What This Chart Shows
This chart shows ROI (return on investment) or lift by marketing activity. When cost data is available, each bar is attributed ROI. A break-even line at 1 shows whether each bar is above or below the threshold. When costs are not in the model, the chart shows marginal lift per target instead of ROI, with a zero baseline. This allows you to see what is above or below break-even and how precise the estimate is.
Key Questions This Chart Helps Answer
- Which channels or levers are above or below break-even ROI (or zero lift)?
- How does attributed ROI (or lift) compare across marketing activities?
- Which groups should we scale, maintain, or cut based on ROI or lift?
Axes, Metrics, and Units
Element | Description |
|---|---|
| X-axis | Lever or group name (channel, series, campaign, etc.). Up to 10, 20, or all items depending on "Show" control. |
| Y-axis | Costs mode: Attributed ROI No-costs mode: Marginal Lift (lift per target); same units as the KPI. |
| Bar color | Green: > 1.5× break-even (strong positive lift). Light blue: above break-even (positive). Yellow: positive but below break-even (marginal). Red: negative. |
| Break-even line | Costs mode: dashed line at 1.0. No-costs mode: dashed line at 0 (zero baseline). |
Values are modeled. They are not raw spend or reported revenue.
Control Options Reference
Control | Meaning |
|---|---|
| KPI classification / hierarchy / drilldown | Which KPI (target) to use for attribution and ROI/lift. |
| Show | Number of bars shown: Top 10, Top 20 (default), or All |
| Grouping mode | Lever: one bar per lever. Series / Channel / Campaign / Funnel / Source type / Topic: aggregate lift (and cost in costs mode) by that taxonomy; one bar per group. |
| Date range | Filters the ledger to periods inside this window. |
| 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
- Bars above the break-even line (1 or 0): That lever or group is estimated above break-even ROI or positive lift. Green = strong; light blue = above threshold; yellow = marginal.
- Bars below: Negative or below break-even. Red = negative ROI or lift. Use to flag underperformers or reallocation.
- Error bars (when shown): Wider bars = more uncertainty in the ROI estimate. If the interval crosses 1.0 (or 0), the estimate is not clearly above or below break-even. Use with the point estimate, not in isolation.
- Order: Bars are ranked by magnitude (top 10/20 or all). Use "Show" to focus on top performers or see the full set.
- No-costs mode: No ROI or CI; you see marginal lift and a zero baseline. Interpret bars as relative lift, not return on spend.
Practical Applications for Marketers
Application | How to use this chart |
|---|---|
| Budget allocation | Prioritize groups or levers with strong ROI (or lift) and acceptable uncertainty. |
| Channel optimization | Switch to channel grouping; use breakdown hooks where the app exposes them. |
| Campaign / lever review | Lever mode + search to focus on a named set. |
| Risk-aware decisions | Prefer bars where 90% CI sits above break-even when available. |
| No-costs environments | Use lift ranking for relative impact until costs are integrated. |
Common Mistakes & Misinterpretations
Mistake | Why it is a problem | How to avoid |
|---|---|---|
| Treating ROI as incrementality | Attributed ROI is model-based, not a randomized lift read. | Pair with tests/holdouts for proof. |
| Assuming every bar has CIs | CIs apply only to lever + costs + full tensor columns. | Read chart title and presence of whiskers. |
| Comparing across different KPIs/dates | Numerator and denominator change with KPI and range. | Keep controls constant when comparing screenshots. |
| Ignoring “top 20” selection | Worst or small levers may be off-screen. | Change filters or use full exports elsewhere. |
Caveats & Considerations
- Taxonomy missing: Informative annotation for channel / funnel / source_type / topic when no taxonomy.
- Statistical note: 90% CI uses normal approx—treat as model uncertainty, not a substitute for experimental error.