Delivery promise statistics at carrier handoff

Updated 13 source citationsSources checked 13 November 2026

On this page
  1. Quick answer
  2. Key Takeaways
  3. How often did orders arrive late when little time remained?
  4. Does adding more promise time reliably reduce late deliveries?
  5. What should a delivery answer check after handoff?
  6. Which orders does this study leave out?
  7. Would a short-window alert find most late orders?
  8. Methodology and Sources
Download study text

Quick answer

Delivery promise statistics at carrier handoff show why a parcel leaving the seller is not enough to reassure a shopper about its arrival date. In an original analysis of 89,712 historical Olist orders, 45.27% of those handed to the carrier with one to three calendar days left before the estimated date arrived late. The corresponding share was 6.31% with eight to fourteen days remaining. These are observed groups of delivered orders, not forecasts for a current store. (Convo Rescue analysis, Olist data, Nov 2026).

The useful question is what the existing estimate still means after dispatch. This study joins the carrier handoff, estimated arrival and actual arrival records for Brazilian marketplace purchases made from January 2017 through July 2018. It reports the remaining time, differences across years and destinations, and the orders the analysis cannot describe. The counterexample matters: in the 2018 group, fifteen or more remaining days did not produce a lower late rate than eight to fourteen.

Key Takeaways

How often did orders arrive late when little time remained?

Orders with little calendar time left at handoff had a much higher observed late rate. This is a warning to check the current arrival estimate, rather than treating a shipping confirmation as an answer to a deadline question.

Delivery promise statistics at carrier handoff: late arrival was 45.27% with 1 to 3 calendar days remaining, 15.09% with 4 to 7, 6.31% with 8 to 14 and 6.15% with 15 or more.
Figure 1: Late arrival differed by remaining calendar time at carrier handoff, with uncertainty and unequal group sizes preserved. Source: Convo Rescue analysis, Brazilian E-Commerce Public Dataset by Olist, Nov 2026. Original aggregate analysis of the cited source. Study populations and limits appear in the text.

Remaining time is the estimated delivery date minus the carrier handoff date, measured in calendar days. An order handed over on Monday with a Thursday estimate has three calendar days remaining. This definition does not assume three working days, a particular collection cutoff or any carrier service level.

Time remaining at carrier handoffDelivered ordersArrived after estimateLate shareDescriptive 95% interval
One to three calendar days34915845.27%40.13% to 50.52%
Four to seven calendar days1,92229015.09%13.56% to 16.76%
Eight to fourteen calendar days13,9558816.31%5.92% to 6.73%
Fifteen or more calendar days73,1784,5026.15%5.98% to 6.33%

The table uses one record per eligible delivered order. Intervals use the Wilson binomial method and describe uncertainty under an independent-order approximation. The small one to three day group deserves a wider interval than the much larger fifteen or more day group. Neither interval measures the chance that a different merchant's next parcel will arrive late. (Convo Rescue analysis, Olist data, Nov 2026).

What if the estimated day had already arrived?

Another 74 orders reached the carrier on their estimated delivery date; 70 were delivered after it. For 234 orders, the estimated date was already past when the carrier handoff was recorded, and all 234 were late. That last result follows from the chronological definition: an order delivered after its handoff cannot meet a date already missed at handoff. It is a useful status check, not an independent discovery that predicts late delivery. (Convo Rescue analysis, Olist data, Nov 2026).

A customer in that position needs an updated answer. Repeating the original date or quoting a generic carrier transit time leaves the missed commitment unresolved.

Does adding more promise time reliably reduce late deliveries?

This analysis cannot show that changing a promise causes better delivery. It also contains a clear counterexample to a simple rule that more remaining time always means a lower late rate.

Purchase yearEight to fourteen days remainingFifteen or more days remaining
2017405 late among 5,675, or 7.14%1,686 late among 36,867, or 4.57%
January through July 2018476 late among 8,280, or 5.75%2,816 late among 36,311, or 7.76%

The direction of this particular comparison reversed between the two purchase periods. A longer estimate can be assigned to a harder destination, a different seller or a product with a less predictable fulfillment path. Pooling those orders does not make their underlying journeys equivalent. (Convo Rescue analysis, Olist data, Nov 2026).

The result argues against taking the pooled graph and turning its group boundaries into a universal promise policy. An estimate can have plenty of calendar time left and still be unreliable for reasons that the order file does not identify.

How much did destination change the comparison?

Among orders with four to seven days remaining, the late share was 10.86% for São Paulo, 183 of 1,685 orders, and 45.15% for the other states combined, 107 of 237 orders. These are not matched routes. The much smaller group outside São Paulo can contain a different mix of distances, sellers and shipping services. (Convo Rescue analysis, Olist data, Nov 2026).

A merchant should therefore examine its own destination and service groups before adopting a cutoff. Even a destination breakdown cannot isolate a carrier effect without the necessary carrier, service and operational records. This dataset does not support ranking carriers or attributing a missed date to a particular company.

What should a delivery answer check after handoff?

Check the current estimate and the deadline the shopper actually needs. Handoff establishes a stage in the journey. It does not establish that the original arrival date is still defensible.

Consider a shopper choosing an item for an event on Friday. The parcel was collected on Thursday, while the checkout still shows a Friday estimate. The useful reply needs the current delivery information for that order and destination. A general statement that standard shipping takes several days cannot resolve whether Friday remains realistic.

