KPI 03 of 07 · Analytics Phase · Merchant-side · Higher is better
Cross-Surface Attribution Score
The percentage of agent-referred traffic attributable to a specific AI surface
Abbreviation: CSASCSAS is the infrastructure metric that enables all other AXD metrics. Without reliable cross-surface attribution, AIR cannot be measured per-surface, AACR cannot be segmented by originating agent, and DCR cannot be tracked across the agentic landscape. CSAS answers the foundational analytics question: when an AI agent sends traffic to your business, can you identify which agent it was?
Traditional web analytics was built for a world of browsers, cookies, and referrer headers. Agent-mediated traffic breaks many of these assumptions. Agents may not execute client-side JavaScript. They may not pass referrer headers. They may access your API directly rather than loading your webpage. Each of these patterns creates an attribution gap that CSAS measures.
The metric is deliberately simple in its formulation - it asks what percentage of agent-referred traffic you can attribute to a specific surface. But achieving a high CSAS requires sophisticated infrastructure: server-side tracking, UTM parameter strategies designed for AI surfaces, API-based attribution from commerce protocols, and cross-device attribution models that account for the agent-mediated journey.
CSAS is the metric you must solve first. A business that cannot attribute its agent traffic cannot optimise its agent strategy. It cannot know which surfaces are driving revenue, which are driving traffic without conversion, and which are invisible to its analytics entirely. CSAS is the foundation on which all agentic commerce optimisation depends.
How protocols improve attribution
Commerce protocols (ACP, UCP, MPP) include built-in attribution mechanisms. When an agent transacts through a protocol, the originating surface is identified as part of the transaction schema. This creates a reliable attribution pathway that does not depend on browser-based tracking.
Protocol-based attribution is inherently more reliable than referrer-based attribution. It is server-to-server, structured, and tamper-resistant. As protocol adoption grows, CSAS will naturally improve for businesses that have integrated. The remaining attribution challenge will be non-protocol traffic - agent referrals that arrive through traditional web pathways.
Numerator
Agent-referred sessions attributed to a specific AI surface
Denominator
Total agent-referred sessions detected
× 100 = CSAS %
Requires server-side tracking. Client-side-only analytics will undercount agent sessions.
Measurement protocol
Implement server-side tracking that captures user-agent strings, referrer headers, and API request metadata. Build a classification model that maps these signals to specific AI surfaces. Validate the model against known agent traffic patterns (e.g., ChatGPT's documented user-agent string).
Implement UTM parameter strategies for AI surfaces where you have control over the referral pathway (e.g., through protocol integrations or structured data that includes tracking parameters). Monitor the ratio of attributed to unattributed agent traffic over time.
CSAS should be reported as both an aggregate score and a per-surface breakdown. The per-surface breakdown reveals which surfaces your attribution infrastructure can identify and which remain opaque. Focus attribution improvement efforts on the surfaces with the highest traffic volume but lowest attribution rate.
Four levels of attribution capability
<20%
Minimal attribution capability. Cannot distinguish agent-referred traffic from organic. Analytics infrastructure does not recognise AI surface referrals. Operating blind to the agentic layer.
20-50%
Partial attribution. Can identify some AI surfaces (typically ChatGPT via referrer headers) but missing others. Attribution gaps make AACR and AIR measurement unreliable.
50-80%
Comprehensive attribution across major AI surfaces. Can segment traffic, conversion, and revenue by originating agent. Attribution data is actionable for optimisation decisions.
>80%
Full cross-surface attribution with real-time dashboards. Can track the complete journey from agent recommendation through to post-purchase behaviour. Attribution feeds automated optimisation.
What moves CSAS up, down, and sideways
UTM parameter strategy for agent surfaces, referrer header analysis, API-based attribution from commerce protocols, server-side tracking that captures agent identifiers, cross-device attribution models.
CSAS is the infrastructure metric that enables all other metrics. Without reliable attribution, AIR is unmeasurable, AACR is unsegmentable, and DCR is invisible. Invest in CSAS before attempting to optimise other KPIs.
Client-side-only analytics (blocked by agent sessions), absence of server-side tracking, no UTM strategy for AI surfaces, reliance on cookie-based attribution in cookieless agent sessions.
Why CSAS matters commercially
CSAS determines whether you can make data-driven decisions about your agentic commerce strategy. Without attribution, you cannot allocate resources to the AI surfaces that drive the most revenue. You cannot identify which surfaces have high AIR but low AACR. You cannot measure the ROI of protocol integrations.
The commercial value of CSAS is indirect but foundational. It does not directly generate revenue - it enables the measurement and optimisation that generates revenue. A business with high CSAS can optimise its agentic strategy with precision. A business with low CSAS is making strategic decisions based on incomplete data.