a2 Milk forecasts sales with AI

The Sydney-based a2 Milk Company has said it's seen more precise sales forecasts from using machine learning algorithms than from its human-generated results.

The diary business said utilising machine learning capabilities from Oracle predictions were 4.5 percent off the real sales compared to its team, which were 7.8 percent of total sales.

James Madigan, chief information officer at the a2 Milk Company, told an Oracle Data & AI Forum audience the business is now embedding the solution into its daily sales dashboard.

“We looked at the number of sales for a period, the Oracle machine learning algorithm was 4.5 percent off the actual sales, the human-generated was 7.8 percent,” Madigan said. 

Madigan added “dirty data” was used and didn’t include such as non-trading days however, was still able to create a more precise response.

 “We're going to embed it into our daily sales dashboard that we have and that will now come with a prediction of sales over a longer period now using that algorithm.”

He said a2 Milk initially began implementing Oracle solutions across its business “about four years ago” when it was looking “to start a genuine digital transformation.”

“A2 was a company that had grown rapidly, but we had a lot to do, to build an IT data analytics platform and we rapidly set about implementing the Oracle Cloud applications- so finance, supply chain, EPM, HR and we're probably going to be doing more with the HR shortly.

“We spent the first few years getting the data right … had we tried to do this before we implemented a consistent set of processes platform it would have been a disaster,

“Two years in we had this problem, what are we going do with data analytics and AI?

“We then set about getting ourselves a data warehouse or lake house and we landed quite quickly on the Oracle autonomous data warehouse”.

Madigan explained the decision made “sense to extend the investments we made in Oracle.”

“We're [also] looking for a platform that will be very open because we estimated only about 50 percent of the data we're ever going to need to make decisions at a2 is going to come out of Oracle sales.”