Side by side
The decision at a glance
Updated September 2026
| GPS revenue attribution | Native ad-platform dashboards | |
|---|---|---|
| Who built the measurement | Independent: GTM, GA4, and Conversion Clarity, tied to actual revenue | The platform selling the ad space being measured |
| What gets credited | Conversions traced to real leads, calls, and closed revenue | Conversions modeled by the platform's own attribution logic |
| Cross-channel view | One system across every channel and touchpoint | Each platform reports only its own claimed credit |
| Bias direction, per field experiments | Reconciled against actual client revenue | Overstates lift by a factor of 3x or more in ~50% of tested cases |
| Setup effort | Configured before launch, per client, by a specialist pod | Default dashboards, on by default, no setup required |
| What peer-reviewed research validates | Randomized incrementality testing | Observational, platform-style attribution |
| Best fit | Any client spending across more than one channel | Single-channel, low-stakes spend with no cross-channel question |
Who built the measurement
GPS revenue attribution
Independent: GTM, GA4, and Conversion Clarity, tied to actual revenue
Native ad-platform dashboards
The platform selling the ad space being measured
What gets credited
GPS revenue attribution
Conversions traced to real leads, calls, and closed revenue
Native ad-platform dashboards
Conversions modeled by the platform's own attribution logic
Cross-channel view
GPS revenue attribution
One system across every channel and touchpoint
Native ad-platform dashboards
Each platform reports only its own claimed credit
Bias direction, per field experiments
GPS revenue attribution
Reconciled against actual client revenue
Native ad-platform dashboards
Overstates lift by a factor of 3x or more in ~50% of tested cases
Setup effort
GPS revenue attribution
Configured before launch, per client, by a specialist pod
Native ad-platform dashboards
Default dashboards, on by default, no setup required
What peer-reviewed research validates
GPS revenue attribution
Randomized incrementality testing
Native ad-platform dashboards
Observational, platform-style attribution
Best fit
GPS revenue attribution
Any client spending across more than one channel
Native ad-platform dashboards
Single-channel, low-stakes spend with no cross-channel question
01
The dashboard that grades its own homework
Every ad platform reports its own performance, and every ad platform has a structural incentive to report that performance favorably. That isn't a conspiracy theory, it's a measurement problem that peer-reviewed academic research has documented directly, using real field experiments rather than assumptions or vendor claims. The gap between what a platform dashboard claims and what independent measurement finds is one of the most rigorously studied questions in digital advertising, and the findings are not subtle.
The foundational study here is Gordon, Zettelmeyer, Bhargava, and Chapsky's 2019 paper in Marketing Science, 'A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook.' The researchers ran large-scale randomized controlled trials on real Facebook ad campaigns and compared the results against observational methods, the same category of last-touch, platform-style attribution logic that drives most ad dashboards today. The finding: those observational approaches consistently overestimated advertising lift relative to the actual randomized experiment, and in roughly half of the studies examined, the estimated increase in purchases was off by a factor of three or more.
That is not a rounding error. A dashboard reporting three times the actual lift isn't a measurement quirk, it's a systemic bias built into how attribution credit gets assigned when the platform doing the measuring is also the one selling the inventory being measured. The same Gordon et al. paper found that in some comparisons, the number of converters attributed to advertising was overstated by 10 to 100 times relative to what the randomized experiment actually showed, a gap large enough to change every budget decision built on top of it.
- Peer-reviewed field experiments, not vendor claims or blog estimates
- Randomized controlled trials compared directly against platform-style observational attribution
- Overestimation off by a factor of three or more in roughly half the studies examined
- Converters overstated by 10-100x in some observational comparisons, per Gordon et al., Marketing Science 2019
02
The eBay experiment that made this concrete
A second, independently run field experiment reached the same structural conclusion from a completely different angle. Blake, Nosko, and Tadelis's study, published in Econometrica in 2015 as 'Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment,' had eBay turn off its own paid search advertising across 68 U.S. markets to measure the real, causal effect. The result for branded keyword ads, the category that shows the strongest numbers in almost every platform dashboard, was stark: brand-keyword ads showed no measurable short-term benefit. When eBay stopped bidding on its own brand name, traffic barely moved, because those searchers would have found eBay through organic results anyway, at no incremental cost.
That study, formally published in Econometrica, matters here for a specific reason: branded search is exactly the kind of activity that looks flawless on a platform dashboard, high click-through, high conversion rate, strong reported ROAS, because the platform is measuring correlation between the ad and the click, not whether the ad caused anything. The eBay experiment is one of the most cited field studies in advertising economics precisely because it demonstrated, with real revenue at stake and a real live business, that conventional attribution overstates effectiveness even when the reported numbers look pristine, echoing exactly what the Facebook field experiments in Gordon et al. found in a different channel entirely.
03
Why the platforms don't fix this themselves
None of this means Meta and Google dashboards are lying outright. They are reporting what their attribution models are built to report: credit assigned by a last-touch or modeled-conversion framework, not causally verified incremental revenue. The distinction between attribution and incrementality is well established in the measurement field: attribution answers 'what touchpoint gets credit for this conversion,' while incrementality answers 'would this conversion have happened anyway.' As Measured's own decision framework on this distinction lays out, attribution, media mix modeling, and incrementality testing all answer different questions, and treating a single platform's attribution number as the full picture is a category error, the same category error the Gordon et al. field experiments were built specifically to expose.
