Platform Conversions vs. Analytics: A 60% Gap

Start with the arithmetic, because one factor explains most of the gap.

• 40% of 512 conversions being view-through means roughly 205 conversions had no click. Click-based conversions are therefore about 307 - within about 3.5% of the GA4 figure of 318. That single decomposition changes the conversation from "the platform over-reports by 60%" to "the platform counts view-through and click conversions together."

Remaining sources to check, in order of likely impact:

1. Attribution model. TTD credits within its own window and rules; GA4 uses a data-driven model that can distribute credit to other channels and only counts what it can observe in the session. These are not the same measurement.
2. Time zone and date range. TTD reports in America/Chicago, GA4 in UTC. A month boundary can move several hours of conversions between periods - usually a small effect, but it must be aligned before any comparison.
3. Pixel placement. The same conversion pixel firing on both the order-confirmation page and the newsletter sign-up page means TTD may count sign-ups as conversions. If the client's definition is purchases only, the TTD count is definitionally inflated. Check whether separate events exist and whether both are mapped to the optimization goal.
4. Identity and deduplication. TTD's cross-device identity graph can match a conversion to a user that GA4 cannot stitch across devices. Conversely, TTD and GA4 dedupe differently within their windows.
5. Consent and measurement loss. Cookie restrictions, consent banners, and ad blockers affect client-side analytics collection differently from platform-side counting.

A 10–20% difference is normal between a platform and an analytics tool. A 60% difference almost always means a definitional mismatch, and here it is largely view-through plus pixel definition.

Build a reconciled view rather than arguing about which number is right.

1. Align the frame. Pull TTD and GA4 for exactly the same date range in the same time zone. Export both at day level so boundary shifts are visible.
2. Split click and view-through. Report click-through conversions (roughly 307) separately from view-through (roughly 205). Finance should see both, labeled, not a single blended 512.
3. Clean the event definition. Segment by conversion event. If the newsletter pixel is conflated with purchase, re-tag or filter it and re-pull the month. This may reduce the TTD count further and bring the click-based figure closer to GA4.
4. Deduplicate at order level. If an order ID is available, compare unique orders rather than conversion events. This removes double-counting from repeat pixel fires and multi-device sessions.
5. Test incrementality rather than attribution. Run a geo holdout or a PSA/blank-ad control for 4–6 weeks and measure the lift in total orders in treated versus control geos. That answers the question finance actually cares about: did the spend cause sales, not just receive credit for them.
6. Triangulate. Use MMM or a media-mix read if available to sanity-check the direction of the holdout result.

Deliverable: a one-page reconciliation showing TTD click conversions, TTD view-through, GA4 conversions, the explained differences, and the residual. The residual is the part that needs a test, not a debate.

Separate three questions that the client is currently treating as one.

• "What did the platform report?" TTD's 512 includes view-through and possibly newsletter sign-ups. It is a platform-attributed number, useful for in-platform optimization but not a finance-grade sales count.
• "What did analytics see?" GA4's 318 is a session-observed, data-driven attribution number. It is credible for cross-channel comparison but under-counts view-through and cross-device by construction.
• "What did the advertising cause?" Neither number answers this. Only a holdout, geo test, or MMM can.

Recommendation: do not cut spend on the basis of a comparison between two numbers that measure different things. Instead:

1. Set the campaign's optimization KPI to the event the business actually cares about (purchases, not newsletter sign-ups), and fix the pixel mapping.
2. Report click and view-through separately, with the attribution window documented in a measurement plan the client signs off on.
3. Run the geo holdout before the next budget review; if incremental orders per dollar are acceptable, keep or grow spend regardless of the attribution gap.
4. If the holdout shows no lift, cut - and cut on that evidence, not on a reporting discrepancy.

Framing for finance: platform and analytics numbers will never match exactly. The decision should rest on incremental return, with attribution used for in-flight optimization and directional reporting.