In Part 1, I argued that the industry conflates two separate decisions — precision targeting (the WHO) and customized messaging (the WHAT) — and that this bundling is where the waste begins. The academic evidence suggests the sweet spot of personalization sits far lower than most practitioners assume and what most publishers would like to sell. Beyond it, brands pay a tax they don’t know about. This article picks up there: steel-manning the best opposing evidence, identifying where personalization genuinely works, and offering a framework for telling the difference.
The Best Cases The Flag-Bearers of Personalization Can Make
Last week's article drew strong responses — some agreeing, building further and many pushing back. Intellectual honesty demands that I steel-man the opposing argument. To be clear : the personalization advocates are not all ad-tech vendors selling their own wares. Some of the evidence is credible, peer-reviewed, and worth examining carefully. But here is what I notice: nearly every pro-personalization study conflates the WHO and the WHAT. They rarely isolate whether the lift came from reaching the right audience or from showing them a different message. When you read the evidence with that distinction in mind, the picture changes considerably.
Five arguments deserve a careful perusal.
The meta-analysis. A 2025 Journal of Advertising Research meta-analysis of 53 studies found personalized advertising 'generally more effective.' But the headline obscures the detail: covert personalization — where consumers don't know they're being targeted — showed only a small effect size (d = 0.20). Overt personalization performed worse. In other words, effectiveness depends partly on consumers not knowing it's happening — a foundation that is actively eroding as privacy regulation advances. And the meta-analysis bundles WHO and WHAT effects. When you cannot tell which lever produced the lift, you cannot tell which lever to fund.
The Matz psychological targeting study. The 2017 PNAS paper from Columbia and Stanford reported psychologically tailored Facebook ads producing up to 40% more clicks. Headline-grabbing — but Eckles et al. (PNAS) challenged internal validity, noting Facebook's own delivery algorithms confounded the results. Sharp, Dawes, and Bounds from Ehrenberg-Bass went further, arguing the findings actually refute the pro-personalization case once you strip away platform optimization. A WHAT study hopelessly entangled with the platform's WHO intelligence.
The McKinsey revenue claim. 'Companies excelling at personalization generate 40% more revenue.' Correlation, not causation. These companies are also larger, better-funded, and more digitally mature. The report doesn't separate targeting from messaging, or service personalization (Netflix recommendations) from advertising personalization (programmatic ads to cold audiences) — four distinct activities collapsed into one number that tells you nothing about which to fund.
The Oxford challenge. Felipe Thomaz of Oxford has contested Ehrenberg-Bass, arguing that optimizing purely for reach produces 'mediocre outcomes.' Fair challenge. But no serious practitioner argues for reach alone. The question is proportionality — whether the industry's overwhelming tilt toward micro-targeting reflects the evidence. Thomaz's forthcoming work may show moderate targeting improves on pure reach. That is entirely consistent with this article's position. The problem is the industry doesn't want to define or stop at moderate.
The funnel-stage argument. A recent field study examining personalization across funnel stages found that effectiveness 'emerges not from maximizing depth or width in isolation, but from aligning message complexity, targeting scope, and user readiness.' Ironically, this is the most sophisticated pro-personalization argument — and it essentially agrees with this article. It also found that excessive personalization reduces trust, confirming the 'creepiness tax' we documented in Part 1.
The pattern across all five arguments: personalization produces small positive effects under controlled conditions, particularly when consumers are unaware of it, within existing relationships, and at the bottom of the funnel. But no study in this body of evidence demonstrates that customized messaging — isolated from improved targeting — justifies the infrastructure the industry has built around it.
The problem is not that personalization can never work. The problem is that an industry has extrapolated from narrow, conditional evidence to justify universal application — and built a multi-hundred-billion-dollar infrastructure on that extrapolation.
Where Personalization Genuinely Earns Its Keep
I want to be precise here because this argument invites a lazy rebuttal: 'So you're saying personalization doesn't work at all?' No. I am saying it works under narrow, specific conditions that most marketers ignore in their rush to personalize everything.
Personalization works when it solves a genuine problem for the consumer rather than merely demonstrating what you know about them. Netflix, Amazon, and Spotify are the canonical examples, and they succeed for a reason the ad-tech industry systematically misunderstands: they are not advertising. They are service interfaces where personalization reduces friction, saves time, and improves the quality of a product the consumer has already chosen to use. When Spotify recommends a playlist based on your listening patterns, it is completing a transaction you initiated. That is fundamentally different from a targeted Facebook ad that follows you across the internet because you searched for running shoes once.
