Australia Post's Machine-Learning Route Planner — Two-Hour Delivery Windows for 500,000 Parcels a Day

Australia Post, the country's postal service and a wholly government-owned corporation, delivers more than a billion items a year to Australian addresses. Facing customer complaints about vague "some time this week" delivery windows, it began piloting a machine-learning-based route planner in October 2020.
The system draws on data from posties' handheld scanners, used to scan each parcel before a delivery run, feeding a route-optimisation model that calculates an estimated time of arrival narrowed to a two-hour window and pushes it to the customer as a notification.
According to Australia Post's own 2020-21 Annual Report, tabled in the Australian Parliament, the ETA notification service reached more than 1,500 drivers across 130 metropolitan sites by the end of the financial year, sending over 2.5 million notifications with 96% of parcels arriving inside the promised two-hour window. The broader route-optimisation software was rolled out to 3,700 drivers nationally, helping move more than 500,000 parcels a day, and a related machine-learning pilot for identifying driver safety exposures reported an 85% reduction during testing.
These figures are self-reported in a statutory annual report rather than an independently audited study, and Australia Post has not published an accuracy breakdown for regional or remote delivery runs, which are historically harder to predict than metropolitan ones. As a government business enterprise, Australia Post must report its performance publicly and answer to Parliament, but no dedicated AI-governance framework specific to this routing algorithm has been made public.

Read the full analysis: https://auspost.com.au/content/dam/auspost_corp/media/documents/2021-australia-post-annual-report.pdf