Reducing measurement discrepancies
RTB discrepancies - the gap between what one side counts and what the other counts or bills are largely driven by how impressions and clicks are tracked on the client side. To keep them low when integrating supply:
Prefer server-to-server (S2S) counting over browser-based redirect chaining. Browsers can collapse intermediate redirect hops, especially Chrome on Android. So an event that rendered may never be logged. This is the biggest single driver of click discrepancies.
Use a header (
302) redirect rather than a meta-refresh redirect it's more reliable for firing tracking calls.Fire the BURL and impression calls correctly, so billable and rendered events are both recorded. See Billable Event Counting (BURL) for OpenRTB 2.4 / OpenRTB 2.5 for how billable-event counting works and how to reconcile it against "Served" metrics.
Reconcile clicks like-for-like. Each side applies its own invalid-click and multiclick filtering, so compare totals on the same basis (valid + invalid together) before treating a gap as an error.
Be aware that some segments run higher on discrepancy notably video formats, and client-side measurement methods such as Client Hints collection,
sendBeacon, and browser "heavy ad" interventions.
Discrepancy bands
Discrepancies are normal in RTB; the goal is to keep them low enough for clean invoicing rather than to eliminate them.
Under 5%
Acceptable
5–10%
Monitor
Over 10%
Investigate
Troubleshooting workflow
Compare your counts against the partner's and use the Served Ratio to see where the gap opens.
Segment by device and browser certain combinations (notably Android + Chrome) tend to run higher on discrepancy, because of how the browser handles navigation and redirects.
Check the redirect method a header (
302) redirect is more reliable than a meta-refresh redirect.Confirm the BURL and impression calls are firing as expected.
For traffic that crosses multiple networks, reconcile each adjacent pair independently. Alignment between two nodes doesn't propagate across the whole chain.
Last updated
Was this helpful?

