In the United States, understocks, overstocks and returns cost retailers a staggering US.75 trillion a year.
Consumer goods companies in China face similar if not bigger challenges, given the rise of mega promotional periods like the 11-11 and 6-18 shopping festivals.
One retailer trying to tackle this head-on is Bear Electric, a leading Chinese home appliance brand. It’s becomingly increasingly difficult to predict sales volume during seasonal promotion periods, according to Yifeng Li, Bear Electric’s general manager.
The power of data
Just a few years ago, with limited access to sell-out and third-party data, it was incredibly difficult for brands to accurately predict future demands. Yet, the rise of mobile internet has led to a proliferation of data. Brands are now able to gain better visibility of sell-through rate and the true driver of sales. The rise of machine learning and AI has also allowed businesses to mine and analyse data in a more systematic way. However, sifting through and analyzing huge amounts of data can be challenging, especially for companies who don’t have an in-house data team.
Which is where Artefact comes in.
Our squad of data scientists, data engineers, data consultants and marketing experts help clients develop demand prediction models using big data and AI technologies.
We recently built a forecasting model for an O2O platform in China, allowing its two groups of users – merchants and consumers – to better connect with each other.
Many of the platform’s merchants suffered from unstable sales volume, resulting in a waste of resources, and failure to support rising demand. To rectify this, we worked with the client to provide individual merchants with daily updates of sales volume prediction for the forthcoming two weeks to help them plan inventory, operation and campaigns.
Better prediction, less waste
You cannot make bricks without straws. Similarly, you cannot predict volume without historical data. One challenge that many merchants face was missing sales data. Indeed, many merchants who’d just set up their business, or who operated sporadically in the past, had never experienced an 11-11 or 6-18 shopping event. Without this data, merchants have little or no overview of the amount of stock they need for promotional periods, potentially leading to a huge blow to revenue.
The question then becomes: how should one predict sales volume, especially for big promotional events, with insufficient historical data?
Our solution was a multi-pronged approach, where we leveraged our expertise in data mining and analyses to help merchants ‘fill the gaps’ during special sales periods. While a classic machine learning model is used to predict sales volume for non-promotional periods, we created a prediction model allows us to combine the merchant’s own historical data with data from other merchants of similar profiles to predict sales volume for promotional periods. This new model improved the platform’s forecasting accuracy by 20%.