Purpose
Price Guidance is an accelerator designed to predict a price-related metric - Unit Price, Price Index, Margin %, Discount %, or Unit Margin - for each transaction, based on past data and the attributes that explain it. The model also quantifies the impact of each attribute on the predicted value, providing a transparent, auditable view of what drives pricing in the business. From the prediction and its drivers, pricing teams can identify undervalued items, audit inconsistencies across segments, and recommend prices with confidence.
Outputs
The main outputs of a Price Guidance model are a predicted value for a specific case and the decomposition of that prediction into per-feature contributions.
-
Predicted Metric: For each input item, a predicted value of the chosen target metric (e.g., Unit Price, Margin %, Price Index…).
-
Feature and Pairwise Impact: Per-feature contributions showing how each attribute moves the prediction up or down, plus combined effects between pairs of features that neither captures alone.
Approach
Price Guidance starts by checking what are the most important attributes in the dataset in order to recommend the features to use and exclude the least important ones. Keeping the most relevant attributes enables to focus on the key ones and also avoid to increase to much the computation time. Then a glassbox, transparent machine learning model learns the patterns and underlying data structure structure to decompose every single impact to provide valuable insights into pricing behavior - for example, how much customer segment or product category shifts the prices.
Key steps in the approach:
-
Data Preparation: Collect transaction history from a Data Source or Datamart. The source should contain enough recent transactions to cover the variability of the business - typically one year of data. Data enrichment may be required before hand to take the best of the approach.
-
Feature Selection: A model is trained first to estimate the relative importance of each candidate attribute. The accelerator then recommends a subset of features, the ones that together explain most of the variance, and flags duplicates and highly correlated features. The user confirms the final feature list before final model training.
-
Model Training & Testing:
-
Transactions are split into training and test windows, which includes the latest transactions.
-
The machine learning model learns the target metric on the training window.
-
The accelerator measures model performance on the test set and surfaces accuracy metrics for review.
-
-
Impact & Evaluation: The trained model can be inspected through impact dashboards, which show how each feature and each significant feature pair contributes to the prediction. Individual predictions can also be examined for any set of input attributes, showing both the predicted target and the drivers behind it.
Positioning relative to Negotiation Guidance
Price Guidance addresses use cases that Negotiation Guidance cannot cover cleanly, including:
-
Pricing contexts that depend on many product or deal attributes that are hard to capture in a fixed segmentation tree.
-
Configured products like for BTO (Build to Order) or MTO (Make to Order), including numeric values defining the products in great details.
-
Predictions for unseen attribute combinations - a feature-based model generalizes better where a segment-based model has no data.
Limitations
-
Data Quality & Coverage: Accuracy depends on the volume and breadth of historical transactions. Thin coverage in certain attribute combinations weakens the model in those segments. Inconsistent or missing values for the chosen target metric (for example, Discount % without reliable list prices) produce unreliable predictions. Choose the Metric Type that matches the data the business actually maintains.
-
Prediction-Focused (v1.0): Version 1.0 focuses on predicting the target metric and explaining its drivers. Adjacent capabilities - confidence corridors (floor/target/ceiling), identification of undervalued items, manual overwrite with simulation, and alignment / optimization across segments - are on the roadmap for later versions.
-
Integration in Quotes & Agreements: Price Guidance exposes a model evaluation API to query predictions from anywhere in the platform, but the integration with Quotes or Agreements is configuration work outside the scope of this accelerator.
-
No predefined extension point: There is no out-of-the-box extension point. Custom features or behaviors require custom code; once added, the accelerator becomes specific and cannot be updated without extra effort to port the modifications. Specific extension-point requests can be reported to Pricefx.
-
Data requirements: See Data Requirements (Optimization - Price Guidance)