Amazon's 90% business-hour delivery gate starts in 34 days

A seller-fulfilled package can arrive before its promised date and still fail Amazon’s next delivery gate. If the first attempt reaches a business after it has closed, the order can miss Business Hour Delivery Rate even though a conventional on-time metric looks healthy.

That distinction becomes enforceable in 34 days. Starting September 30, 2026, Amazon will require US sellers to keep BHDR at 90% or higher for seller-fulfilled shipments to Amazon Business customers. The metric uses a rolling 14-day period. Sellers below the line on September 30 will be notified; if the rate has not improved by October 30, Amazon says affected seller-fulfilled offers may be deactivated for business customers.

This is a first-attempt window, not another on-time score

BHDR asks whether the first delivery attempt happened inside the customer’s operating hours. It is not the same question as whether the parcel arrived by the promised date, whether the seller confirmed shipment on time, or whether a carrier eventually completed delivery.

The scope is also narrower than an account-wide selling suspension. Amazon’s announcement covers seller-fulfilled shipments to Amazon Business customers in the US store. It says FBA and retail offer eligibility will not be affected. The stated downside is loss of seller-fulfilled visibility to business buyers, which can still be material for a catalog with concentrated B2B demand.

Do not combine BHDR with Late Shipment Rate or On-Time Delivery Rate in one blended scorecard. Put the promised date, first-attempt timestamp, customer hours, and final delivery timestamp in separate fields. A shipment can pass one test and fail another.

The defect budget is smaller than 90% sounds

The percentage hides a shipment-count problem. To remain at or above 90%, failed eligible shipments cannot exceed the floor of 10% of the window’s total. This table is arithmetic on Amazon’s threshold, not an Amazon-issued tolerance schedule:

Eligible shipments in the 14-day window	Maximum failures that still pass
1–9	0
10–19	1
20–29	2
30–39	3
100	10

Low B2B volume is therefore not automatically low risk. With nine eligible shipments, one closed-door attempt produces 88.9%. With 19, two failures produce 89.5%. The rate will also change as old shipments leave the rolling window, so a green dashboard today is not a permanent cushion.

Track both numerator and denominator. A percentage without shipment count cannot tell an operator whether the problem is one disputed scan or a repeatable carrier-service pattern.

Let the report decide the carrier-service map

Amazon directs sellers to Account Health to review BHDR. In an official forum response, an Amazon moderator said the downloadable report includes first-attempt delivery time, failure reason, carrier and ship method, and the customer’s business hours on the attempt date.

Use those fields as a routing dataset. Group failures by carrier plus service, not carrier name alone. Then split by weekday, origin warehouse, destination region, and customer-hours pattern. “UPS is good” is not a control; “this service from this origin stays inside the recorded window on these lanes” is testable.

Reconcile exceptions against carrier tracking before changing the map. If Amazon records a closed day or outside-hours attempt that conflicts with the carrier event, preserve both records and open a case with order-level evidence. One screenshot of the dashboard cannot show whether the defect was routing, customer-hours data, a scan mismatch, or a genuine late-day attempt.

Amazon’s protected path has three parts

Amazon says shipments fulfilled with Automated Handling Time, Shipping Settings Automation, and Amazon Buy Shipping together are guaranteed to meet the BHDR requirement. Treat “together” as the control boundary. Buying a label alone is not the same configuration described in the announcement.

AHT aligns the handling promise with recent operations. SSA calculates delivery promises using locations and carrier performance. Buy Shipping selects the label inside Amazon’s workflow. Before relying on the path, capture whether all three were enabled when the order was placed and retain the purchased label. Do not assume a setting turned on later protects earlier orders.

The protected path is still a business decision. Automation must reflect real cutoffs, warehouse days, and carrier pickups. If a service is materially more expensive, compare the incremental shipping cost with the contribution margin and B2B revenue at risk rather than treating compliance as costless.

Use the 14-day window to work backward

September 30 is 34 calendar days from publication, and October 30 is 64 days away. The 14-day measurement means the operational change should be stable before the requirement date, not merely approved on it.

Work backward in four passes. Pull the baseline now. Within a week, identify the carrier-service and weekday cells causing most failures. By mid-September, route eligible orders through the chosen map or Amazon’s three-part path. During the final two weeks, review the report daily enough to see which shipments are entering and leaving the denominator.

That cadence is a recommendation derived from the rolling window, not an Amazon deadline. Amazon has not published a seller-specific sample-size exception in the announcement. Use the current help page and the notice in your own account as authority if their language changes.

The Operator Read

BHDR moves a carrier’s clock into the seller’s eligibility system. Complaining that the final attempt is outside the warehouse’s control may be understandable, but it does not create an operating answer.

The controllable layer is the route: which orders qualify, which service is selected, which customer-hours record is used, which Amazon settings were active, and which evidence survives when the scan looks wrong. Build that layer while the metric is still observable rather than after a business offer disappears.

Ninety percent is not a generous average for a small denominator. Count the shipments, calculate the failure budget, and make every exception explainable before September 30.