New ZealandFood and restaurants2024 – ongoing

The a2 Milk Company

Machine learning sales forecasts in the daily sales dashboard

a2 Milk's Oracle machine learning forecast came within 4.5% of actual sales, against 7.8% for human forecasts

In progressCompany's own claim

Problem

The a2 Milk Company relied on human-generated sales forecasts. It wanted forecasts that tracked actual sales more closely.

Approach

The company used machine learning algorithms from Oracle to forecast sales. The model was fed imperfect data that left out items such as non-trading days. At the time of the report, a2 Milk was adding the forecasts to its daily sales dashboard, and the tool was also meant to predict sales over longer periods.

Results

4.5%gap between the machine learning forecast and actual sales, in one comparison, per a2 MilkCompany's own claim[1]

7.8%gap between the human-generated forecast and actual sales, in the same comparison, per a2 MilkCompany's own claim[1]

Takeaway

The model beat the human forecast even with imperfect input data. The report gives one comparison, not a long-run track record.

Sources

a2 Milk forecasts sales with AI
iTnews12 Apr 2024News article

Source summaries

[1]iTnews12 Apr 2024

iTnews reports that a2 Milk uses Oracle machine learning to forecast sales and is embedding the forecasts in its daily sales dashboard. In one comparison the machine learning forecast was 4.5% off actual sales and the human forecast 7.8% off. The model worked from imperfect data.

Full details are in the original sources.

Added 2 Nov 2026 · Updated 2 Nov 2026