Rory Sutherland’s “Doorman Fallacy” is the best argument against AI marketing automation — and the best argument for doing it right. The fallacy: a consultant sees a hotel doorman, defines his job as “opening doors,” replaces him with automatic doors, and destroys the security, the taxi-hailing, the guest recognition, and the status signaling that made the hotel worth staying at. Most AI marketing tools are making the same mistake. They automate the visible task and destroy the invisible value. But the solution isn’t less automation — it’s knowing which door to automate and which doorman to keep.

Key Takeaway: Sutherland’s Doorman Fallacy, his warning about bad marketing, and his rules for good copy all point to the same truth: marketing’s value is psychological, not mechanical. AI that automates the mechanics (scheduling, distributing, A/B testing) creates value. AI that automates the psychology (brand voice, audience intuition, creative judgment) destroys it. The line between the two is where most AI marketing tools get it wrong.

In a two-hour conversation with Shane Parrish on The Knowledge Project, Ogilvy Vice Chairman Rory Sutherland delivered what amounts to a masterclass in why efficiency-obsessed automation fails. We’ve written about Sutherland before — on why cost reduction isn’t a strategy and why irresponsible experimentation drives breakthroughs. This time, he gives us the framework for understanding exactly where AI marketing should stop.

And where it should go further than anyone expects.

The Doorman Fallacy: Any Idiot Can Cut Costs

Sutherland’s setup is deceptively simple. A management consultant walks into a five-star hotel. They see a doorman earning $30,000–$40,000 a year. They define his job as “opening doors.” They recommend automatic doors and infrared sensors. They pocket the savings.

What they destroyed: security (the doorman kept vagrants out), taxi-hailing (guests didn’t wait in the rain), luggage handling (first impression of service), guest recognition (regulars felt known), and status signaling (the uniformed presence said “this hotel takes you seriously”).

Sutherland cites Roger Martin’s observation: “Any idiot can cut costs. What takes real skill is cutting costs in a way that doesn’t destroy value.”

This is the most important sentence in AI marketing right now. Because nearly every AI marketing tool is doing exactly what the consultant did — defining marketing as “producing content,” automating the production, and destroying everything else.

Where Most AI Marketing Tools Are the Automatic Door

Consider what happens when you replace a marketing team with a content generator:

What the tool automates	What it destroys
Writing blog posts	The editorial judgment about what to write about
Generating social copy	The intuition about when to stay quiet
Creating ad variations	The brand taste that knows which variation feels right
Scheduling posts	The awareness that your audience just experienced a crisis and today isn’t the day

The content generator looks at the marketing team and says: “Your job is producing content. I can produce content faster and cheaper.” That’s the automatic door. It solves the visible task and misses the invisible value.

The invisible value in marketing is judgment. When to post. When not to post. Which competitor move deserves a response and which to ignore. Whether this week’s content should be ambitious or empathetic. Those decisions require understanding the brand’s audience at a level that most AI tools don’t even attempt.

The Danger of Bad Marketing: When Automation Destroys Brands

Sutherland’s analysis of marketing failures is equally instructive. He distinguishes between genuine missteps and manufactured outrage. The Bud Light controversy, he argues, was “confected outrage” — an influencer partnership amplified by groups on both sides for signaling purposes. A rational analysis of the actual campaign wouldn’t predict a brand crisis.

Gillette’s campaign was different. It deliberately challenged its core audience’s identity. The ad wasn’t misunderstood — it was understood perfectly, and the audience rejected it.

The lesson for AI marketing: bad marketing isn’t about being offensive. It’s about misunderstanding who you’re talking to.

This is where autonomous AI marketing gets genuinely dangerous. An AI optimizing for engagement metrics might produce content that generates clicks but alienates the core audience. An AI trained on generic best practices might push a brand into messaging that sounds like everyone else — what Sutherland calls the trap of benchmarking against competitors: “Don’t benchmark against your most obvious competitor. All you’ll do is make yourself a copy.”

The Royal Mail example drives this home. When researchers measured postal service quality, they found that brand perception had almost nothing to do with delivery reliability. The major determinant of whether customers liked Royal Mail was whether they liked their postman. High-reliability districts with unfriendly carriers scored lower than unreliable districts with friendly ones.

Sutherland calls these moments “brand quakes” — a single remarkable interaction that fundamentally reshapes someone’s perception of a company. A postman who holds a parcel in the rain, remembers your name, or does a small unprompted favor creates more loyalty than a decade of 99.9% delivery reliability. The brand quake isn’t efficient. It doesn’t scale in a spreadsheet. But it causes a complete reevaluation of the customer’s relationship with the brand — and that’s worth more than any metric.

Sutherland points to Dyson as a company that gets this right. James Dyson reframed customer contact as an honor, not an interruption — and argued the customer service team should sit next to the boardroom, not in a basement call center. When every customer interaction is treated as intelligence rather than cost, the entire organization learns faster. Most companies do the opposite: they optimize customer service for speed and deflection, turning every touchpoint into a brand quake in the wrong direction.

As Sutherland puts it: we have very little evolved experience evaluating postal efficiency, but we have half a million years of evolved experience deciding who to like and trust.

This is the deepest challenge for AI marketing. An AI can optimize send times, subject lines, and channel mix with inhuman precision. But can it create a brand quake? Can it recognize the moment when a customer needs something unexpected — a personal touch, a surprising response, a gesture that says “we actually see you”? The brands that win long-term are built on brand quakes, not on optimized funnels. An AI that only optimizes for delivery metrics is running the map. An AI that understands how people actually form brand relationships is reading the territory.