The platform has no structural reason to close that gap on its own. A dashboard that reports lower, more conservative numbers doesn't help the platform sell more inventory, and there's no independent auditor forcing the correction. That's not a criticism specific to Meta or Google, it's an unavoidable conflict of interest built into any system where the measurement tool and the thing being measured are owned by the same company, which is exactly the structural condition both the Facebook field experiments and the eBay paid search study were designed to test around.
04
What GPS actually measures instead
Conduit's GPS framework exists specifically to close that gap. Before a campaign launches, the pod configures [Google Tag Manager](/glossary/google-tag-manager-gtm) and [GA4](/glossary/ga4) to trace actual site behavior back to source, and [Conversion Clarity](/glossary/call-tracking) to capture phone conversions that ad platforms structurally cannot see at all, since a phone call never reports back into Meta's or Google's own pixel. That combination produces a client-side, cross-channel view of what actually turned into a lead, a call, or closed revenue, reconciled against the client's own numbers rather than any single platform's self-reported credit, the exact reconciliation the Gordon et al. research shows platform dashboards alone cannot provide.
This is the difference between [multi-touch attribution](/glossary/multi-touch-attribution) built independently of the platforms and a set of siloed dashboards each claiming its own share of credit. When Meta and Google each report strong numbers for the same client, in isolation, there is no way to know whether the campaigns are additive or whether both platforms are claiming credit for the same conversion twice, the exact overlap problem the Gordon et al. field experiments were designed to isolate and measure, and that the eBay branded search study demonstrated in the starkest possible terms.
05
A worked example
Picture a client running both Google Search and Meta retargeting to the same audience segment. Google's dashboard reports 40 conversions at a strong ROAS. Meta's dashboard, independently, reports 35 conversions from retargeting the same visitors, also at a strong ROAS. Added together, the platforms claim 75 conversions. The client's actual CRM shows 50 new customers that month. Both dashboards are technically reporting what their own attribution logic assigned, but at least 25 of those claimed conversions are double-counted, credited by two platforms to the same underlying customer, exactly the mechanism the eBay branded-search experiment exposed: a channel getting credit for a customer who was already converting through another path entirely.
A revenue-attribution layer built outside both platforms, tied to the CRM and call tracking rather than either platform's pixel, resolves that discrepancy by tracing the actual path to the actual closed customer. That is the entire function GPS is built to serve, and it's why the reconciliation has to happen at the GA4/GTM/Conversion Clarity layer, not inside either platform's own reporting, which the Gordon et al. study found systematically overstates its own contribution when left unchecked.
06
How media mix modeling fits into the picture
Field experiments like Gordon et al. and the eBay study are the gold standard for causal proof, but running a true randomized holdout on every channel, every month, isn't practical for most agency clients. This is where media mix modeling (MMM) fits in as a middle layer: a statistical model that estimates each channel's contribution to revenue using historical spend and outcome data, without requiring a live experiment every time. As Measured's decision framework lays out, the practical approach most measurement-literate teams take is to treat platform attribution as a generous upper bound, occasional incrementality tests as the conservative lower bound, and let an MMM-style model estimate the number in between, recalibrated whenever a real holdout test is run.
GPS is built around the same logic, just implemented through infrastructure an agency actually owns rather than a standalone modeling exercise: [GTM](/glossary/google-tag-manager-gtm) and [GA4](/glossary/ga4) provide the cross-channel behavioral data, [Conversion Clarity](/glossary/call-tracking) closes the phone-conversion gap, and the resulting revenue picture is reconciled against the client's actual numbers rather than left to whichever platform's dashboard reports the most flattering figure. It's the same corrective the Gordon et al. research argues for, applied at the reporting layer instead of a one-off academic study.
07
Objections agencies raise, and how to answer them
The most common pushback on moving away from platform-native reporting is setup cost: configuring GTM, GA4, and Conversion Clarity properly before a campaign launches takes real time, compared to a dashboard that's on by default. That objection is fair, and it's exactly why GPS treats this as pre-launch infrastructure rather than an afterthought bolted on once a client asks why the numbers don't add up. The second objection is usually that the client already trusts the platform numbers and doesn't want to hear about discrepancies. That's a real risk, but it's smaller than the alternative: the Gordon et al. field experiments show the gap between platform-claimed and actual lift doesn't close on its own, it just surfaces later, usually when a client's own finance team asks why reported ROAS doesn't match revenue on the books.
See how this runs under your brand
Twenty minutes with the pod that runs it. Bring one client and we will tell you if it is a fit.
08
What platform dashboards still do well
None of this makes native dashboards useless. They remain the fastest way to see campaign-level signals, click-through rate, cost-per-click, frequency, and delivery pacing, that matter for day-to-day optimization inside a single channel. For a client running one channel with no cross-channel question and no material revenue at stake, the platform's own numbers are directionally fine for tactical decisions like pausing an underperforming ad set. The failure mode is treating those same numbers as a reliable measure of total incremental revenue, especially once a client is spending across more than one channel, which is precisely where the field experiments above show the overstatement compounds.