Personalization also works in contexts where the consumer has an existing, trust-based relationship with the brand and has voluntarily shared information. The HBR research by John, Kim, and Barasz found that when a trusted platform transparently uses first-party data — information the user knowingly provided on that same site — click-through rates increased by 11%, time spent viewing products rose by 34%, and revenue from those products grew by 38%. Those are meaningful numbers. But the operative words are 'trusted,' 'transparent,' and 'first-party.' Remove any one of those conditions and the entire equation reverses.
Luxury brands offer another legitimate use case. When a fashion house uses CRM data to invite a long-standing client to a curated preview matched to their style history and past purchases, that personalization feels like service, not surveillance. The client has an established relationship; the data exchange feels reciprocal. But this works precisely because it is high-touch, human-mediated, and rare. The moment you attempt to industrialize that intimacy — scale it, automate it, apply it to cold audiences — you get the uncanny valley of marketing, where your ad knows too much about someone who never asked for your attention.
The pattern is clear: personalization works within existing relationships, with voluntarily shared data, in service contexts, and when the consumer perceives a clear and immediate benefit. I would estimate that no more than a fifth of the scenarios in which marketers currently deploy personalization meet these conditions.
The 'Publicness Premium': Why Large Brands Cannot Be Built in Private Conversations
Here is the argument that I believe is most consequential and least discussed.
Great brands — Coca-Cola, Nike, Apple, the brands that command pricing power and cultural resonance — are fundamentally public constructs. They derive their value from shared memories, collective experiences, and cultural meaning that exists in the space between people, not inside any individual's browser history. When everyone knows what Coca-Cola means, when the brand occupies a shared position in collective consciousness, that shared understanding is itself the source of brand value.
There is a deeper structural argument here, one rooted in game theory. In behavioral economics, 'common knowledge' occurs when everyone knows X, and everyone knows that everyone else knows X — the concept formalized by Michael Chwe in Rational Ritual (Princeton, 2001). A unified, broad-reach campaign creates exactly this condition. Its power comes not just from you seeing the message, but from you knowing that millions of others saw the exact same message. That shared awareness is itself a signal of brand confidence, scale, and legitimacy. I call this the 'Publicness Premium'
Hyper-personalization, by definition, fragments this shared experience. When every consumer receives a different message, a different creative, a different value proposition — tailored to their individual data profile — you are not building a brand. You are conducting a million separate direct-response campaigns wearing a brand's logo. The cumulative effect is not stronger brand equity; it is the slow dissolution of whatever shared meaning the brand possessed.
Snickers understood this. 'You're Not You When You're Hungry' has run since 2010 — a single, universally relevant human insight executed broadly, consistently, and with creative boldness. No algorithm needed. No customer data platform required. Just a truth about human experience that everyone recognizes and that builds brand meaning through shared cultural participation. IKEA's advertising speaks to all parents without attempting individual targeting. Orange's campaign challenging perceptions of women's sport resonated broadly because its insight was human, not algorithmic.
These are not nostalgic examples from a pre-digital era. They are current proof that broad, culturally resonant communication builds brands in ways that hyper-personalized digital campaigns structurally cannot. WARC's Multiplier Effect study confirms what practitioners have long suspected: blending targeted performance tactics with broader brand equity advertising improves overall sales impact. The key word is 'blending' — not replacing brand building with targeting, which is precisely what many organizations have done under the guise of data-driven marketing.
WPP's Brand Asset Valuator data shows differentiation and relevance scores declining across brands in North America. I would argue this is not despite our investment in personalization — it is partly because of it. We have traded the hard, slow work of building shared meaning — the Publicness Premium — for the easy, measurable comfort of behavioral targeting. And the brands are getting weaker as a result.
A 6-Point Framework: Just Because We Can Does Not Mean We Should
So what should marketers actually do?
The answer starts with a deceptively simple principle: personalized messages should only cater to differences that matter.
Not differences that exist in the data. Not differences that can be detected by an algorithm. Differences that are genuinely large enough, commercially significant enough, and behaviorally distinct enough to justify the additional cost of a separate message — in money, in production time, and in organizational attention. That is a much higher bar than the industry currently applies — and it is the bar that every point in this framework is designed to enforce.
