Spreadsheet Workbook (Google Sheet): product-interaction-diagnostics

The Decision

Browsing interactions (clicks, views, scrolls) are often read as product demand. The risk is exposure-driven interaction bias: platform mechanics such as product placement and navigation flow get mistaken for genuine customer intent. The question this diagnostic answers: are browsing signals safe enough to guide product push, marketing spend, and inventory promotion?

Ineligible signals trigger a hard stop:

• Browsing interaction data must not influence product or marketing decisions.
• Proving the lack of reliable signals is a valid result for governance.

Stress Testing

The data was subjected to three stress tests to see whether interactions came from product interest or platform mechanics.

Exposure Dominance

Assumption: Interaction volume reflects product preference.

Interaction concentration follows exposure position: 35.5% early, 44.3% late, 20.2% mid. Category navigation shifts the mix, with earlier category pages skewing early exposure while deeper pages skew late.

Insight: Interaction volume is driven by how products are shown, not by what customers prefer.

Session Length Dependence

Assumption: Repeated interactions indicate sustained product interest.

Repeated interactions rise with session depth: 6.6% of short sessions contain repeats, 22.6% of medium, 52.2% of long. Longer sessions also touch a broader set of products (152 vs 213 unique repeated products).

Insight: Repetition is a byproduct of time spent browsing, not product interest.

Coverage Collapse

Assumption: Stable signals exist broadly enough to support decisions.

Only 20.1% of sessions contain any repeated interaction; 79.9% have none. Session mix is short-heavy at 45.7%, with medium at 20.9% and long at 33.4%.

Insight: Potentially interpretable signals exist in a fifth of sessions, not enough coverage for product push decisions.

Recommendation

No product-level push decisions are allowed based on current browsing interaction signals.

• Interaction is driven by placement, navigation depth, and session length, not product interest.
• Cross-product comparison is impossible due to uneven exposure.
• Future analysis must correct for exposure effects before browsing data counts as evidence.

Evidence and Artifacts

• Root diagnostic workbook (Google Sheet)
• Signal evaluations
• Decision gate outcome