09
When platform dashboards are enough
A single-channel client with modest spend, no cross-channel overlap, and decisions limited to within-platform optimization doesn't need a full GPS build. The cost of independent attribution infrastructure should scale with what's actually at stake in the reporting; a client spending a few hundred dollars a month on one platform doesn't need the same rigor as one running five figures across three channels simultaneously, where the bias documented in Gordon et al. has the most room to distort real budget decisions.
10
The agency credibility angle
There's a reputational dimension to this that's easy to underweight. An agency that reports platform-native numbers uncritically, without flagging the known overstatement problem the Gordon et al. research documents, is implicitly vouching for numbers it didn't produce and can't fully defend if a client's finance team pushes back. An agency that reports through independent, reconciled revenue attribution instead is making a claim it can actually stand behind, backed by a measurement approach the underlying academic research supports rather than one the platforms have every incentive to keep generous.
11
What a discrepancy actually looks like when it surfaces
In practice, this problem rarely announces itself as an academic finding, it shows up as a finance team asking why reported ad platform revenue doesn't reconcile with what actually closed. That reconciliation gap is exactly what Gordon et al. measured directly with randomized experiments rather than a guess, and it's exactly what the eBay branded search study demonstrated with real revenue on the line: numbers that look strong on a platform dashboard can be substantially, sometimes dramatically, disconnected from what the ad spend actually caused. An agency that has already built independent reconciliation into its reporting isn't caught flat-footed by that question when it comes; an agency relying solely on platform-native dashboards has no good answer prepared.
12
Why this matters more as budgets scale
The dollar impact of the overstatement documented in Gordon et al. scales directly with spend. A small, single-channel account misreporting lift by a factor of three is a modest optimization error. A client running five or six figures a month across Google, Meta, and a third channel, misreporting at the same rate the field experiments found common, is making real budget-allocation decisions, shifting spend toward whichever platform's dashboard looks best, based on a number the underlying research shows can be off by 3x or more, and in some comparisons by an order of magnitude beyond that. The bigger the account, the more expensive it becomes to keep trusting a number platforms have every incentive to inflate and no structural incentive to correct.
This is precisely why the reconciliation work matters more, not less, as an agency's client base grows. An agency running a handful of small, single-channel accounts can reasonably rely on platform-native numbers for tactical decisions. The same agency managing larger, multi-channel budgets is making increasingly consequential allocation calls on numbers the eBay field experiment and the Facebook field experiments both suggest should not be trusted without independent verification.
13
What actually gets configured, and when
The practical difference between GPS and a default dashboard setup comes down to timing and ownership. A platform dashboard is on the moment an ad account is created, no configuration required, which is exactly why it's the path of least resistance and why so much agency reporting defaults to it. GPS instead configures [GTM](/glossary/google-tag-manager-gtm) and [GA4](/glossary/ga4) before a single ad launches: defining what a real conversion means for that specific client, wiring [Conversion Clarity](/glossary/call-tracking) into the same reporting view for phone leads, and setting up the cross-channel view that the Gordon et al. research shows no single platform dashboard can produce on its own. That upfront work is the entire reason the resulting numbers can be reconciled against a client's actual revenue rather than reflecting whichever platform's attribution model is most generous to itself.
The client-facing version of that data ships under the agency's own reporting system, not a raw Meta Ads Manager export or a Google Ads dashboard screenshot. That distinction matters for the same reason the eBay branded search study matters: a client looking at platform-native numbers alone has no way to know whether what they're seeing is causally real or simply well-attributed. A reporting system built on [multi-touch attribution](/glossary/multi-touch-attribution) and reconciled revenue gives the agency something defensible to stand behind in a client meeting, rather than a number borrowed from a platform with every incentive to inflate it.
14
When independent revenue attribution wins
Once a client spends across more than one channel, or once phone calls are a meaningful conversion path, platform dashboards stop being sufficient on their own. That's also the point where [customer lifetime value](/glossary/customer-lifetime-value-ltv) conversations with the client become impossible to have using platform numbers alone, since no single platform sees what happens to the customer after the reported conversion event, the exact blind spot both the Gordon et al. and Blake, Nosko, and Tadelis studies were built to surface.
This is also the point where an agency's own credibility is on the line. Reporting platform-claimed numbers that later don't reconcile with a client's actual revenue is a fast way to lose trust in the account. Independent revenue attribution, reconciled at the GA4/Conversion Clarity layer before it ever reaches the client report, is what keeps that reporting defensible month over month, not just at the moment a campaign looks good on a dashboard built to make it look good.
For agencies delivering this through Conduit, GPS reporting ships under the agency's own brand: the client sees the agency's reporting system, not a Conduit-branded dashboard or a raw export from Meta and Google. See [white label reporting](/white-label-reporting) for what that actually looks like, and the [GPS](/gps) overview for how the GTM, GA4, and Conversion Clarity configuration comes together before a single ad launches.
