Google Said 2,267 Orders. The Real Number Was 2,684.
Confidential Retail Brand, an established, multi-region B2B/B2C retail brand running large-scale Google Ads campaigns across several countries, with a customer buying cycle that can stretch anywhere from two weeks to six months between orders.
Permanent identity tracking that stitched orders to first-touch regardless of time elapsed.
Shared the verified number up the org to rescope Google's attribution window.
The Setup
Before the dashboard, this brand's marketing team had no visual way to separate new customers from returning ones. Whatever reporting existed showed end-result purchases and not much else: no clean read on new-visitor volume, no way to tell if a campaign was actually bringing in new people or just re-monetizing the same buyers who were already coming back on their own.
Since tracking began, every visitor to the site has been tagged with a permanent identity cookie the moment they land, whether they convert that day or not. That single change made it possible to answer a question the team had never been able to answer cleanly before: who's actually new, and where did they come from.
The Break
Walking through the new dashboard live, the brand's marketing team noticed something that didn't match the usual story. Every other reconciliation case study in this category runs one direction: the platform claims more than what actually happened. Here it ran the opposite way. For the same new-customer window, the verified dashboard showed 2,684 real orders, more than the roughly 2,267 orders implied by Google's own reported conversion value. One of the marketing stakeholders on the call caught it immediately: "It's quite a big difference on revenue between what Google says versus what this says." For once, the platform hadn't inflated its own number. It had missed real orders entirely.
The Diagnosis
This is the same mechanism behind every case study in this series, a platform's own attribution logic drawing a hard line where the real customer journey doesn't draw one, just running in the opposite direction. Google's attribution window credits a conversion back to a campaign only if it lands inside a fixed window from the click. A customer might see an ad, browse, and not come back to actually order for weeks or months later, well inside a normal buying cycle for this business, but well outside the window Google is willing to credit. When that customer finally buys, Google's own system has already moved on. The order happens. Google's own report never counts it.
The permanent cookie doesn't expire on Google's clock, so it never runs into that problem. If someone's first touch was tracked in one month and their completed order lands months later, the system stitches that order back to the same person regardless of how much time passed in between, which is exactly why the verified number came in higher than Google's, not lower. Nothing was inflated. Google was just structurally blind to orders that took longer than its own window allows.
The same walkthrough surfaced supporting numbers that reinforce the same point: cross-device identity stitching running at 98-99% ("less than 2% variance," well above the 90%+ threshold most attribution setups consider strong), 74% of non-paid new customers had a paid touch somewhere in their journey even when it wasn't their first or last click, and 23% of new customers convert more than 30 days after their first visit, a meaningful share of revenue that a platform's own shorter attribution window would simply never see.
Where It Stands
The reconciliation gap is now visible and explained, not just noticed. The dashboard is being shared further up the org for review. A handful of refinements are already in motion off the same call: a device-level breakdown surfacing the cross-device stitching stat directly, campaign drill-down by Google's own "Goals" column, and a fix to the target-CAC/margin input fields, which weren't triggering a recalculation. Additional ad-platform and international-region data connections are still pending on the client's side, which will only deepen the picture once connected. The regions already fully connected are tracking well.
Why It Matters
The category's whole premise is telling you the real number, whichever direction it points, not just "we catch platforms overclaiming." A platform that can quietly overclaim can just as easily quietly undercount. Both are just what happens when a model draws an arbitrary line and calls it the truth. This brand needed a number not bound by any platform's own clock, not to be rescued from a broken one, so a real order gets counted as a real order no matter how long the customer took to come back.
That's only possible with a data layer the brand owns permanently. A rented tool's attribution window is set by the vendor's own product roadmap, not the brand's actual buying cycle, and there's no version of 'rent a slightly longer window' that fixes a structural blind spot like this one.