Since launching locally in late 2017, Amazon Australia has already generated some interesting insights into the spending habits of Aussie consumers.
Speaking at Amazon’s Innovation Day in Sydney this week, Jenny Freshwater, the global company’s leader of forecasting and capacity planning, discussed the issues of supply chain optimisation and forecasting product demand.
She revealed some interesting examples of the way Australian consumers behave, which differs between regions and also from what happens in similar countries like the US.
Ms Freshwater said that part of forecasting future demand involves looking back at past demand. For example, she said that Australians buy a lot of water, and that Mount Franklin sells more than any other brand, making it a safe assumption for future demand.
When Amazon Australia launched, she explained, it was clear that Australians consumed large amounts of Vegemite, but that the retailer had severely underestimated just how much demand there was for the spread.
According to Ms Freshwater, Amazon Australia has initially targeted Sydney and Melbourne, and even between the two cities, there are noticeable differences in what consumers buy most, such as demand for music accessories and vinyl records being far more pronounced in Melbourne than its northern neighbour.
However, she said that overall shopper demand has been greater in Sydney than in Melbourne, despite the first distribution centre being in Victoria.
Another interesting finding is that Australians are quite different in how they buy Mother’s Day gifts than Americans.
She noted that, in the US, demand stays virtually flat between two weeks out from Mother’s Day and the actual week of it.
In Australia, though, Amazon has seen demand roughly double in the week prior, before dropping back to a similar level in the week of Mother’s Day — clearly placing more strain on Australian retailers and suppliers to cater to this sudden spike.
Ms Freshwater admitted that forecasting demand is “not an exact science”, and that even a retailer of Amazon’s size and scale sometimes gets it wrong, such as purchasing an oversupply of mattresses which, because of their size, meant storing was not economical, had to be sold at a steep discount to clear the stock.
As such, she said that forecasting still require the input of people and can’t be left solely to machine learning and predictive technologies, because forecasting models can’t tell everything, or do so 100 per cent correctly.
She also cautioned about the types of data being inputted into digital forecasting systems.
“The saying ‘garbage in, garbage out’ is none truer than in machine learning,” she said.
Adam Zuchetti is the editor of My Business, and has steered the publication’s editorial direction since early 2016.