Revenue Correlation answers a hard question: which of your marketing metrics actually track with revenue? It measures how each metric moves together with revenue across the locations in a project, at several time lags, and flags the relationships strong enough to trust.
Running an analysis
The analysis is computed on demand: choose your filters and press Run. The controls at the top shape what gets compared.
Business and Date range scope which locations and weeks feed the analysis. A wider date range yields more paired weeks, which you need for reliable numbers.
Metrics can stay on All metrics, or be narrowed to the ones you care about.
Lags (weeks) are the time offsets to test: 0, 1, 2, 4, and 8. A lag of 0 compares a metric and revenue in the same week; a lag of 4 compares this week's metric against revenue four weeks later, which is how you spot leading indicators.
The location-weeks counter (locations x weeks) is your sample size. Each cell needs at least about 30 paired location-weeks; below that it shows a dash. If it reads 0 weeks, widen the date range or relax the filters.
Grouping
Grouping decides how locations are combined before correlating:
Within location (default) removes each location's own baseline first, so you correlate the movements of a metric and revenue rather than the fact that a bigger location has bigger everything. This is the best default for a mixed portfolio.
Pooled raw pools every location's raw values together with no baseline removed. Simpler, but size differences between locations can create or inflate a correlation.
Portfolio aggregate sums all locations per week into one portfolio revenue series and one portfolio metric series, then correlates those two. It answers "does the portfolio as a whole move together?" rather than "do individual locations?"
Transform
Levels uses raw weekly values.
WoW change uses the week-over-week change (this week minus last week). It strips out slow trends and seasonality, so it answers "when this metric jumped, did revenue jump too?" Reach for it when everything looks correlated under Levels because all the metrics drift upward together.
Reading the results
Rows are metrics, columns are time lags, and each cell is the correlation, from -1 to +1.
Green is positive: the metric tends to rise when revenue rises.
Red is negative: they move in opposite directions.
Dark border means the cell survives the multiple-testing (FDR) correction; these are the most trustworthy. With many metrics and lags tested at once, some will look correlated by chance, and the border marks the ones that hold up.
Dash means there were not enough paired weeks for that cell.
The Pearson / Spearman toggle switches the coefficient. Pearson measures a straight-line relationship; Spearman ranks the values first, so it catches relationships that move together but not in a straight line and shrugs off outliers. A big gap between the two usually means the relationship is non-linear or driven by a few extreme weeks.
Switch between the Heatmap and Table views (the table lists both coefficients side by side), and select any cell to open a detail with the exact coefficient, its 95% confidence interval, p-value, and sample size. If a data source cannot be loaded for the selected window, a notice names it and its metrics are left out of the grid.
Getting the most out of it
Start with the defaults (Within location, Levels, Pearson) and a date range of at least a few months, so you clear the ~30 paired-week minimum.
Trust the dark-border cells first; treat the rest as hints worth checking.
Use lags to find leading indicators: a metric that correlates with revenue four to eight weeks later is something you can act on before revenue moves.
Flip to WoW change when Levels shows everything correlated, to see which relationships survive once the shared trend is removed.
Compare Pearson and Spearman: agreement is reassuring; a big gap points to outliers or non-linearity.
Remember that correlation is not causation - a strong cell is a lead to investigate, not proof that changing the metric will change revenue.