Australia Post using machine learning to tell you when to expect a delivery

It will give customers a two-hour window for when a delivery will arrive, although it has mapped it down to plus or minus 15 minutes.
Written by Asha Barbaschow, Contributor

Over the last six months, Australia Post has experienced a 20% decline in letters, but a boost in households shopping online. During that period, the postal service had 8.1 million households within its 12 million delivery points that ordered goods online.

People were also buying bulkier and heavier items, such as gym gear, office supplies, and in true Australian form, alcohol, so Australia Post also had to change its mode of delivery.

Australia Post executive general manager of transformation and enablement John Cox said this meant the organisation had to transition many of its bike delivery posties into van delivery roles.

"The challenge with all of that is our most experienced drivers know their route, they work out exactly where to go and they optimise very quickly the pathway. The posties have a different way of delivering, they actually have a very safe route where they always turn left to avoid going into traffic and it's the same path pretty much every day," Cox said.

"Going to a van and delivering parcels, actually requires a much more dynamic route that they need to select."

Speaking at the Digital Transformation Agency's Disruption and Change Digital Summit on Tuesday, Cox said the organisation took the scanner app that it built and fed it delivery route information. Using machine learning, Australia Post was provided with the best routes for delivery, he said.

See also: Australia Post using its data to reject claims of unsafe postie practices

But what that gave Australia Post was a very accurate view of where a postie is at all times, so it's now trialling an estimated time of arrival feature.

"So based off when the postie scans the parcel in the morning to put in their bag, we'll be able to notify the consumer that they will be delivered within a certain time," he said.

The delivery window will be set at two hours, even though Cox said the organisation has it down to within plus or minus 15 minutes.

"We want to build in a little bit of flexibility there," he explained. "But what we've been able to do is take the machine learning models, been able to take the scanning platform, we've been able to take the fact that in real time that is now feeding up into the cloud and then provide all of that insight and make sure that it's accurate throughout the day."

Australia Post is also using machine learning to predict mail volumes.

"We got it to a point now off that data analytics that we can see plus or minus 5% on a daily basis what we think the volumes are going to be and it's proven to be very accurate," he said. "We've also done forecasts into what we think is going to happen over Christmas and what would happen in another state lockdown."

The postal service has been investing in a telco transformation over the last few years, rolling out upgrades to 4,500 post offices as well as equipping delivery staff with better mobility capabilities, including a single scanning platform.

"We had multiple devices using different technologies that were really difficult to deploy change on, in fact we could only deploy change once every six months across the fleet, which in a moment where you've got to pivot very quickly, it just isn't feasible, so we moved it onto a very modern scanning platform -- its actually an Android based technology and the app is available then to be deployed and very easy to make change," Cox said.

The organisation is also in the process of shifting information from its existing data warehouse onto the Google Cloud platform.

During Senate Estimates last month, it was revealed that four Australia Post executive-level staff in 2018 received Cartier watches from the organisation as gifts, totalling AU$12,000.

Minister for Communications, Cyber Safety and the Arts Paul Fletcher later announced an investigation into the claims and the organisation's CEO Christine Holgate has since resigned from her post.


Editorial standards