A filmed conversation published on July 3, 2026 has Ogilvy vice chairman Rory Sutherland arguing that marketing measurement rewards whatever is cheap to count and quietly assigns a value of zero to everything else, from word of mouth to the reversion rates that decide whether a new technology sticks at all.

The video, titled "Rory Sutherland: Good marketers watch what people do | SaaSy AF", was posted to the YouTube channel of Aaron Gibson, founder and chief executive of Hurree. According to the channel description, the discussion was recorded in May 2026 in front of a live audience, and covered what modern marketers build their ideas around, why stated and observed behaviour diverge, and how artificial intelligence and data fit into that. By the time the page was captured, the recording had drawn 5,350 views, 94 likes and 15 comments from a channel with 380 subscribers. The description also noted that both participants planned to attend MAD//Fest London in July.

Sutherland covered a wide field over roughly 49 minutes. The through-line was narrower than the range suggests: almost every example returned to the same defect, which is that organisations construct their models from whatever data happens to be lying around, then treat the resulting picture as complete.

The gap between what people say and what they pay

The clearest case Sutherland gave concerned environmental claims. He described a figure familiar from consulting presentations, that 75% of consumers say they will pay extra for a product that is kind to the environment. According to Sutherland, that number is wrong by roughly a factor of nine. The real proportion, he said, is about 8%.

Tons of decisions have been taken on the basis of the higher figure. And the error does not stop at the headline number, because the framing itself carries a second-order penalty. Sutherland argued that a large group of consumers will infer from an environmental claim that a detergent has been compromised on cleaning power. The claim intended to add value subtracts it.

His agency's behavioural science practice reached an awkward conclusion from cases like this: that virtue is sometimes better delivered without disclosure. The first problem the practice was handed, roughly 13 years earlier, involved a Belgian biscuit. A low-fat version had been launched and sales had collapsed, yet blind tastings showed no detectable difference from the original. Sutherland recounted the diagnosis he offered the agency in Belgium: "You didn't put low fat on the packaging, did you?" The manufacturer had spent heavily reducing fat content and wanted credit for it.

Sutherland extended the pattern to low-calorie ice cream, which he said has not really succeeded outside the United States despite being, on paper, an ideal product. Some categories appear to require an element of guilt to function. The distinction between stated preference and revealed preference, he said, "is sometimes absolutely massive."

That distinction has direct consequences for how generative systems are used. What concerns him about large language models, he said, is "the language bit, which is what people say and what people do." A model trained on declarations inherits the declarations.

Where the data stops, the model stops

Sutherland reached for the physicist Murray Gell-Mann to describe why marketing resists engineering treatment. Told that someone wanted to discuss economics, Gell-Mann is said to have replied: "imagine how difficult physics would be if atoms could think."

The operational version of that problem is mundane. Physics and engineering work because there is a number for everything that matters. A marketplace contains forces that nobody measures, word of mouth being the obvious one, and unmeasured forces get treated as zero. Sutherland compared the omission to attempting a lunar trajectory without accounting for gravity. Models get assembled from convenient inputs rather than relevant ones.

He also flagged a failure mode specific to trained systems, citing the economics writer Tim Harford. A classifier trained to recognise animals was shown white noise and reported an animal, because it had never encountered an input that was not an animal. It optimised for the most plausible animal in a field of static. Sutherland connected this to human confabulation, the same tendency that produces faces in clouds.

His position on the technology is not uniformly sceptical. He said he had been considerably more sceptical until roughly six to twelve months before the recording, and had changed his view after the models started "making points I never would have thought of" and reaching for analogies he would not have reached for himself. One of those analogies became a central part of his argument.

Convergent evolution and the quarterly target

The analogy concerns vultures. New World vultures, found in North and South America, and Old World vultures, found in Africa and parts of Europe, look nearly identical: bald necks, highly acidic stomachs, the same unlovely silhouette. They are not closely related. According to Sutherland, the Old World birds descend from hawks and the New World birds from cranes, and their resemblance is a product of both lineages being optimised for the same task, which is eating decomposing flesh.

Businesses, he argued, are being subjected to the same optimising pressure by financial intermediaries. Capital itself is diverse, with widely varying time horizons and risk appetites. Intermediaries who must justify their own existence month by month need comparability instead, so they apply identical metrics across every company in a sector. The result is convergence, and convergence pushes firms into head-to-head price competition on identical terms.

