What This Chart Shows
This chart shows how attribution changed between two time windows and which levers or groups were responsible for the move. A Current Period and a Prior Period are selected via date pickers. The first bar is Prior Total (total attributed lift in the prior window), the middle bars are the change for each marketing activity (current minus prior), and the last bar is Current Total. So you see both direction (who gained vs lost share of impact) and how the totals reconcile across the two ranges.
Key Questions This Chart Helps Answer
- Which channels, campaigns, levers, or taxonomy groups gained or lost the most attributed impact vs the prior window?
- Is the overall modeled contribution up or down between the two periods, and is that driven by a few levers or many small ones?
- How does period choice (e.g. last month vs month before) change the story for the same KPI?
- Do event/funnel/search filters isolate the subset of levers that actually moved?
Axes, Metrics, and Units
Element | Description |
|---|---|
| X-axis | Category labels. |
| Y-axis | Change in Marketing Contribution (or Change in Lift in no-costs mode). Prior Total and Current Total are in the same units (attributed lift); middle bars are Δ = current − prior for that lever/group. |
| First bar (Prior Total) | Absolute starting level: sum of attributed lift in the prior period. |
| Middle bars | Relative change. Green = positive change. red = negative change. |
| Other bar | Sum of changes for levers/groups not in the top 15. Shown only if that sum is non-zero. |
| Last bar (Current Total) | Total level: Prior Total + sum of all change bars. Matches the current-period total for the filtered set. |
Values are modeled. They are not raw reported KPI.
Control Options Reference
Control | Meaning |
|---|---|
| KPI classification / hierarchy / drilldown | Which KPI (target) to use in both periods. |
| Grouping mode | Lever (default): one bar per lever. Series / Channel / Campaign / Funnel / Source type / Topic: aggregate by that taxonomy; one bar per group. |
| Prior Period | Start and end date of the "prior" window (e.g. last month). Required to render the chart. |
| Current Period | Start and end date of the "current" window (e.g. this month). Required to render the chart. |
| Event categories | Filter by specific event categories (include/exclude). |
| Funnel stages | Filter by specific funnel stages (include/exclude). |
| Search terms | Text search on lever/series labels. |
How to Interpret the Results
- Prior Total → bars → Current Total: This is a reconciliation story: middle bars are incremental changes that bridge the two anchor levels (within filtered levers).
- Large positive bar: That lever/group’s attributed impact increased a lot vs the prior window—not automatically “spend up” or “caused sales.”
- Large negative bar: Attributed impact decreased; could be lower activity, mix, model, or real performance—investigate with context.
- Other: Aggregates remaining movers; a big Other means many small changes, not one hidden hero.
- Compare periods fairly: Same length windows and similar seasonality reduce misleading calendar effects (e.g. 28 days vs 31 days).
- No uncertainty bands: The chart does not show confidence intervals; treat small Δ as noisy.
Practical Applications for Marketers
Application | How to use this chart |
|---|---|
| Budget reallocation | Identify groups with the largest positive or negative Δ before moving dollars. |
| Channel / tactic review | Switch grouping mode to see whether channel or campaign narrative matches the lever view. |
| Campaign evaluation | Set Current to in-flight dates and Prior to a pre or control window (design windows thoughtfully). |
| Experiment / validation | Use as hypothesis input; confirm with geo/holdout or lift tests when claiming incrementality. |
| Forecasting / planning | Understand recent momentum in attributed impact vs a baseline period. |
Common Mistakes & Misinterpretations
Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Calling Δ “incrementality proof” | Values are model attribution differences, not randomized causal impact. | Pair with tests or explicit lift studies for proof claims. |
| Ignoring filters | Prior/Current totals on the chart reflect filtered levers when filters apply. | Note active search/event/funnel when sharing screenshots. |
| Equating attribution change to spend change | φ can move with mix, quality, competition, and model updates. | Cross-check media and ops data. |
| Using incomparable periods | Holidays, launches, or different window lengths distort Δ. | Align calendar and business context. |
Caveats & Considerations
- Required inputs: Both date ranges must be set; otherwise: “Select both a Current Period and a Prior Period…”
- Bridge failure: Error figure with initialization message.
- No meaningful change: “No change detected between the selected periods…” when all Δ are ~zero.
- Top 15 + Other: Detail is truncated; tail is rolled into Other.
- No-costs mode: Axis/labels switch to Lift; interpret as lift delta, not ROAS-style contribution unless costs exist.
- Causal language: Use for measurement storytelling and prioritization; high-stakes decisions should combine with other evidence.