Revenue Intelligence & Vector Attribution
Understanding which marketing touchpoints drive revenue is critical for optimizing return on ad spend (ROAS). Keystone's Revenue Intelligence platform provides comprehensive attribution modeling, transitioning businesses from legacy single-touch heuristics to high-dimensional vector trajectory attribution.
1. The Revenue Intelligence Dashboard
Navigate to Analytics -> Revenue Intelligence (/analytics/revenue-intelligence) to inspect multi-channel revenue attribution.
Click to enlarge
Key Metrics:
- Attributed Revenue by Channel: Revenue mapped across Google Search, Meta Ads, Email Marketing, Affiliate, and Organic.
- Assisted Conversions: Touchpoints that contributed to multi-session journeys without claiming final credit.
- Attribution Window: Configurable lookback period (7 days, 30 days, 60 days, 90 days).
2. Supported Attribution Models
Keystone allows growth teams to compare attribution models side-by-side to understand channel bias:
| Model | Logic | Best Used For |
|---|---|---|
| First-Touch | 100% credit to the very first discovery interaction. | Top-of-funnel brand awareness & prospecting channels. |
| Last-Touch | 100% credit to the touchpoint immediately preceding checkout. | Direct conversion & bottom-of-funnel channels (Retargeting, Branded Search). |
| Linear | Equal credit distributed across all observed touchpoints. | Broad baseline evaluation across all active channels. |
| Time-Decay | Exponentially weights touchpoints closer to the conversion moment. | Consideration cycles spanning 14 to 30 days. |
| Vector Trajectory (1536D) | Computes geometric cosine similarity between touchpoint vectors and conversion state. | Enterprise multi-touch journeys with high complexity. |
3. Vector-Based Trajectory Attribution (1536D)
Traditional attribution models rely on arbitrary weights (e.g. 40% First, 40% Last, 20% Middle). Keystone replaces arbitrary rules with Geometric Vector Trajectories:
graph LR
subgraph Latent Embedding Space
T1["Ad Click Vector (t=0)"]
T2["Blog Read Vector (t=2d)"]
T3["Product Demo Vector (t=4d)"]
Conv["Order Completed Vector (t=5d)"]
end
T1 -. Cosine Distance .-> Conv
T2 -. Cosine Distance .-> Conv
T3 -. Cosine Distance .-> Conv
- State Embeddings: Every user touchpoint (video view, blog article, search ad click, email open) is transformed into an embedding vector in a 1536-dimensional space.
- Trajectory Mapping: The sequence of touchpoints forms a directed behavioral trajectory through latent space.
- Similarity Calculation: The model computes the directional cosine velocity and alignment between each marketing touchpoint and the terminal conversion point.
- Attribution Weight: Marketing touchpoints that fundamentally altered the user's trajectory toward purchase receive proportionally higher attribution credit.