The cost of that pressure shows up in a specific statistic he cited: 80% of managers in very large organisations will cancel a project they are confident will be very profitable, purely to meet a quarterly forecast or target. He invoked Goodhart's law, the observation that a measure adopted as a target ceases to function as a measure, and argued that investors, businesses and consumers all lose, the last through reduced choice.

He drew a related argument from Joseph Fishkin, a law professor he identified with UCLA, whose work on bottlenecks he called among the most important books of the past fifty years. Imposing a single fair test on everyone reorders who succeeds without increasing the number of routes to success; Fishkin's preferred objective is plurality of opportunity rather than equality of it. Applied to commerce, Sutherland's version is that companies which behave interestingly are almost always family-owned or founder-led.

On artificial intelligence specifically, he predicted a split. Asked what is easiest to sell to a large public company, he answered cost savings, and said quarterly-reporting cultures treat cost reduction as an addiction. Founder-led and family-owned businesses, working to different time frames, are more likely to ask what new thing the technology makes possible. Both framings are legitimate, he said. The problem is that most firms will pick the same one.

When an option becomes an obligation

Sutherland described himself as having become more cautious with age about a particular pattern: "a lot of technology arrives as an option and ends up as an obligation."

The examples were retail and transport infrastructure rather than advertising. Self-service checkouts, ticket machines at railway stations, parking applications and screen-only ordering at McDonald's all arrived as alternatives and were welcomed as such. Then, in his account, the accountants arrived and removed the alternative, because staffing the till, emptying coin machines and manning the ticket office all cost money. He noted that his father, in his late eighties, was effectively unable to park a car in the final years of his life for want of a smartphone.

The retail consequences he described were not the ones the business case anticipated. Shoplifting rose sharply once shoppers scanned their own goods, with produce being weighed as cheaper items; Sutherland said some supermarkets found themselves selling more carrots than they had bought. Separately, self-scanning from a full trolley proved impractical, which undermined the weekly shop that large-format supermarkets exist to serve.

His structural point concerned accountability. Procurement, he argued, can claim credit for a cost reduction without being held responsible for the value that reduction destroys. Marketing sits under the opposite arrangement.

The asymmetry in how marketing is judged

Marketing is charged for every penny of cost, Sutherland said, while being permitted to claim only a portion of the upside, and starting again from zero each year regardless of how long an idea keeps working.

He illustrated this with the naming campaign devised by Ogilvy in Australia, which put individual first names on Coca-Cola cans. According to Sutherland, the idea was still running a decade later, recurring roughly every year or two across about 180 countries, and had generated a billion dollars for the client. The agency was paid A$350,000 for it. The Australian marketers who conceived it received no share of the upside either, and the incremental sales it continues to produce, which he put at around $100 million a year, are now credited elsewhere in the organisation.

The underlying distribution is what he called fat-tailed, closer to venture capital or film than to a process industry. Half the value of marketing is incremental and reliable. The rest arrives perhaps once every two years and cannot be identified in advance. Across fifteen years working on the American Express account, he said, three things mattered more than everything else combined, and none of them was knowable beforehand.

Two consequences follow in his account. Businesses systematically overspend on acquisition and underspend on retention and customer experience, because one is fast and easy to measure and the other is slow and hard. And finance functions tend to eliminate exactly the details that make brands memorable, because those details look unnecessary: the meerkat's accent and smoking jacket, the hotel cookie, staff and client entertainment cut after a bad month for what he characterised as symbolic reasons within a rounding error.

Fame, he argued, is the compounding asset that this accounting cannot see. "If you're not famous, you have to find all your customers," he said. A firm that is known receives approaches it could not have forecast. He described a coffee format built for airports and railway stations, on the premise that some customers want a standard drink without queueing behind a complicated order. Within two days of the launch publicity, conference organisers approached with a use case nobody involved had considered, because plenary sessions empty all at once. Sutherland tied this to Adam Smith's division of labour, extended into what he called a division of attention.

Process, randomness and what the industry pretends

Sutherland was blunt about how agencies present their work. Advertising has to pretend there is a process, he said, because agencies are paid by the hour and procurement wants an account of every minute. Checklists and groundwork exist and matter. A linear method that reliably converts inputs into ideas does not, and the pretence has made an iterative activity worse by forcing it into a straight line.

He offered the meerkat campaign as evidence, suggesting its origin lay in a creative team member from Barcelona and the Catalan word for market. No process reaches that. What can be managed, in his framing, is the amount of randomness in the system and the ability to recognise significance when it appears. Penicillin and Viagra both arrived that way, the latter having been developed as a treatment for angina.

