Rebuilding attribution from zero
Half of all contacts were landing in ‘Other Campaigns’ and no channel report could be trusted. We rebuilt UTM governance, repaired source integrity, and got every rupee of spend traced to pipeline.
Scaffolding. This case study is a template showing the structure that works. Rewrite it with a real engagement and real figures, then remove template: true from its frontmatter.
- 52% → 97%est
- Contacts with a usable source
- 100%
- Spend traced to pipeline
- −11 hrsest
- Reporting time per month
What was broken
Original Source is written exactly once, at contact creation, and HubSpot never revisits it. That single property was quietly governing every channel report, every attribution conversation, and every budget decision — and roughly half of it was wrong.
The causes were unremarkable and cumulative: hand-typed UTM parameters, four spellings of linkedin, campaign links shared internally and re-shared without tags, and paid landing pages that stripped query strings on redirect.
What I built
- A canonical UTM taxonomy. One fixed set of sources and mediums, documented, with every legal combination enumerated. No improvisation at the point of link creation.
- A validator. A script that rejects a malformed link before it goes out, because a wrong tag cannot be corrected after the fact — the contact keeps it forever.
- A repair pass. Historical contacts re-sourced where the evidence supported it, and explicitly marked unknown where it did not. Guessing would have been worse than admitting the gap.
- One reporting layer. CRM, web and ad platform data unified in BigQuery so the number in the board deck and the number in the CRM are the same number.
What changed
Channel reporting became arguable in the good sense: people disagreed about strategy instead of about whose figure was correct. Two channels that had looked marginal turned out to be carrying pipeline that was previously credited to direct traffic.
The uncomfortable finding: one channel had been cut the previous quarter on the strength of a number that the rebuild showed to be wrong.