How to Write Good Copy — And Why AI Gets It Backwards

Sutherland’s copywriting rules are deceptively simple:

Use verbs over adjectives, adjectives over adverbs. Verbs create movement and action. Adjectives describe. Adverbs qualify. The hierarchy matters because persuasion is about getting someone to do something, not admire your description of it.
Prefer Anglo-Saxon words over Romance-language equivalents. “Buy” over “purchase.” “Help” over “facilitate.” “Start” over “commence.” Shorter, punchier, more direct.
Convert features into benefits. The classic copywriting discipline: nobody cares that your product has a 50,000-word context window. They care that it remembers their brand voice across every conversation.
Write conversationally. As David Ogilvy put it: use the language your audience uses every day, the language in which they think. Arthur Conan Doyle’s Sherlock Holmes stories never make you backtrack to figure out who’s speaking. Clarity is respect.
Sometimes, just tell people a fact. Sutherland’s most counterintuitive advice: persuasion doesn’t always require rhetoric. Sometimes stating a fact plainly is enough to shift belief.

Here’s why this matters for AI marketing: most AI-generated content does the exact opposite of every rule on this list.

AI defaults to adjectives. “Our innovative, cutting-edge, AI-powered platform delivers seamless, end-to-end marketing automation.” Count the adjectives. Now count the verbs. The ratio is backwards.

AI defaults to Romance-language words. “Leverage,” “utilize,” “facilitate,” “optimize.” These are the hallmarks of AI-generated corporate prose — and they’re exactly what Sutherland says kills copy.

AI struggles with features-to-benefits conversion. It can list features endlessly. Converting them into the specific emotional benefit for a specific audience requires understanding who you’re talking to and what they care about — the territory, not the map.

This is the copilot ceiling in action. AI copilots produce drafts fast. But fast drafts that violate every rule of good copy create more work, not less. The human still has to rewrite the adjectives into verbs, replace “utilize” with “use,” and figure out whether this piece should be empathetic or ambitious.

The Real Doorman in Marketing

So where’s the line? What should AI automate in marketing, and what should it leave to humans?

Sutherland gives us the framework without knowing it. The doorman’s visible job — opening the physical door — is the commodity task. Automate it freely. The doorman’s invisible job — security, recognition, status, judgment — is the value. Preserve it at all costs.

In marketing:

Automate freely (the automatic door):

Scheduling and publishing across channels
A/B testing headlines and subject lines
Distributing content to the right platforms at the right times
Monitoring performance metrics and flagging anomalies
Resizing and reformatting creative assets
Managing email sequences and drip campaigns

Preserve at all costs (the doorman):

Brand voice and tone decisions
The judgment to not post when timing is wrong
Creative direction and editorial vision
Crisis response and sensitivity
Audience intuition built from years of relationship
The “taste” that knows which version feels right even when the data says otherwise

The problem with most AI marketing tools is that they automate the doorman’s job (writing, creating, deciding) while leaving the automatic-door work (scheduling, distributing, monitoring) to the human. It’s exactly backwards.

A Brand Parent inverts this. It handles the mechanics — the scheduling, distributing, testing, monitoring, and adjusting — while the human retains the judgment that Sutherland correctly identifies as irreplaceable. Not because AI can’t generate words, but because the choice of words, the timing of words, and the silence between words is where marketing value actually lives.

What Sutherland Would Say About the Trust Maturity Model

Sutherland’s entire body of work circles one idea: humans are not rational optimizers. We’re social animals who make decisions based on trust, status, habit, and herd behavior. Any system that ignores this will fail — no matter how efficient it is.

This is why the trust model matters. You wouldn’t hand a new CMO the keys on day one. They observe first. Then advise. Then co-pilot. Then earn autonomy within boundaries. The progression isn’t a feature — it’s how trust actually works in human relationships.

“Character is a proxy for quality,” Sutherland says. “When people can’t judge a product’s quality, they judge the seller’s character.”

An AI that earns trust through demonstrated judgment — that shows it understands when to speak and when to stay quiet, that learns a brand’s specific audience over time, that flags risks before they become crises — is building the kind of character-based trust that Sutherland describes.

An AI that generates 50 blog posts and dumps them into a scheduling tool is the automatic door.

The Doorman Stays. The Paperwork Goes.

Sutherland’s Doorman Fallacy isn’t an argument against automation. It’s an argument against stupid automation — the kind that optimizes for visible costs while destroying invisible value.

The marketing doorman — creative judgment, brand intuition, audience empathy, the human taste that no algorithm replicates — stays. The marketing paperwork — the scheduling, distributing, monitoring, testing, reformatting, and reporting that consumes 80% of a marketing team’s time — goes.

That’s not a contradiction. That’s the whole point.

The consultants who replaced the doorman with automatic doors weren’t wrong about doors being openable by machines. They were wrong about what the doorman was actually doing. The same mistake is happening in AI marketing right now — and Sutherland just gave us the framework to avoid it.

Want AI that handles the paperwork, not the doorman? Get early access to Lane — the AI CMO that automates execution while you keep creative control. Or see how it works.

Got questions? Reply to the email that brought you here, or reach out at hello@luminarylane.net.

References

Rory Sutherland — How to Think Like a World-Class Marketer (The Knowledge Project with Shane Parrish, Dec 2025)
Related: Cost Reduction Isn’t a Strategy — Our first analysis of Sutherland’s behavioral science framework
Related: Spend 5% Irresponsibly — Why an AI agent makes experimentation free
Related: Anthropic’s Team Hits the Copilot Ceiling — Why copilots produce drafts but don’t do the job
Related: The AI CMO: From Marketing Tools to Brand Parent — Our pillar piece on the Brand Parent category