Most sales forecasting relies on one thing: what happened before. Historical data, seasonal patterns, last year's numbers. It's backward-looking by design — and in a fast-moving consumer goods environment where demand can shift in days, that's a problem.
For my MSc dissertation at the University of Sheffield, I wanted to test something different. Could Google Trends — the volume of what people are actually searching for right now — act as a predictor variable for future sales? Not as a replacement for historical data, but as an additional signal. A real-time window into consumer intent before it shows up at the checkout.
I tested it on real FMCG retail data from a major Greek retailer — three years of weekly SKU-level sales across four packaged salad products. Four machine learning models: Support Vector Regression, Random Forest, Decision Tree, and Artificial Neural Networks. Each one trained with and evaluated against Google Trends data as an independent variable.
The finding wasn't dramatic. Google Trends alone won't replace your forecasting model. But it can improve it — particularly at country-level aggregation, and particularly when the keyword selection is right. SVR performed best overall. The margin of error, when measured against a retailer with hundreds of stores, becomes meaningful at scale.
What this taught me about marketing is simpler than the methodology suggests: data is an unbiased truth teller. Search behaviour doesn't lie. People type what they want, when they want it, before they buy it. That signal exists whether you use it or not.
The question isn't whether Google Trends is a perfect forecasting tool. It isn't. The question is whether you're leaving a real-time demand signal on the table while making decisions based only on what already happened.
In my experience — in forecasting, in campaign strategy, in deciding where to focus budget — the discipline is the same. Face the data. Face reality. Then decide.