Guankeyun
Define the session, period and record type before comparing invitation sources. Distinguish invitation relationships, entrance visits, actual viewing and order outcomes. Compare only explainable fields and retain unknown sources separately rather than guess a channel to fill the table.
Define what “source” means in the system
A source may mean an entrance, inviter information or an existing tag. Guankeyun’s private-community data overview describes inviter information, source tags and audience-source analysis. Confirm how fields arise, update and change in the actual version. Similar names do not make fields directly comparable.
Keep field definitions, extraction locations and ownership. Describe invitation relationships if that is all available, or total visits if only visits are available. Do not automatically call an inviter the sales owner or treat a tag as proof that someone learned of an event exclusively through that channel.
Inspect test paths rather than assume attribution
Use agreed fictional test identities and entrances in the permitted environment and inspect the resulting records. Check what changes when the same identity uses another entrance. This tests product behavior; it should not bypass restrictions or manufacture campaign activity.
First-source, recent-source or other attribution should follow actual product rules. Refer unexplained changes to product or implementation staff rather than choose a convenient interpretation during review. Keep test records separate from production campaign data.
Align time, identities and units
Use matching extraction cutoffs and campaign ranges in a comparison. Explain new versus existing users, visits versus deduplicated people, and livestream versus replay. Different extraction times can explain apparent channel differences and do not directly establish operational quality.
Use people-based metrics only where supported by relevant deduplication and identity handling. Do not promise complete cross-device merging without verification. Figures from different tools can each be useful, but should not be subtracted or divided into a precise conclusion before their definitions are checked.
Separate stages between source and purchase
Sharing or visiting is not viewing, and viewing is not buying. Separately record verified viewing, inquiries and orders. Confirm external store or customer-system matching independently. A working redirect does not establish automatic order attribution across the whole journey.
Before calculating a rate, align numerator, denominator and period. Where a component is missing, retain confirmed records rather than invent conversion. Do not infer spending power, private circumstances or motivation from the source of someone who stops viewing or does not purchase.
Finish with one testable improvement
For low entrance traffic, inspect message clarity, link validity and audience relevance. For low viewing after visits, inspect viewing conditions and content promises. Do not discard a channel on volume alone or add unrelated broadcasts to inflate numbers. Carry evidence gaps into the next campaign as questions to test.
Retain definitions, extraction time, limitations and decisions. Recheck by the same rules next time to understand real change. Source review should explain entry paths and improvement opportunities rather than force every viewer into a single exclusive origin.