Using AI Forecasting to Generate Suggested Orders |
The existing 'Suggested Orders' functionality in the Ordering app has been enhanced by using Fourth's new AI-powered forecasting engine |
Release date: April 23rd (US) and 24th (EMEA & APAC) 2024
- Enabled by default? - No
- Set up by customer admin? - No
- Enable via support ticket? - Yes
- Affects configuration or data? - Yes
What's Changing?
The existing 'Suggested Orders' functionality in the Ordering app(based on forecast sales) will now use Fourth's new AI-powered forecasting engine.
Reason for the Change
Machine learning, in combination with sales data and external variables such as weather and local events, will improve the accuracy of forecasts and the resulting suggested orders and prep plans.
Customers Affected
Customers who subscribe to Fourth's Performance edition of Inventory, which allows access to Suggested Ordering and Prep Planning.
Due to the inherent changing nature of the sales mix when working with menu cycles, Forecasting is not suitable for customers using the Menu Cycles application.
Release Note Info/Steps
The existing Suggested Ordering functionality is described in this previous release note.
To utilise the new AI forecasting engine, the following settings need to be applied - this will be carried out by Fourth's implementation team. Please raise a request for this to be done.
The following settings will be applied by Fourth's implementation team.
In the Core Inventory Application
- Implement suggested ordering - (this may already be enabled if you have previously used Suggested Ordering with the legacy forecasting engine)
- Use AI Forecast - this will use the new AI engine to generate the forecasts. If set to 'No', the legacy forecasting engine will be used (not recommended)
- AI Forecast Sales Export Start Date - this determines how much historical sales data is used to train the AI models. Typically this would be set to something like 6-8 years ago. The more historical data available the better the forecasts (lost sales during the Covid pandemic will not skew the forecasting)
Fig.1 - Enabling the use of AI Forecasting for Suggested Ordering
Access to 'Suggested Orders' in the Ordering app is controlled by a user-level (Role) setting.
Fig.2 - Enabling a user to be able to see Suggested Orders in the Ordering app
In Recipe and Menu Engineering (RME, Starchef)
Not all ingredients are suitable for Suggested Order calculations so they are omitted by default.
- To include an ingredient in the Suggested Order calculation it should have its Order Type set to SO in RME (see Fig.3)
Fig.3 - Marking an ingredient to be included in Suggested Orders
Forecasting
Once the above is configured the following will happen:
Each night, after Inventory has received the latest sales from the POS provider it will send the sales per sales item, per location to the AI Forecasting engine.
The AI Forecasting engine will generate updated forecasts for every sales item in every location for the next 14 days (see notes below on how the AI forecasts are generated).
Inventory will ingest the updated forecasts, by sales item, which are then "exploded" into the constituent ingredients required to produce the sales item (recipe).
The Suggested Order job will run and will calculate and publish Suggested Orders which will be visible within the Ordering app. See notes below on how the Suggested Order quantities are calculated.
Calculation/Generation of Suggested Orders
When the Suggested Orders job runs it will create a suggested order for each ingredient as follows:
Suggested order quantity = quantity required based on sales forecast plus any safety stock defined less any stock on hand, less pending delivery quantities where:
Stock on hand is calculated as the current stock on hand, less the forecast sales between the order date and the next available delivery date plus any outstanding pending delivery quantity.
In other words: what you have now, less what you are likely to sell before the next available delivery, less any that is already on order and likely to be received before the next available delivery.
For the above calculation, any negative stock on hand will be considered as zero stock on hand.
The pending delivery quantity only considers pending deliveries due 2 days before the order date or less, so does not consider old open orders that have either been physically received but not marked as received or are unlikely to be delivered.
Display of Suggested Orders
The orders generated by the above process will be shown in the Ordering app within the 'Suggested Orders' section.
Here, you will see a suggested order for each supplier and delivery day.
- If placing an order, select the suggested quantities for each item, either by using the right arrow button line-by-line or all in one go by selecting Use Suggested
- + and - buttons can be used to change the quantities if needed
Fig.4 - Item quantities in a suggested order
AI Forecasting Modelling Algorithm
Advances in the field of machine learning over the past few years have enabled businesses to leverage their accumulated data in making better-informed decisions about the future.
Fourth's AI Forecast Engine is powered by machine learning, a branch of artificial intelligence where data sits at the centre of the solution. We are leveraging the latest developments and the use of algorithms that are powerful in automatically learning historical trends and patterns. These algorithms are routinely outperforming expert rules as well as simple time-series algorithms.
The algorithms consider multiple years of data to detect and understand the most predictive factors and their importance for future sales. Multiple data points are created, and outliers (exceptionally high or low sales), and missing data are automatically handled by the engine.
A model is trained for each individual location, allowing the algorithms to be location-specific and acknowledge the differences between types of locations. A competition of multiple algorithms and machine learning models is run, and the best-performing one is selected to produce the forecasts available in the system.
Each day, the engine generates forecasts for fourteen days in the future – from the current day +1 (tomorrow) to the current day +14 (14 days from today).
Data Features
All historical sales data for each location is considered. The algorithm automatically detects the most important patterns (be they weekly, seasonal, or daily trends) which are most predictive of future sales. Multi-year history of sales data is also considered.
The system uses a range of intrinsic and external data to generate the feature space that the models are trained and predicted on. These include:
- Intrinsic: historic lookups, historic error values, rolling stats, etc.
- External: public holidays data, school holidays, Covid indicators
Feature generation is the primary focus for improving the forecasts, so Fourth constantly experiment with adding/changing features to improve performance.
While we are constantly improving the algorithms and data sources, the accuracy of the algorithm can be impacted when external factors influencing the forecast haven’t been available to the engine, or in a situation of highly volatile sales, when it is difficult to detect the signal from the noise.
Comments
0 comments
Please sign in to leave a comment.