Incrementality and signal impact — baseline vs campaign, with contributions on the second tab.
Data for analysis
Full report as baseline and campaign to measure model calibration (expect lift ≈ 0%)
📁
Incrementality Analysis
Requirements: Daily data is required. The baseline report needs at least 75 daily time-series rows (~11 weeks, allows quarter-adjacent ranges that land at 88–89 days); the campaign window can be shorter for in-flight reads. Weekly or monthly aggregates are not supported.
📋Loaded from saved analysis
📊
Lift Metrics
🔬 Self-test (A/A) — Full report modeled against itself. Expected lift ≈ 0%. Deviation indicates model calibration.
Calibration:
+0.0
Total lift (primary KPI)
+0%
Lift Percentage
—
Confidence
0
Days Analyzed
—
Model fit (tier)
📊
KPI Comparison
📈
Daily View
click to collapse
Counterfactual (Old Mix)
Actual (New Mix)
Lift
🎯
Signal impact — contributions
Data for analysis (Light MMM — tabled)
📈
Light MMM (tabled) — form
(tabled) Panel summary not loaded.
(Tabled) Channel selection.
Panel support (tabled)
Channel
Impressions (prox)
Spend (prox)
Sales (mix)
📊
Sales Impact Analysis results
—
Model fit (R²)
0
Days in model
—
Panel sales ($)
—
Sales in media decomposition ($)
Vendor-style readout: Contribution % and Decomposed sales apply the in-sample media share to total panel sales (sums to panel sales when contributions are identified). mROI (prox) is decomposed sales divided by panel spend proxy for that channel — a simple scale readout, not a marginal response curve from a full hierarchical MMM.
Channel contribution % is not shown when the model does not identify positive media effects on sales (see warning above). The chart below shows panel activity mix instead (supporting view only).
Primary chart = attributed contribution % when the model identifies positive media marginals; otherwise activity mix (% of impression proxy across channels) so the slot is never empty — activity mix is not attribution. Diagnostics (below) = in-sample fit and residuals.
Model diagnostics (in-sample)
Appendix — full channel table
Channel
Impressions (prox)
Spend (prox)
Sales (mix)
Contribution %
Decomposed sales ($)
mROI (prox)
Adstock half-life (d)
Confidence
📋 How this model works
In-house geometric adstock + Hill saturation + Bayesian Ridge on standardized features (aligned with incrementality) + block bootstrap — not Google LightweightMMM.
📊
Ready to Analyze
Select two reports with time series data to compare signal mixes using synthetic control.
Baseline = old signal mix (trains the model). Campaign = new signal mix (compared against the counterfactual).
Baseline needs at least 75 days of daily data. Campaign can be shorter.
🎯
Signal impact
Same baseline and campaign as the first tab. Run Incrementality on the first tab first; contribution bars for the campaign period appear here after a successful run.
SignalCraft Assistant
Welcome! I can help you understand your Signal Lift analysis, explain Ridge regression results, and answer questions about campaign performance. What would you like to know?