The Personalization Industrial Complex

Somewhere in the last decade, the marketing industry decided — with near-religious conviction — that the future of advertising lies in personalizing every message to every individual consumer. The logic sounds irresistible: if we can tailor a message to exactly who you are, what you browsed last Tuesday at 11pm, and what your friends liked on Instagram, surely that message will convert better than a broad one aimed at everyone.

Surely.

Except the evidence for this belief is, at best, inconsistent — and at worst, actively contradicted by a growing body of academic research. We have built an entire infrastructure of customer data platforms, AI-driven recommendation engines, and real-time behavioral targeting systems atop what is essentially an unproven assumption: that more personalization always equals better outcomes.

I’ve spent 27 years in this industry. I’ve sat in enough boardrooms and pitched enough campaigns across markets from Singapore to New York to know what happens when a shiny new capability arrives: we fall in love with the ‘can’ and forget to interrogate the ‘should.’

And right now, the fever-dream of personalization has engulfed the entire ecosystem. Ad-tech vendors sell their latest 'innovations', anxious agencies jostle in with complex retainer setups, while Google and Meta sell privacy-invasive micro-targeting tools requiring constant budget pumps ! Even by our industry's standards — this is the most expensive example of 'adopt first, interrogate later' impulse I have ever witnessed. Almost everyone in this ecosystem profits from more personalization. Everyone except the brand.

Before we go further, let me define the problem precisely — because the industry’s sloppy use of the word ‘personalization’ is itself part of the problem.

When marketers say ‘personalization,’ they are actually conflating two entirely separate decisions, each with its own cost structure, its own evidence base, and its own ROI calculus:

The first is precision targeting — the WHO question. Using data to identify and reach audience segments you could not isolate or reach cost-effectively before. A mid-cap beverage brand reaching lapsed light buyers in a specific metro through first-party purchase data, for instance. This is a media decision.

The second is customized messaging — the WHAT question. Tailoring the creative, the copy, the offer, or the experience for those segments instead of — or in addition to — the broader campaign message. This is a creative and production decision.

The second is impossible without the first. But the first does not automatically require the second.

There are very weighty questions to ask before you do precision targeting and even when it is justified and done well, it is routinely assumed that once you have targeted precisely, you must also customize the message. That assumption is almost never interrogated, almost never costed, and almost never tested against the alternative of reaching the same precise audience with the same broad creative that everyone else sees.

Each of these decisions carries a cost — in money, in production time, and in organizational attention. The discipline the industry lacks is brutally simple: the additional benefit of each layer must outweigh its additional cost. If precision targeting does get you to the audience you need but the broad message would have worked just as well — why are you paying for 500 creative variants? And if neither the precision targeting nor the customization can demonstrate incremental lift over a well-crafted broad campaign — why are you paying for either?

This is not an argument against personalization.

It is an argument against personalization without strategy, without evidence, and without the intellectual honesty to admit that in many cases we are burning cash on a hypothesis that flatters our ego more than it serves our brands.

Let me explain where I think this is working, where it is failing, and why the distinction matters more than our industry is willing to admit.

What the Research Actually Says (And What We Choose to Ignore)

Let’s start with what the personalization advocates rarely cite: the tipping point.

A rigorous 2022 study published in the Journal of Electronic Commerce Research by De Keyzer, van Noort, and Kruikemeier examined how consumers respond to different levels of personalized advertising on social networks. The study used a factorial survey design with 619 participants across four different source contexts. The central finding should give every CMO pause: the ‘sweet spot’ of personalization — the point beyond which you are paying more for diminishing or negative returns — sits far lower than most practitioners assume. Moving from low to moderate personalization significantly increased what the researchers call ‘perceived creepiness’ without a corresponding increase in perceived relevance. And moving from moderate to high personalization? No further increase in relevance whatsoever. The costs went up. The benefits flatlined.

The study found that perceived creepiness was a stronger predictor of consumer attitudes than perceived relevance at every personalization level tested. The perceived costs of personalization outweighed the perceived benefits.

Read through our WHO/WHAT lens, this finding becomes even more damning. The De Keyzer study was testing what happens when you increase the WHAT — more personal data elements layered into the message itself. The WHO was held constant (Facebook users). The result: the customized messaging layer added cost and creepiness but no relevance. The targeting may have been justified; the message customization was not. This is the distinction the industry refuses to make.

This is not an isolated finding. Research by John, Kim, and Barasz from Harvard Business School, published in Harvard Business Review, demonstrated that when consumers become aware that third-party data or inferred information is being used to target them, purchase interest drops by 17–24%. That is not a rounding error. That is a quarter of your potential conversions evaporating because your targeting was too clever for its own good.

Even Forbes contributor Erica Gunn, hardly an industry skeptic, acknowledged the paradox: while Epsilon’s research found 80% of consumers prefer personalized experiences, Gartner found that more than half will unsubscribe and 38% will stop doing business with a company if personalization feels ‘creepy.’ The gap between preference and tolerance is where billions in marketing spend disappear.

