Rebuilding Forecasting Models for a Global Crisis

Rapid analytics to survive the pandemic.

Founder's Track Record

Services:Predictive Forecasting & Simulation, Advisory & Strategy

When COVID-19 hit in March 2020, every forecasting model at eBay Classifieds Emerging Markets broke overnight — historical patterns became meaningless as real estate viewings stopped and car dealerships closed. Our founder rebuilt the forecasting infrastructure from scratch, creating "no-COVID" baselines for comparison, weekly-refresh scenario models, and Tableau dashboards that gave CFOs and GMs across 5+ markets the tools to plan budgets and allocate resources through the crisis.

Key Results

Markets Covered

Mexico, South Africa, Poland, Ireland, Argentina — each with market-specific scenarios

Weekly

Forecast Refresh

Models updated every week to capture rapidly changing pandemic recovery patterns

Scenario Tracks

Pessimistic, baseline, and optimistic forecasts per market enabling range-based planning

The Transformation

Before

After

Before

After

All forecasting models broken by COVID

Weekly scenario models across 5+ markets

All forecasting models broken by COVID

Weekly scenario models across 5+ markets

No historical precedent to train from

"No-COVID" baselines for impact comparison

No historical precedent to train from

"No-COVID" baselines for impact comparison

Budget decisions frozen — no data to act on

Weekly reallocation based on recovery signals

Budget decisions frozen — no data to act on

Weekly reallocation based on recovery signals

One global model for all markets

Market-specific scenarios at different recovery rates

One global model for all markets

Market-specific scenarios at different recovery rates

Months to retrain complex ML models

Rapid-prototype Tableau models deployed in days

Months to retrain complex ML models

Rapid-prototype Tableau models deployed in days

The Challenge

COVID-19 invalidated eBay Classifieds' entire forecasting infrastructure in a matter of weeks:

Every existing model was trained on historical data that assumed normal demand patterns — when real estate and auto categories collapsed, the models produced meaningless outputs
CFOs and GMs across 5+ markets needed to make urgent budget decisions but had no reliable forward-looking data to work with
Traditional model retraining was impossible — there was no historical precedent for the demand patterns COVID created, so standard approaches couldn't learn from the past
Each market (Mexico, South Africa, Poland, Ireland, Argentina) was hit differently and recovering at different rates, requiring market-specific scenario planning rather than one global model

Our Approach

**Baseline Reconstruction:**

Created "no-COVID" baseline models by patching historical data to remove pandemic distortion — establishing what demand would have looked like under normal conditions for comparison
Used Google Analytics attribution data as the primary signal source, since web traffic patterns provided the earliest indicators of demand recovery by category and market

**Scenario Planning Framework:**

Built rapid-prototype forecasting models in Tableau that generated multiple scenarios (pessimistic, baseline, optimistic) for each market and category
Designed models for weekly refresh cycles — fast enough to capture the rapidly changing reality, simple enough to maintain without a data science team

**Executive Decision Support:**

Delivered Tableau dashboards to CFOs and GMs showing current performance against "no-COVID" baselines and forward scenarios
Enabled weekly budget allocation reviews where leadership could see which markets and categories were recovering and adjust spend accordingly

The Outcome

**Financial Resilience:**

CFOs and GMs across 5+ markets had weekly scenario-based forecasts throughout the crisis — replacing broken models with actionable planning tools
Budget allocation shifted from frozen (no data to decide) to dynamic weekly reallocation based on market-specific recovery signals

**Organizational Impact:**

"No-COVID" baseline technique gave leadership a reference point to measure actual pandemic impact rather than comparing against meaningless pre-COVID forecasts
Rapid-prototype approach — Tableau models refreshed weekly — proved more valuable than complex ML models that would have taken months to retrain
Forecasting framework built for crisis survival continued as the permanent planning infrastructure for eBay Emerging Markets

Want similar results for your organization?

Get in touch