> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getinteract.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Channel Attribution
> Analyse which marketing channels drive conversions across a chosen date window, using Markov chain and Shapley value attribution.
The **Channel Attribution** page is the main attribution dashboard. Open it from **Insights → Attribution**.
## Selecting a date window
Use the date range picker at the top of the page to set the analysis window. Only customer journeys that **started** within this window are included.
**Quick presets** are available for common windows:
| Preset | Typical use |
| ------------- | ----------------------------------- |
| Last 30 days | Recent performance, fast to compute |
| Last 90 days | Standard operational window |
| Last 180 days | Longer-term patterns |
| Last 365 days | Full-year analysis |
After selecting a range, click **Run attribution**. If results for this exact window (and any active label filters) are already cached, they load immediately. Otherwise, the model runs and results appear within seconds to a few minutes depending on data volume.
Use **Run all presets** in the Runs settings page to pre-compute the standard windows in the background, so the dashboard always loads instantly.
## Label filters
If **label splits** are configured in your attribution settings, a multi-select filter appears next to the date picker. Select one or more segment values (e.g. `NL`, `DE`) to view attribution for that combination.
When multiple labels are selected, each label is modelled independently and results are shown separately or aggregated — depending on the visualisation.
## Dashboard sections
### Channel removal effects
The removal effect chart shows the Markov model's primary output: **how much conversion rate drops when each channel is removed from all journeys**.
* A tall bar means the channel is structurally important — many journeys rely on it
* A short bar means journeys frequently convert even without this channel
* Channels not appearing in any journey during the window show zero removal effect
This view helps identify which channels are essential to the path (often mid-funnel channels) vs. which are replaceable.
### Attributed revenue
The attributed revenue chart distributes total journey revenue across channels according to the credit share from each model.
* **Markov** credit is based on removal effects (re-normalised to sum to 100%)
* **Shapley** credit is based on cooperative game-theory coalition values (only shown when Shapley is enabled)
* **Last-touch** is shown as a reference baseline
Use this chart to compare which model paints a more realistic picture of your channel mix — particularly to identify channels that last-touch over- or under-credits.
### Journey flow (Sankey)
The Sankey chart visualises the most common customer journey paths as a flow from first touchpoint to conversion or dropout.
* **Node width** is proportional to the number of journeys passing through that channel
* **Link width** is proportional to flow volume between two channels
* Journeys that do not convert flow to a "No conversion" terminal node
Use this to understand typical conversion paths, spot where journeys drop off, and identify which channel combinations most frequently lead to conversion.
### Coverage and quality indicators
Two metric indicators appear alongside the charts:
| Indicator | What it means |
| ------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Coverage** | Percentage of journeys where at least one touchpoint matched a known channel. Low coverage (below \~80%) suggests many journeys contain channel values not in your configured channel list — check for naming inconsistencies. |
| **Single-step journeys** | Percentage of journeys with only one touchpoint. Very high values (above \~60%) can affect Markov model quality, as there are fewer multi-channel paths to learn transition probabilities from. |
## Model comparison
Both Markov and Shapley results are shown side-by-side when Shapley is enabled. Key differences:
| | Markov | Shapley |
| ------------------------ | ------------------------------------- | -------------------------------------------------- |
| **Approach** | Removal effect from transition matrix | Marginal contribution across all channel orderings |
| **Sensitivity to order** | High — channel position matters | Low — all orderings are considered equally |
| **Handles synergies** | Partially | Yes |
| **Computation speed** | Fast | Slower (exponential with channel count) |
| **Best for** | Operational decisions, regular runs | Deep analysis, model validation |
When the two models agree on channel ranking, confidence in the results is higher. When they disagree significantly, investigate journey structure: highly sequential journeys favour Markov; channels with strong synergistic effects show up more clearly in Shapley.
## Next steps
Drill into per-touchpoint credit at the individual journey level.
Configure the paths table, models, spend data, and scheduling.