How Todd Solved Forecasting for Amazon Agencies and Brands

Part 1: The Problem & Building the Model

Todd Vanderstelt describes the frustration that sparked development: "For years, I watched brands and agencies make gut calls on ad budgets and sales goals, knowing that both sides were guessing." A brand's question - what happens to revenue if we reduce TACOS by two points? - became the catalyst for building a mathematically sound forecasting system.

Model Development Process

After testing correlations between Revenue and TACOS, Vanderstelt encountered endogeneity issues and restructured the approach to predict Revenue directly through multiple regression. Key variables tested included Impressions, Traffic, Clicks, Conversion Rate, Ad Conversion Rate, and Spend. The final model simplified to three core predictors: Spend, Trend (Time), and Seasonality.

Monthly multiple linear regression proved superior to more complex alternatives, balancing accuracy with interpretability. The prototype evolved into MixShift's Revenue Forecasting Shift tool, expanding into annual budget planning and reverse-calculation models for determining required ad spend to achieve sales targets.

Key Use Cases

Anchoring Performance: Forecasts establish baselines for identifying extraordinary performance or deficits
Building Trust: Month-by-month accuracy comparisons against actuals increase confidence
Mid-Month Budget Justification: Quantifies incremental sales from additional spending
Model Fit Analysis: Low fit signals external factors; high fit confirms spend/time/seasonality drivers
Ad Coefficient Exposure: Reveals total sales per advertising dollar, exposing true incrementality
Market Optimization: Identifies where ad dollars generate highest efficiency
Saturation Tracking: Monitors declining ad coefficients indicating market saturation
TACOS Balance: Simulates lower TACOS scenarios with quantified sales impacts
Promotion Measurement: Isolates true promotional lift from historical patterns
Seasonality Visualization: Exposes timing and amplitude of demand peaks and valleys
Executive Storytelling: Transforms complexity into clear visual narratives
Annual Planning: Determines ad spend needed for growth targets
Monthly Allocation: Uses seasonality and efficiency data for optimal distribution
Scenario Planning: Models High/Medium/Low outcome ranges
Supply Chain Alignment: Translates sales forecasts into PO decisions
ROI Analysis: Projects payback periods using margin assumptions
Benchmarking: Compares regions and sub-brands side-by-side
Competitive Detection: Identifies unexpected misses signaling competitor action
Launch Evaluation: Measures incremental lift from new product introductions

Future enhancements target repeat-purchase behavior modeling, where delayed customer revenue impacts current attribution. Unit-level forecasting is also in development for detailed PO planning.

Part 2: Why I'm Qualified to Solve It

Professional Background

Vanderstelt's credentials span finance, operations, and strategy roles. At Amazon, he served as COO of the Textbook division and headed the "Hands-off-the-Wheel" forecasting initiative. Subsequently, he founded and led two Amazon Marketplace Services agencies, working directly with hundreds of brands.

Where Other Solutions Fail

Bottoms-Up Overconfidence: SKU-level models compound errors during aggregation, producing volatile forecasts. Top-down modeling at aggregate levels delivers tighter confidence intervals and faster trend adaptation.
Complexity Without Clarity: Machine learning and Marketing Mix Models operate as "black boxes." The forecasting approach prioritizes transparent ad coefficients visible to users, showing exactly how spending drives results.
Disconnected Marketing Reality: Traditional tools ignore advertising relationships while ad platforms ignore seasonality. This unified model connects spend directly to total sales.
Unsustainable Maintenance: Most systems require constant retraining and SKU cleanup. This approach uses standard data already available - revenue and ad spend - updating automatically without manual rework.
Misaligned Business Cadence: Daily/weekly bottoms-up forecasts create noise incompatible with monthly/quarterly planning. Monthly-level modeling matches actual business decision cycles.

Why the Top-Down Approach Succeeds

Starting from total revenue and working backward captures macro relationships more effectively. Advantages include reduced noise, transparent inputs, team alignment across finance/operations/marketing, quick deployment using existing data, and stability through market changes without manual retraining.

The Value of Experience

Most forecasting tool builders come from either data science or marketing backgrounds - working in isolated silos. Vanderstelt's dual experience includes building bottoms-up SKU-level algorithms at scale while understanding practical business constraints. This combination informs designs that are accurate, interpretable, and flexible enough for real-world adjustments, prioritizing shared business truth over mathematical perfection.