AI Sales History
When you introduce a new product, there is an immediate forecasting challenge: there is no historical sales pattern to forecast from.
AI New Item Forecasting helps solve this cold-start problem by using the sales patterns of similar products already in your assortment to create an AI-generated sales history for the new item.
That history gives the forecasting engine a starting point for purchasing and replenishment from day one. As the product begins generating real demand, AGR automatically calibrates the AI-generated history against its actual performance.
The goal is simple: make better initial inventory decisions, maximise sales opportunities and reduce the risk and cost of carrying the wrong amount of stock.
Beta feature: AI New Item Forecasting is currently in Beta. We recommend reviewing the generated sales history, similar items and resulting forecast as part of your normal planning process.
Why use AI New Item Forecasting?
Without a meaningful demand forecast, it can be difficult to determine how much stock to purchase or replenish. Ordering too little can result in lost sales, while ordering too much can tie up capital and increase the risk of excess stock.
AI New Item Forecasting gives you a powerful starting point by answering:
"Which products in my existing assortment are most similar to this new item, and how did those products sell?"
AGR uses the answer to create an artificial sales history that feeds into the automatic forecasting engine.
How to get started
Contact your CSM to enable the AI Beta feature, through the chat or direct e-mail.
Once it has been enabled on your setup, you can activate the feature on individual items or in bulk for multiple items.
For an individual item
Open an item and go to:
Item Info → Forecast Settings → AI Sales History
Enable the setting and review the similar products AGR identifies.
For multiple new items
A useful workflow is:
Add the AI Sales History column to your item list.
Filter using First Sale Date to identify new or early-stage products.
Select the relevant items.
Use Bulk Update to enable AI Sales History.
Review a sample of the items after processing.
Check the generated history, similar items and resulting forecast.
You can disable AI Sales History again at any time.
How a new item with no sales history and no stock or undelivered looks before and after
How do I find AI Sales history items
To surface an item list of all items using AI sales history, use the column and/or filter: AI Sales history and find items with Yes
How it works
AGR uses an AI model built with Azure OpenAI to compare information about the new item with products already in your assortment.
The comparison uses product information such as:
Item name
Item description
Supplier/vendor
Cost Price
Product categorisation
AGR searches for up to 10 similar items within the same location that have usable sales history.
To keep the comparison as relevant as possible, AGR first searches within Item Group 2.
If there are not enough suitable items available, the search is broadened:
Item Group 2 → Item Group 1 → All items within the location
AGR excludes items that would not provide a useful demand signal, including:
Special-order items
Items without sales history
Each similar item receives a similarity weight. Products that are more closely related to the new item have a greater influence on the generated history.
This means AGR does not simply copy the sales history of one product. It combines demand patterns from multiple relevant products.
From this AGR generates up to 25 months of an artificial sales history reflecting both:
Demand level — approximately how much this type of product would sell.
Demand shape — how demand could change over time, including seasonal patterns.
Adapting to real sales
As real sales start arriving, AGR automatically uses them to calibrate the generated history.
After the item has completed at least one full forecasting period — one full month for monthly forecasting or one full week for weekly forecasting — AGR compares the item's actual demand with the AI-generated baseline.
Example
Imagine the similar products suggest demand of approximately 50 units per week, but the new item actually sells 100 units during its first complete week.
The new product would automatically scale to 2× the expected level.
As more real sales history becomes available, the forecasting process increasingly has the item's own demand information to work with.