Record or questionWhat it can establishWhat still needs checking
Recorded carrier handoffThe order file contains a handoff eventWhether a later scan or exception changed the journey
Original estimated dateThe date in the order's estimate fieldWhether that estimate is current or was revised elsewhere
Shopper's required dateThe deadline that makes the purchase usefulWhether the verified delivery option supports it
Updated order or carrier recordThe latest available estimate or exceptionThe conditions and uncertainty attached to that update

The table is a decision procedure, not another result from the archive. It keeps each claim tied to the record that could support it. If an estimate has changed, the answer should say which date is current and what the next useful step is. It should not invent certainty to keep the sale moving.

For the distinction between an end-to-end estimate and the transport stage, see checkout delivery date statistics. Different buyer needs across delivery options are covered in delivery option statistics.

Which orders does this study leave out?

The purchase window contains 92,580 orders. Of these, 89,860 have delivered status. The final timing comparison retains 89,712 after excluding nine delivered orders with a missing required timestamp and 139 with an inconsistent chronological sequence. (Convo Rescue analysis, Olist data, Nov 2026).

Orders not marked delivered are outside the rate calculation. They include canceled, unavailable, shipped and processing records. A delivered-only late rate cannot describe the full risk of an order failing to reach the shopper, and unresolved shipment records must not be counted as successful on-time deliveries.

The dataset also does not preserve every estimate the shopper saw over time. The field supplies an estimated date, not a complete history of checkout wording or later revisions. That distinction prevents the analysis from measuring whether a specific customer-facing sentence was accurate when it was sent.

Would a short-window alert find most late orders?

An alert can identify a high-risk group and still miss most of the operational workload. Combining the already-past, same-day, one-to-three-day and four-to-seven-day groups gives 2,579 orders, or 2.87% of the eligible cohort. That queue contains 752 of the 6,135 observed late arrivals, or 12.26% of all late orders. The other 5,383 late arrivals had at least eight days remaining at handoff. These are our calculations from the six published groups, not a newly trained prediction model. (Convo Rescue analysis, Olist data, Nov 2026)

Orders with at most seven calendar days remaining at carrier handoff account for 2.87 percent of the Olist cohort and 12.26 percent of observed late arrivals. Orders with at least eight days remaining account for the other 87.74 percent of late arrivals.
Figure 2: Late risk within a small group and the share of all late orders captured by that group use different denominators. A useful queue needs both. Source: Olist handoff groups, original aggregation, Nov 2026. Original aggregate analysis of the cited source. Study populations and limits appear in the text.

The short-window queue's late share is 29.16%, compared with 6.18% for orders with at least eight days left. Those observed group rates make short remaining time a useful reason to check an order. They do not make the remaining orders safe to ignore. The larger group contains far more orders, so a lower rate still generates most late arrivals. In particular, 4,502 late orders came from the fifteen-or-more-day group alone, 73.38% of all observed late arrivals. (Convo Rescue analysis, Olist data, Nov 2026)

A merchant building a support queue should therefore decide which problem it wants to solve. A small queue can concentrate checks on concerning records when staff capacity is limited. A programme intended to cover nearly every late delivery needs broader signals, such as lack of progress, exceptions and updated route information. These archive counts do not establish how many late arrivals a live team could prevent or resolve.

There is an additional timing limit. The already-past and same-day groups include cases where the delivery commitment was missed or nearly exhausted at handoff. A queue that detects them is partly identifying an existing problem, rather than predicting a future one far enough in advance to intervene. Measure lead time separately from retrospective accuracy before calling an alert proactive.

For a current operation, record the number of eligible orders, the number flagged, late arrivals among flagged orders, late arrivals outside the queue and how long before the deadline each alert appeared. Review the actual intervention alongside those measurements. A high-risk flag is useful only if the next check or action improves the answer the shopper receives.

Methodology and Sources

We analyzed the public Olist archive and selected orders purchased from 1 January 2017 through 31 July 2018. Eligible orders have delivered status, complete purchase, carrier handoff, customer delivery and estimated delivery fields, a purchase timestamp no later than handoff, handoff no later than delivery, and an estimated calendar date no earlier than purchase. Remaining time is estimated date minus handoff date. Late means recorded customer delivery date after estimated date. We retained separate groups for an already-past estimate and a same-day estimate. Purchase-year and destination checks use the same eligible orders and customer join. The analysis is observational and does not adjust away seller, route, product or service differences.

The source covers historical Brazilian marketplace orders from 2016 through 2018. Its current catalog was last updated in October 2021; the aggregate calculations here were made in November 2026. This article uses the month of the original analysis in citations that name Convo Rescue analysis. The archive contains no current merchant performance claim. Aggregate tables and the reproduction script preserve the exclusions and source hashes.

Glossary

  • Carrier handoff: The recorded point when an order moved from the seller to the carrier.
  • Remaining time: Calendar days between the recorded handoff date and estimated delivery date.
  • Late arrival: A recorded customer delivery date after the order's estimated date.
  • Calendar day: A date boundary, including weekends and holidays.
  • Delivered-only cohort: Orders already marked delivered, excluding other final or unresolved statuses.
  • Wilson interval: A binomial interval that makes sample size visible around an observed proportion.
  • Selection: The fact that orders in different groups may differ in ways beyond the variable being compared.

Sources

Cite this page

Use this citation when referring to the analysis. Retain the original source population and period with any quoted finding.

Convo Rescue. "Delivery promise statistics at carrier handoff." Convo Rescue, November 2026. https://convorescue.com/guides/delivery-promise-time-at-handoff-statistics/
<a href="https://convorescue.com/guides/delivery-promise-time-at-handoff-statistics/">Delivery promise statistics at carrier handoff</a> (Convo Rescue, November 2026)