Event quality drives everything downstream
Platforms optimise toward the events you send. Report a form open as a conversion and you will get more form opens. Report completed, qualified applications and the targeting shifts toward people who complete them.
Identity and context matter
Complete, consented information sent with each event improves attribution accuracy. Partial signals produce confident-looking reports built on guesswork.
Reconcile, do not trust
Compare platform-reported conversions with real applications in your pipeline every week. The gap between them is the most useful number nobody looks at.
Build an event dictionary first
Name every event, trigger, required properties, owner and destination. ‘Lead’ is too vague when one tool fires it on form open and another on completed application. Prefer explicit events such as ApplicationStarted and ApplicationCompleted.
Keep business outcomes separate from interface behavior. Button clicks help diagnose a page; qualified applications and held calls guide budget.
Preserve attribution without inventing certainty
Capture landing page, referrer and campaign parameters on the first visit, then attach them to the borrower record where consent and policy allow. Use platform click identifiers and server events according to each platform’s rules.
Attribution is a model, not absolute truth. Report direct observations separately from modeled credit, and never overwrite the original source merely because a later session occurred.
Send high-quality downstream events
Advertising systems learn from the outcomes supplied to them. A completed qualified application is generally a more useful optimization signal than a page view, and a held appointment or funded outcome can help evaluate quality when volume and policy permit.
Deduplicate matching browser and server events with a shared event identifier. Validate timestamps, currency and values, and do not resend stale events with altered dates to make them appear current.
Reconcile every week
Compare platform events, site events and pipeline records for the same date and event definition. Investigate missing identifiers, blocked requests, duplicates, time-zone differences and stage changes.
Publish a simple discrepancy report: expected, received, matched and unresolved. A stable measurement system makes uncertainty visible instead of hiding it behind precise-looking dashboards.
