# Operations & Data

> What the change means for web and marketing analytics and the martech/CDP operations that feed personalisation.

Digital/Web Analysts and CDP/Martech Ops must make data, attribution, memory and agentic traffic legible without promising precision that does not exist. (Finance, HR, Procurement and Supply Chain remain outside the scope because the corpus does not address them.)

## What changes for operations & data

* **Attribution breaks, but not as usually said.** AI-influenced traffic arrives much less as "direct" (19.9% vs 38.8% of ordinary traffic) and much more as branded search (55.9% vs 40.4%): the user gets the recommendation in chat, then searches the brand on Google. Invisible to a last-click model. You need dedicated AI-referrer channel groups and an upstream measure of visibility in answers. See [The customer journey shortens](https://www.shaped-by-ai.com/en/impacts/what-is-happening/search-becomes-conversation/the-customer-journey-shortens/).
* **2026 official source.** Since 3 June, Search Console offers *Search Generative AI performance* reports — impressions inside AI Overviews, AI Mode and generative Discover, for now impressions only. It integrates monitoring, does not replace it.
* **The data ceiling of memory.** Consent and minimisation against the expectation that "the system already knows who you are" is a martech decision shared with [Product & Design](https://www.shaped-by-ai.com/en/impacts/roles/product-design/). See [Memory and context adaptation](https://www.shaped-by-ai.com/en/impacts/what-is-happening/the-conversational-interface/memory-and-context-adaptation/).

## Responsibilities and actions

1. Define channel groups and attribution windows for AI-influenced traffic.
2. Pair downstream measures with upstream share of presence, stating the limits and surfaces covered.
3. Protect data quality, consent and traceability when automated systems act.
4. Give Marketing and Leadership value scenarios, not just visit or open counts.
5. Define events, states and stable selectors with Product & Design so task success, failures, confirmations, cancellations and audit trails are measurable.
6. Keep agent identity separate from user identity, apply the sanitisation/prompt-injection boundary, and document what the automated system can observe or execute.