But the most damning critique does not come from privacy researchers or consumer psychologists. It comes from the field’s own growth scientists. Byron Sharp and the Ehrenberg-Bass Institute have spent decades demonstrating that brands grow primarily through broad reach and mental availability — by reaching light buyers and non-buyers across an entire category, not by micro-targeting heavy users. Hyper-personalization rests on the flawed assumption that targeting small, high-propensity segments yields the highest return. The data says the opposite: real growth comes from converting the vast periphery of light and non-buyers that personalization, by design, intentionally excludes.

Some practitioners have been more blunt about than others, calling hyper-personalization 'one of the biggest myths in modern marketing'. Their argument cuts to the bone: personalization confuses tactical activation with strategic target selection. The industry has traded sophisticated mass marketing — reaching the broad category with clear, unified messaging — for narrow digital performance traps. And generative AI has not made personalization more effective; it has simply made generating endless variations of mediocrity cheaper.

Producing 500 hyper-personalized variations of a weak idea still yields a weak campaign. It just yields it faster and at greater scale.

Where Personalization Burns Cash (And Nobody Admits It)

Now for the uncomfortable part. Once you separate WHO from WHAT, you can see precisely where each layer fails — and how the assumed bundling of both creates compounding waste.

First, the retargeting absurdity. As Fleischmann put it with admirable bluntness in Inc.: serving someone ads for down pillows five minutes after they already bought down pillows is what passes for personalization today. This is not intelligence. This is an algorithm chasing its own tail with your media budget. And yet retargeting remains one of the largest categories of programmatic spend, sustained by attribution models that credit the last click before a purchase that was going to happen anyway.

Second, the creepiness tax. The De Keyzer study found no positive effect of higher personalization through perceived relevance. Let me repeat that: in the commercial advertising context most relevant to brand marketers, increasing personalization from moderate to high produced zero additional relevance benefit while continuing to accumulate creepiness costs. You are paying more to make people more uncomfortable, and the incremental relevance is imaginary.

Third, the context-blindness problem. The same research demonstrated that personalization effects are profoundly context-dependent. Health, governmental, news, and commercial sources each produced different privacy calculus outcomes. Yet most programmatic buying treats personalization as a universal strategy — the same targeting logic applied indiscriminately across contexts, categories, and audience relationships. This is not strategic marketing. This is a machine doing what machines do: optimizing within the narrow parameters it was given, oblivious to the broader context of how consumers actually experience brand communication.

Fourth, the trust deficit. Kantar’s data showing declining ad attention and persistent distrust of digital advertising tells a story that should alarm anyone investing heavily in personalized digital campaigns. When 62% of consumers do not trust your medium, making your messages more personally targeted does not build credibility — it amplifies suspicion. You are spending more to be more precisely distrusted.

Fifth, the generational warning. Leo’s HumanKind study found that only 13% of people trust that brands have their best interests at heart. Among Gen-Z, the combination of data sensitivity, digital detox trends, and ad-blocker adoption means that the audiences we are spending the most to personalize for are precisely the audiences most resistant to being personalized at. The entire strategy is chasing a demographic that is actively running away from it.

And finally, the opportunity cost nobody talks about. Every dollar spent on ad-tech infrastructure, data management platforms, and personalization engines is a dollar not spent on product innovation, creative development, or brand building. When all marketers use the same platforms’ algorithms in similar ways with content of similar format, the result is not differentiation. It is homogeneity — an arms race in which everyone spends more to achieve the same tepid outcomes.

And now generative AI has poured accelerant on the fire. The promise is seductive: AI can produce thousands of hyper-personalized ad variants at near-zero marginal cost, tailored to micro-personas across every platform and language. But cheaper production does not equal better outcomes. Managing thousands of AI-generated assets introduces what I call ‘variant entropy’ — operational overhead, brand fragmentation, and an exponentially growing surface area for inconsistency, without any evidence that the micro-variation genuinely lifts baseline sales versus a single, well-crafted message.

The honest question every marketing leader should ask of their GenAI-powered personalization program is simple: does this micro-variation genuinely lift incremental sales, or does it merely shift attribution around within existing demand?

So are there any benefits of personalization at all ?

Absolutely !

In this article - I focused on aspects that we all need to be cautious about while going down the personalization pathway.

_______________________________________________________________________________________

In Part 2 of this series, I’ll lay out exactly where personalization genuinely earns its keep, the specific conditions under which it works, and a practical framework for distinguishing signal from noise. I’ll also steel-man the best evidence from the opposing camp and show why even the strongest pro-personalization research supports restraint, not the current frenzy.

But first — I’d love to hear where you think I’m wrong.

Has anyone run a controlled incrementality test on their personalization program versus a broad-message alternative? What did you find?

Previous
Previous

Deploying AI or Being Deployed by AI ? Learning from Singapore

Next
Next

Managing the Personalization Industrial Complex : Part 2