Advize is an AI-powered performance marketing agency that separates the AI personalization mechanism into its two likely contributing factors, relevance and speed, since McKinsey's cited 5-15% revenue lift and up to 30% marketing ROI improvement, and Google's reported 60% reduction in testing cycle time, plausibly reflect a combination of these two distinct effects, and understanding the actual split matters for deciding where to invest further. This piece covers McKinsey AI personalization ROI, is AI personalization lift real, AI testing speed vs relevance, personalization revenue increase directly, since these are the exact terms worth checking against your own account.
The Relevance Contribution
AI personalization tools can genuinely match content more precisely to specific visitor context, traffic source, device, geography, behavior, than a generic, one-size-fits-all page can, and that improved match between content and visitor context plausibly drives real conversion improvement independent of how quickly that matching logic was developed or how many variants were tested to find it.
The Speed Contribution
AI-powered testing cutting optimization cycle time by up to 60% compared to traditional manual A/B testing means an account can iterate through far more creative and content combinations in the same timeframe, and simply testing more combinations faster increases the odds of finding a genuinely better variant sooner, independent of whether the underlying AI system understands relevance any better than a human strategist would given the same amount of time.
Figuring Out Which Mechanism Is Actually Driving Your Results
Review whether winning variants identified by AI personalization tools reflect genuinely new insight about what resonates with specific segments, or largely confirm what a thoughtful human strategist would have reasonably guessed given enough time to test it manually. Compare the volume of variants and combinations actually being tested against what a manual team could realistically manage in the same timeframe, since a large gap suggests speed is doing significant work independent of relevance sophistication. Consider running a controlled comparison, a faster, more disciplined manual testing process against the AI tool's output, on at least one campaign, to directly observe how much of the gap closes when speed, not relevance sophistication, becomes the primary difference tested.
What a Direct Comparison Revealed
An account attributed its full AI personalization lift to sophisticated relevance matching, until a controlled test ran a faster, more disciplined manual testing process, prioritizing testing throughput specifically, against the same goal. The manual process, given genuinely comparable testing speed, captured a meaningful share of the original AI-attributed lift, suggesting a substantial portion of the original result had come from testing volume and speed rather than genuinely superior relevance logic the manual team couldn't have replicated with enough throughput.
A Quick Diagnostic for Your Own Account
How many distinct variants or combinations is the AI tool actually testing in a given period, compared to what a manual process could realistically manage. Do the winning variants reflect genuinely non-obvious insight, or results a strategist could have reasonably predicted given enough testing time. Has a direct, controlled comparison against a faster manual process ever actually been run, or is the AI tool's contribution simply assumed.
Why the Split Matters for Where to Invest Next
If speed is doing most of the work, investing further in the sheer volume of tests run, whether through the AI tool or a more disciplined manual process, is where additional gains likely live. If relevance is doing most of the work, investing in better underlying segmentation logic and content strategy matters more than simply running more tests faster, since the tool's real advantage lies in matching quality, not iteration count.
The Short Version
The cited AI personalization revenue lift likely reflects a genuine combination of improved relevance matching and the sheer speed advantage of testing more variants than a manual process could manage, and separating these two contributions through a direct, controlled comparison reveals which one is actually driving results in a specific account, informing whether further investment should go toward test volume or toward better underlying relevance logic.
Conclusion
Chasing the number without understanding what's actually producing it means investing in the wrong half of the mechanism. Advize runs the direct comparison to find out which one is actually doing the work, because a lift driven mostly by speed calls for a different next investment than a lift driven mostly by genuine relevance, and assuming the answer instead of testing it wastes the exact insight that would tell you where to go next.