Module: schema.processors.budget_optimization_processor

Defines a processor for budget optimization inference on a Meridian model.

This module provides the BudgetOptimizationProcessor class, which is used to perform marketing budget optimization based on a trained Meridian model. The processor takes a trained model and a BudgetOptimizationSpec object, which defines the optimization parameters, constraints, and scenarios.

The optimization process aims to find the optimal allocation of budget across different media channels to maximize a specified objective, such as Key Performance Indicator (KPI) or Revenue, subject to various constraints.

Key Features:

  • Supports both fixed and flexible budget scenarios.
  • Allows setting channel-level budget constraints, either as absolute values or relative to historical spend.
  • Generates detailed optimization results, including optimal spends, expected outcomes, and response curves.
  • Outputs results in a structured protobuf format (BudgetOptimization).

Key Classes:

  • BudgetOptimizationSpec: Dataclass to specify optimization parameters and constraints.
  • BudgetOptimizationProcessor: The main processor class to execute budget optimization.

Example Usage:

  1. Fixed Budget Optimization: Optimize budget allocation for a fixed total budget, aiming to maximize KPI.

    from schema.processors import budget_optimization_processor
from meridian.analysis import optimizer
from schema.processors import common

# Assuming 'trained_model' is a loaded Meridian model object

spec = budget_optimization_processor.BudgetOptimizationSpec(
    optimization_name="fixed_budget_scenario_1",
    scenario=optimizer.FixedBudgetScenario(total_budget=1000000),
    kpi_type=common.KpiType.REVENUE, # Or common.KpiType.NON_REVENUE
    # Optional: Add channel constraints
    constraints=[
        budget_optimization_processor.ChannelConstraintRel(
            channel_name="channel_a",
            spend_constraint_lower=0.1, # Allow 10% decrease
            spend_constraint_upper=0.5  # Allow 50% increase
        ),
        budget_optimization_processor.ChannelConstraintRel(
            channel_name="channel_b",
            spend_constraint_lower=0.0, # No decrease
            spend_constraint_upper=1.0  # Allow 100% increase
        )
    ],
    include_response_curves=True,
)

processor = budget_optimization_processor.BudgetOptimizationProcessor(
    trained_model
)
# result is a `budget_pb.BudgetOptimization` proto
result = processor.execute([spec])

  2. Flexible Budget Optimization: Optimize budget to achieve a target Return on Investment (ROI).

    from schema.processors import budget_optimization_processor
from meridian.analysis import optimizer
from schema.processors import common
import meridian.constants as c

# Assuming 'trained_model' is a loaded Meridian model object

spec = budget_optimization_processor.BudgetOptimizationSpec(
    optimization_name="flexible_roi_target",
    scenario=optimizer.FlexibleBudgetScenario(
        target_metric=c.ROI,
        target_value=3.5  # Target ROI of 3.5
    ),
    kpi_type=common.KpiType.REVENUE,
    date_interval_tag="optimization_period",
    # Skip response curves for faster computation.
    include_response_curves=False,
)

processor = budget_optimization_processor.BudgetOptimizationProcessor(
    trained_model
)
result = processor.execute([spec])

Classes

class BudgetOptimizationProcessor: A Processor for marketing budget optimization.

class BudgetOptimizationSpec: Spec dataclass for marketing budget optimization processor.

class ChannelConstraintAbs: A budget constraint on a channel.

class ChannelConstraintRel: A budget constraint on a channel.