Any marketer presented with a personalization proposal, should work through these six simple thumb rules :
1. Demand / create econometric proof for personalization — separately for each layer. Before committing significant budgets to any personalization initiative, require the same standard of evidence you would demand for any other investment. Not vendor case studies. Not platform-reported metrics like CTR or ROAS, which measure activity within a walled garden, not actual business impact. Demand controlled incrementality testing and marketing mix modelling to prove actual sales lift versus a broad-message / broad-reach alternative. And demand the proof separately: did the targeting produce incremental lift? Did the customized message produce incremental lift beyond that? If the proof doesn't exist today - work with your research division to do a pilot and that creates the proof points
In my experience, most personalization programs cannot survive this unbundled test — which is precisely why it is rarely applied or even planned.
2. Apply the 95:5 rule ruthlessly — it is a WHO discipline, not a WHAT one. Binet and Field's data tells us that roughly 95% of potential buyers are out of market at any given time. Reserve hyper-targeting — a WHO decision — strictly for the 5% of in-market buyers at the bottom of the funnel, where personalization can earn its keep within existing customer relationships built on voluntarily shared data and earned trust. For the other 95% — the vast majority of future buyers who are out-of-market today — invest in broad, unifying creative that builds the memory structures and mental availability that will pre-dispose them when they do enter the market. The 95% are where your future growth lives. Stop ignoring them. And note: even for the in-market 5%, the question of whether a customized WHAT outperforms the broad creative still requires its own proof as #1 above.
3. Distinguish service personalization from advertising personalization. When Netflix recommends a show, that is service — a WHAT that the consumer explicitly invited by subscribing and sharing viewing history. When a targeted ad follows you across the internet, that is surveillance dressed as relevance — an uninvited WHAT driven by inferred data. The former earns loyalty because the consumer initiated the relationship. The latter earns ad-blockers because the consumer never asked for your attention. Personalizing a post-purchase experience, a support interaction, or app functionality where tailored utility is expected is genuine service. Personalizing ad copy based on scraped web history and inferred micro-demographics is not. These are fundamentally different activities and should never share a strategy document, a budget line, or a KPI framework.
4. Reinvest in creative — move budget from the WHAT infrastructure to the WHAT itself. Take a portion of what you are spending on data management platforms and programmatic optimization — the infrastructure of customized messaging — and redirect it toward the thing that actually differentiates brands: ideas. A brilliant creative insight applied broadly will outperform a mediocre message hyper-targeted to perfection. We have known this for decades. We have just allowed the technology to distract us from it. The irony is sharp: the industry has built an enormous WHAT infrastructure (CDPs, DMPs, DCO engines) whose purpose is to vary the message — while underinvesting in the message itself.
5. Audit the cultural context. Personalization is not culturally neutral. The De Keyzer study showed that the same level of personalization produces different responses depending on the source context. The IJFMR research by Srivastav documented significant cultural variation in receptivity to personalization across collectivist and individualist societies. A global brand running the same targeting logic across Jakarta, Stockholm, and Lagos is not being data-driven — it is being data-blind. Both the WHO (which segments are reachable and receptive) and the WHAT (what degree of message customization is welcome versus intrusive) shift with cultural context. A single global personalization playbook is not a strategy; it is an actual oxymoron !
6. Pass the watercooler test — protect the Publicness Premium. Before approving any top-of-funnel creative strategy, ask a disarmingly simple question: can two consumers who saw this campaign discuss it together? If the answer is no — because dynamic creative optimization served them entirely different messages — then you have not built a brand. You have built a targeting exercise. The most powerful advertising in history worked precisely because it created shared cultural moments that people could recognise, reference, and bond over — the common knowledge premium in action. A thousand personalised dark posts will never replicate what one brave, broadly relevant idea achieves.
And finally, have the courage to say no. The most strategic thing a marketer can do in 2026 is look at a personalization capability and say: 'We could do this, but we choose not to, because the evidence does not support it for this context, this audience, and this brand.' That is not 'being a Luddite' - that is discipline the industry needs today. The discipline is the one thing that separates marketing strategy from marketing expenditure.
The personalization industrial complex will continue to grow. The technology will continue to improve. But technology is never a strategy — it is an enabler that requires strategic judgment about when and how to deploy it. Right now, too many organizations are substituting technological capability for strategic thinking, and paying a 'personalization tax' — the hidden cost of doing something because you can, not because you should.
The brands that will win the next decade are not the ones that personalize the most. They are the ones that ask, at every decision gate: does this difference in the data represent a difference that matters — a difference large enough to justify a separate message, a separate cost, a separate claim on organizational attention? And when the answer is no, they will have the discipline to reach that audience with the same bold, culturally resonant idea that reaches everyone else.
That restraint is not a limitation. It is the strategy.