YUMBI has introduced a sales forecast feature, now available on the Performance Graph in the Store Management Dashboard Report. This feature leverages a machine learning model that analyzes the past 26 months of sales and related data to predict daily revenue for the current month.
The model is currently in its beta phase and will be refined over time for improved accuracy. (See Figure 1)
Figure 1
How the Forecast is generated on the current model:
The forecast is based on sales revenue data along with the following key factors:
Pricing & Transaction-Level Features
Pricing and transaction data drive the core of revenue forecasting. By monitoring changes in average item prices and the number of items per order, the model can assess the direct impact of pricing strategies on total revenue. When prices fluctuate or order sizes shift, forecasts adjust accordingly to reflect anticipated changes in sales volumes.
Customer Engagement & Demand
Understanding customer behavior is crucial in predicting future sales. Key variables that help the model anticipate transaction volumes and upselling opportunities include:
- Active Customers: The number of customers who place orders within a given period.
- Web Traffic: Metrics such as average daily users
- Customer Sentiment: based on ratings provide insight into customer satisfaction, influencing repeat purchases and long-term demand.
Marketing & Promotions
Marketing efforts play a significant role in revenue fluctuations. Factors such as:
- Store & Brand Level Campaigns: Promotional activities can drive temporary spikes in sales.
- Advertising Impact: Targeted marketing initiatives contribute to increased visibility and customer engagement. While these variables may rank slightly lower than pricing or demand data in long-term forecasting, they provide valuable insights into short-term revenue surges.
Seasonality & Calendar Effects
Consumer purchasing behavior often follows predictable seasonal trends. To account for these fluctuations, the forecasting model considers:
- School Holidays & Public Holidays: Increased foot traffic during breaks impact demand, especially in the fast-food industry.
- Payday & Month-End Trends: Many customers make larger purchases around payday, influencing revenue patterns.
- Regional & National Events: Province-level school closures and public events contribute to variations in customer traffic.
By analyzing these factors collectively, the forecasting model generates precise revenue estimates. Observing interactions between broad seasonal trends, local marketing initiatives, and customer engagement allows businesses to optimize planning, staffing, and resource allocation across all stores.
Understanding and leveraging these insights ensures that businesses remain agile, adapting to shifts in demand while maximizing profitability.
Understanding the Historical Forecast Accuracy
The historical forecast revenue for your store can fall into three reliability states:
- Reliable (Green) → Historical forecast error < 10% (highly accurate).
- Caution (Orange) → Historical forecast error 10% - 20% (moderate deviation).
- Uncertain (Red) → Historical forecast error > 20% (significant deviation).
NB: The historical forecast error is calculated as the percentage difference between the forecasted and actual daily revenue.