His closing position kept both halves. A reasonable amount of order is necessary, and a framework of order is necessary. Eliminating randomness entirely, he said, is "really really dangerous in the medium to long term."

He raised one further concern about labour. For roughly 150 years, businesses operated under an unwritten assumption that a company doing well meant employees doing at least adequately. Sutherland argued that automation has begun to break that assumption, and that the emerging narrative treats staff reduction as the route to corporate performance. He cited a figure of 20% of young people sabotaging their employers' artificial intelligence efforts, and noted that the behaviour is not necessarily irrational from their position.

Why the argument lands now

The measurement critique arrives into an industry that has been documenting the same shortfall from the inside. A March 2026 Brandwatch survey of 1,028 marketers found that only one in four say they understand their audiences very well, with predicting future needs or behaviours cited as the leading challenge at 60% and understanding the reason behind audience decisions at 40%. That last figure maps directly onto Sutherland's stated-versus-revealed distinction: the reason is precisely what a click-through rate cannot supply.

Confidence in measurement has not moved with data volume. TransUnion research reported in October 2025 found platform-provided attribution still the most common methodology at 65.8%, with the lowest confidence scores attaching to influencer marketing at 44.4% and in-store activity at 38.3%, and nearly 30% of respondents facing cuts to measurement budgets. A MiQ study of 53 million households published in April 2026 found only 43% of marketing professionals confident in their measurement capabilities, with fragmented screens and supply paths identified as the primary blockers.

The specific bias Sutherland describes, in which slow effects are treated as absent, has been quantified more than once. Research from Google and partners including Nielsen, Kantar and WARC, covered in March 2025, found that returns in months five through twenty-four match those of the first four months, and that a 1% increase in brand awareness typically drives a 0.6% lift in long-term sales alongside a 0.4% short-term lift. TransUnion and MMA Global research from October 2025 put the long-term effect at up to six times the short-term conversion lift at Kroger, with 1.8 times at Ally Bank and 2.5 times at Campbell's, and suggested conventional tools may undervalue brand contributions by as much as 83%. The measurement community itself remains divided over whether declining effectiveness is real or an artefact of the instruments.

The tooling question is separate from the framing question. An Adobe study reported in February 2026 found marketers automating only 18% of their daily tasks, while capital continues to flow toward decision systems: $3.7 billion into artificial intelligence sales and marketing startups during 2026 by early May, against a Gartner forecast that more than 40% of agentic projects will be cancelled by the end of 2027. Vendors have begun pitching directly at the demographic-proxy problem, with Youtility closing a $4.2 million seed round in April 2026 on a behavioural decisioning thesis. On the comparability point Sutherland raises indirectly, AudienceProject argued in April 2026 that the industry's difficulty is a lack of comparable methodology rather than opposed interests.

Fractl research covered in June 2026 found YouTube dominating affinity across most business-to-business categorieswhile open web traffic fell 46% over three years, another sign that teams misread where buyers spend attention.

Sutherland's own recent record on this subject is consistent. PPC Land reported on June 13, 2026 that he had warned advertising-funded artificial intelligence would repeat the incentive failures of search, with financial pressure gradually displacing relevance. The May conversation extends the same logic inward, to the metrics marketing departments apply to themselves.

Timeline

Summary

Who: Rory Sutherland, vice chairman of Ogilvy, in conversation with Aaron Gibson, founder and chief executive of the analytics firm Hurree.

What: A recorded discussion arguing that marketing and corporate measurement systems privilege easily counted inputs and treat unmeasured forces as zero, illustrated with a stated-preference gap of 75% against roughly 8% on environmental premiums, a claim that 80% of managers in very large organisations cancel projects they expect to be profitable in order to meet quarterly targets, and the A$350,000 fee Ogilvy Australia received for a naming idea that generated a billion dollars for its client.

When: Recorded in May 2026 and published on July 3, 2026, reaching 5,350 views and 15 comments by the time the page was captured.

Where: Filmed before a live audience and posted to Aaron Gibson's YouTube channel, with both participants stating an intention to attend MAD//Fest London in July 2026.

Why: The argument bears directly on how advertisers, media buyers and marketing technologists allocate budget, because survey-derived intent, short attribution windows and quarterly reporting cycles all encode the same assumption Sutherland disputes, that what cannot be counted quickly does not count at all.