Unlike most other industries, tea traders don’t directly deal with tea manufacturers. Instead, they bid to purchase their tea at the Colombo Tea Auction held each week. This presents them with the challenge of foreseeing the price for a particular category of tea they are interested in purchasing.
Price variations largely depend on the availability of the specific tea type and the total demand at the time of bidding. Therefore, predicting how much a quantity would cost to an acceptable degree of accuracy is difficult.
It is much trickier to handle is the decision of ‘when to buy and how much to buy’ to minimize their overall cost.
Why is making the decision to purchase complicated?
Tea buyers (both local buyers and tea exporters) buy large quantities to manage the demand for the upcoming weeks. This can amount to thousands of rupees per auction. On top of the purchase costs, these companies also must manage their manufacturing and inventory costs for bulk purchases.
Tea sold at the auctions is extremely price-sensitive, especially because one’s own purchasing decisions contribute to the price of a particular type of tea. Unlike in regular situations where economies of scale apply when there is a high demand for a certain grade of tea and the buyer purchases larger quantities the price automatically goes up.
Purchase decisions are also directly associated with that of the Competitor. One misjudged decision can let the competitors take over the market space and leave long term damage to the business.
The market dynamics are hard to understand. It requires years and years of experience to understand the potential market instabilities and what buying decision to make.
What industry-specific parameters are considered?
When manually predicting price variations, industry experts look into the following;
- Next season’s yield
- Foreign market performance
- Local market stability
What does FORECAST Squared do?
FORECAST Squared analyzes historical data and builds forecasting models. These time series models can incorporate external factors that drive the forecasting variable, allowing users to generate customized models that produce highly accurate forecasts for a significantly long-term horizon.
How does FORECAST Squared work?
The system needs to absorb historical data; weekly prices, quantity supplied and the total demand seen at each auction. With the use of Machine Learning algorithms, FORECAST Squared builds models that capture the trends, seasonal patterns, and calibrates the effect of supply and demand on price. These models can accurately estimate the most likely price value for each week by tea category. It also recognizes that the price can still vary due to unforeseen factors. Therefore, it estimates not only the price but also the variability of the forecast.
An added benefit of having the demand and supply is that users can generate different scenarios for comparison. While FORECAST Squared can produce accurate forecasts for demand and supply figures for the forecasting period, it can give visibility to how the price may react under varying conditions, by simply keying in expected values for the supply and demand for future auctions. Running such scenario analysis can help tea purchasers to make data-driven decisions on when to buy and how much to buy.
- Achieve over 85% accuracy in tea prices for any given type of tea
- Predict the monthly supply of tea with 80% accuracy
- Plan purchasing plans across weekly auctions to minimize the buying cost using the insights on auctions prices
- Minimize high inventory buffers and wastage