Orchestrating carriers through a single layer produced 33% faster deliveries and 15% lower fulfillment cost, and customer support resolution became 10 to 20 times faster. That last figure matters for promise accuracy specifically, because it describes the recovery path on the occasions when a promise does break, which is the interval where retention is either preserved or lost.

Request a Locus delivery promise accuracy assessment to measure your promise-to-on-time gap, detect buffer inflation in your checkout windows, and establish whether your promised dates are computed or assumed.

Run the segmentation before the next board deck

The fastest way to find out whether this matters in your business takes one analyst and existing data.

Pull twelve months of orders. Split them by whether the delivery arrived inside the last window communicated to the customer, not the carrier service level. Then compare repeat purchase rate, order frequency, and average order value between the two groups over the following six months.

If the two groups behave identically, promise accuracy is not a retention lever in your category and you can stop. In most consumer categories they do not behave identically, and the difference is large enough that the number belongs next to your on-time rate rather than instead of it.