This is a classic recommendation problem, or a Market Basket. Since you do not have a target column yet, the solution is in how you prepare the data. You will need to work more on the data preparation to create a target column. If I can summarize in steps, I would say this:
1. Create your Target
You need to define a target by using your historical data. Split your data into two time periods. Use the first period (the past) to build your customer features, and use the second period (the future) to identify what they actually purchased. That future purchase becomes your Target Column.
Qlik Predict needs a specific outcome to learn from. You should transform your data so each row is a unique customer. Use a "Point-in-Time" split:
The Past: Aggregate their activity (total spend, frequency, last category) up to a specific date.
The Future: Look at what they actually bought after that date. This becomes your Target Column (e.g., "Next_Product_Category").
2. Engineering Features in Qlik
To give the model the best chance, calculate these attributes for each customer in your script:
Recency: Number of days since their last transaction.
Frequency: Total number of distinct orders.
Monetary: Average or total spend.
Last Purchase: The specific category or item from their most recent order.
3. Running the Experiment
Once your table is ready with the features and the Target Column, load it into Qlik Predict. Select your "Next Purchase" column as the target.
If the data is very simple, you can also use Market Basket Analysis (Association Rules) directly in Qlik using set analysis to identify items that are frequently bought together.
This is a classic recommendation problem, or a Market Basket. Since you do not have a target column yet, the solution is in how you prepare the data. You will need to work more on the data preparation to create a target column. If I can summarize in steps, I would say this:
1. Create your Target
You need to define a target by using your historical data. Split your data into two time periods. Use the first period (the past) to build your customer features, and use the second period (the future) to identify what they actually purchased. That future purchase becomes your Target Column.
Qlik Predict needs a specific outcome to learn from. You should transform your data so each row is a unique customer. Use a "Point-in-Time" split:
The Past: Aggregate their activity (total spend, frequency, last category) up to a specific date.
The Future: Look at what they actually bought after that date. This becomes your Target Column (e.g., "Next_Product_Category").
2. Engineering Features in Qlik
To give the model the best chance, calculate these attributes for each customer in your script:
Recency: Number of days since their last transaction.
Frequency: Total number of distinct orders.
Monetary: Average or total spend.
Last Purchase: The specific category or item from their most recent order.
3. Running the Experiment
Once your table is ready with the features and the Target Column, load it into Qlik Predict. Select your "Next Purchase" column as the target.
If the data is very simple, you can also use Market Basket Analysis (Association Rules) directly in Qlik using set analysis to identify items that are frequently bought together.
