Growth Systems Library
Modeled Conversions
Modeled conversions are platform estimates of the conversions that occurred but were not directly observed — Google and Meta fill in the signal gap left by iOS privacy restrictions using statistical inference, and understanding what to trust in their reported numbers is essential for accurate capital allocation.
Modeled conversions are platform-generated estimates of conversion events that the platform could not directly observe due to privacy restrictions, ad blockers, or browser limitations. Rather than reporting zero conversions for unobserved events, Google and Meta use statistical models — trained on historical conversion patterns, device signals, and aggregate behaviour — to infer how many conversions likely occurred that the platform could not see. These modeled estimates are then added to the directly observed conversions to produce the total reported conversion count.
Exactius treats modeled conversion data with calibrated scepticism — using it as a directional signal while cross-validating against independent measurement sources that the platform does not control.
Modeled conversions matter because they now represent a significant share of reported performance on both Meta and Google. Following iOS 14, Meta estimates that 20–40% of reported conversions in many accounts are modeled rather than directly observed. If the model is accurate, this is a helpful recovery of otherwise lost signal. If the model is systematically biased — which it may be, since the platforms have an incentive to show strong performance — it inflates reported ROAS and creates a false impression of campaign efficiency.
The capital allocation risk: if you are making budget allocation decisions based on platform-reported ROAS that includes a significant modeled component, and that model overstates performance, you are systematically over-investing in channels that are performing worse than they appear. This is the mechanism behind many post-scaling efficiency collapses: the reported metrics looked good, scaling decisions were made, and then the 'real' performance eventually surfaced.
Assessing modeled conversion reliability: compare platform-reported conversions against your own analytics (GA4, Shopify, or backend purchase records) for the same period. A discrepancy of 10–20% is typical due to attribution window differences and de-duplication. A discrepancy above 30–40% suggests the platform's modeled component is significant and may not be reliable. Track this ratio consistently over time: if it widens after a budget increase, the modeled contribution is growing and the data quality of your scaling decisions is declining.
Cross-validation framework: Media Efficiency Ratio (MER = total revenue ÷ total ad spend) is platform-agnostic and cannot be manipulated by modeled data. Compare your MER trend against your platform-reported ROAS trend. If MER is flat or declining while ROAS is rising, modeled conversions are likely inflating the platform's reported numbers. Use incrementality testing as the ultimate validation: the holdout group conversion rate is immune to platform modelling because it is measured independently.
Exactius does not make capital allocation decisions based on platform-reported ROAS alone — precisely because of the modeled conversion problem. The Growth Operating System, developed by David Manela, uses MER as the primary allocation signal (platform-agnostic, model-proof) and incrementality testing as the validation mechanism. Platform-reported data, including modeled conversions, is used directionally — for creative testing and campaign structure decisions where independent validation is impractical at the required cadence.
The practical implication: if your business is growing, modeled conversions are an acceptable trade-off — imperfect signal is better than no signal for algorithm optimisation. If your business is trying to determine whether a specific channel is worth its budget, modeled conversions are not a reliable basis for that decision. Exactius embeds growth operators who maintain an independent measurement layer alongside platform reporting, so the two signals can be continuously compared and discrepancies flagged before they propagate into scaling decisions.
→ Learn more about the Growth Operating System at davidmanela.com/frameworks/growth-operating-system
What are modeled conversions in Google Ads?
Modeled conversions in Google Ads are Google's statistical estimates of the conversions that occurred but could not be directly measured due to privacy restrictions — primarily from users who declined consent in the EU under GDPR, or from Safari and Firefox users where third-party cookies are blocked. Google uses Consent Mode and machine learning models to infer conversion rates for unobserved users based on observable signals. These estimates are added to directly measured conversions to produce the total reported conversion volume. The proportion of modeled conversions in a Google Ads account varies by geography (higher in EU due to consent requirements) and by audience composition.
Can you trust modeled conversion data for budget decisions?
Modeled conversion data can be trusted directionally — as an input to platform optimisation algorithms and as one signal among several for relative performance comparison. It should not be trusted as the sole basis for absolute budget allocation decisions — specifically, whether a channel deserves more or less of your total media budget. The risk is that platform models may be systematically biased toward showing better performance (platforms have an incentive to attribute conversions to themselves). Cross-validate platform-reported conversions against your own analytics and MER trend before making scaling decisions. Where the two signals diverge significantly, trust your own analytics.
How much of my Meta ROAS is modeled?
Meta does not disclose the exact proportion of modeled conversions in individual accounts, but the industry estimate is that 20–40% of reported conversions in many accounts are modeled rather than directly observed, particularly for accounts serving significant iOS traffic without CAPI implemented. Improving your CAPI implementation reduces the modeled proportion — because more conversions are directly observed server-side and less estimation is required. Checking your Event Match Quality score in Events Manager and comparing Meta-attributed conversions against your backend order count gives you an indirect estimate of how much of your reported volume is modeled versus directly measured.
Exactius Growth
Your growth system is either compounding or degrading.
Book a diagnostic call. We'll identify where your growth system is breaking.
Book a call