Case Study
Rebuilding Forecasting Models for a Global Crisis
Rapid analytics to survive the pandemic.
At eBay Classifieds Emerging Markets, Arturo rebuilt forecasting models during COVID, helping CFOs and GMs plan amid unprecedented disruption.
Key Results
The Transformation
The Challenge
When COVID lockdowns hit in March 2020, every forecasting model at eBay Classifieds Emerging Markets became instantly useless. Real estate viewings stopped. Auto dealership traffic vanished. The demand curves that had been reliable for years showed patterns that no historical model could predict.
The CFO and country GMs across Mexico, South Africa, Poland, Ireland, and Argentina needed updated forecasts immediately — not for precision, but for survival. How deep would the drop go? When would recovery start? Which categories and markets would bounce back first? The existing models, built on years of steady growth data, couldn't answer any of these questions because nothing in the training data resembled a global pandemic. Finance was making multi-million-dollar budget decisions blind, and every week without a workable forecast increased the risk of either over-cutting (damaging long-term market position) or under-cutting (burning cash during the worst possible moment).
Our Approach
We threw out the old models and built new ones from scratch, optimized for speed and scenario flexibility rather than precision. The first step was creating a 'no-COVID baseline' — we used pre-pandemic trend data to project where each market and category would have been without the disruption. This baseline became the reference point for measuring actual impact and projecting recovery trajectories.
We built scenario models in Tableau that let finance teams toggle between recovery assumptions: V-shaped (quick bounce), U-shaped (gradual), and L-shaped (prolonged depression). Each scenario was calibrated per market and per category, because real estate in Mexico was recovering at a completely different pace than autos in Poland. The models pulled from Google Analytics traffic data, internal listing volumes, and revenue metrics, refreshing weekly.
The rapid prototyping approach was deliberate. Instead of spending months building a statistically rigorous model, we built something useful in days and refined it every week as new data came in. Each Monday, we updated the models with the previous week's actuals, recalibrated the scenario assumptions, and had updated forecasts ready for the Wednesday finance review. This weekly cadence meant the models got progressively more accurate as the pandemic unfolded and real recovery patterns emerged.
We also built market comparison views that showed GMs how their market tracked relative to peers. When South Africa's lockdowns eased but Poland's tightened, the comparative view helped leadership allocate resources to markets showing recovery signals first. The cross-market view was particularly valuable for the CFO, who needed to make portfolio-level investment decisions — which markets to sustain, which to pull back, and which to accelerate based on recovery trajectory rather than gut feeling. Each market's model included a sensitivity analysis showing how budget changes would impact projected recovery timelines, giving finance a quantitative basis for trade-off decisions.
The Outcome
The scenario models didn't predict the pandemic — nothing could. But they gave the CFO and GMs a structured framework for making resource allocation decisions under extreme uncertainty. Instead of guessing, leadership could say: 'Under our base case recovery scenario, Mexico real estate breaks even in Q3 — let's maintain investment. Under the same scenario, Poland autos doesn't recover until Q1 next year — let's reduce spend now and reinvest when traffic signals improve.'
The models supported budget planning across 5+ markets through the most volatile period in the company's history. Weekly refreshes meant forecasts improved continuously as real data replaced assumptions. The 'no-COVID baseline' technique proved particularly useful — it became the standard way to measure pandemic impact and recovery across all markets. Several GMs cited the scenario planning framework as critical to their ability to navigate the crisis without making panic-driven cuts that would have damaged long-term market position. The approach demonstrated that forecasting during uncertainty isn't about prediction accuracy — it's about giving decision-makers a structured way to reason about an uncertain future and make defensible choices under pressure.
Want similar results for your organization?
Get in touchReady to turn data into decisions?
Let's discuss how Clarivant can help you achieve measurable ROI in months.