Product performance / 42.60s

Which recommended product was chosen?

Recommendation measurement becomes useful when impressions and clicks remain attached to the product, SKU, placement, strategy, rank and recommendation set.

Silent film with burned-in narrationOpen WebM directly

What You Will See

  1. Product identityName and SKU keep performance attached to something stakeholders recognize.
  2. Exposure and responseImpressions, client-reported clicks and CTR appear together.
  3. Decision contextAverage rank, placement and strategy explain where and how the product was served.
  4. One accepted tupleThe view isolates a specific click by product and recommendation-set context.

Why It Matters

An aggregate CTR without product identity cannot tell a merchandiser what was actually seen or selected.

Product-level facts make instrumentation inspectable while preserving the distinction between a recorded click and a trustworthy human preference.

The System Underneath

Served setProduct, placement, strategy and rank
Clicked eventSet ID + selected product tuple
ClickHouse factsItem-grain event identity
Product viewImpressions, clicks, CTR and recency
Measurement events contain no customer PII. The dashboard joins product identity for reporting.

What This Story Proves

Visible proof

Product-aware recommendation instrumentation appears in the BI dashboard.

Not claimed

The synthetic click proves event capture, not human intent, causal lift or commercial impact.

Read the measurement contract.

AI engineering connectionExecutable feedback constrains autonomous change