Data Requirements (Optimization - Price Guidance)

A Price Guidance model needs to fetch data from either a Data Source or a Datamart. The source should contain enough recent transactional history to cover the variability of the business, typically at 1 year of data. Each record should be one transaction line.

Required and optional columns:

Field	Required?	Comment
Metric	Yes	The field the model predicts. Must be numeric or monetary — typically Unit Price, Price Index, Margin %, Discount %, or Unit Margin. Rows where this field is empty are excluded when the transaction data is loaded.
Product	No	Typically Product ID or SKU. Recommended, most pricing patterns are anchored on product but not required for example for use case of configured products, in this case, product attributes are crucial.
Customer	No	Typically Customer ID. Recommended where pricing varies by customer.
Transaction Date	Yes	Used to split training and test windows.
Revenue	Yes	Extended to the quantity.
Quantity	Yes
Profit	Yes	Extended to the quantity. Cost is then derived as `Revenue − Profit`.
Currency	Yes	The model's currency. Conversion is applied for Datamart sources only. For a Data Source, values are assumed to already be in this currency.
Feature Candidates	Yes	Attributes that may influence the predicted metric - categorical (product hierarchies, customer segment, region, etc.) and/or numerical (size, customer revenue, etc.). This is the crucial part, the model will learn the pattern from those attributes, so make sure to include: product attributes that define the product, size, power… customer attributes, like customer segment, type, size … any additional context, like the deal environment, the competition landscape…

The choice of Metric and Metric Type is the most consequential decision when setting up a Price Guidance model. Choose a Metric whose definition matches what the downstream consumer is deciding - for example, the price at quote time, not a post-sale net price that includes rebates or corrections the sales rep cannot anticipate. A model trained on a field that includes later adjustments will predict values that cannot be reproduced at decision time. The underlying data also needs to be consistent: Discount % requires reliable list prices, Margin % requires reliable costs, and so on.