← All case studies
▦Data · ADAData & AI 2023–Present

A trend model built from public data

Reviews, restocks and ranking shifts, turned from internet noise into an early warning system.

90%accurate trend reports
4 of 6signals tracked sales
5categories using it
metrichow closely it tracks sales (r)Metric A0.82Metric B0.71Metric C0.65Metric D0.58Metric E0.22Metric F0.15these two track traffic, not sales10%30%50%Yr 1Yr 2Yr 3Yr 4ShopeeLazadaTikTokthe shift starts heremarket share by quarter (illustrative)
6 candidate signals, tested against real sales
!The problem

Lots of data. Very little signal.

Every review, stock change, livestream and ranking leaves a footprint. Collecting it is easy. Finding the real signal in the noise, across markets, without chasing every data point by hand, is the hard part.

▦What I did

Test every signal against sales.

With the Analytics team, I helped design a market-share product built on trend models. We combined six candidate signals and checked how each one tracked sales.

Only 4 of the 6 really did. The other two followed traffic instead, which was a useful finding too. Then we used models that read each data point’s behavior to find the moments that signal a real shift.

✓What happened

A report leadership actually uses.

The market trend report is about 90% accurate. It’s one of the most-used products by senior management looking for an edge, across five categories:

NutritionF&BPersonal CareHome CarePet Care

Charts on this page use illustrative data. Real client data is confidential.

Built with
Trend modelsPythonMarket analysisReporting