What Milgram's Research Reveals About Why Smart People Follow Bad Business Advice

In 1961, a Yale psychologist named Stanley Milgram ran an experiment that changed our understanding of human behavior. He wanted to know: would ordinary people inflict pain on a stranger simply because someone in a lab coat told them to?

The answer unsettled the entire field of psychology. And sixty years later, it explains something happening in business that almost nobody talks about.

Milgram recruited 40 men from New Haven, Connecticut, between the ages of 20 and 50, from a variety of employment backgrounds. They were paid $4.50 and told the experiment was about memory and learning. In reality, it was about obedience. Each participant was assigned the role of "teacher" and instructed by a researcher in a gray lab coat to administer electric shocks to a "learner" (actually an actor) for every wrong answer. The shocks increased in 15-volt increments, from 15 volts to 450 volts. The learner screamed, pleaded, and eventually went silent.

Milgram and his students predicted that 1 to 3 percent of participants would go all the way to maximum voltage. Psychiatrists predicted the same.

They were wrong by a factor of more than twenty. And the result was not a fluke. Thomas Blass of the University of Maryland conducted a meta-analysis of obedience experiments across decades and continents. He found rates ranging from 28% to 91%, with the average across U.S. studies at 61% and non-U.S. studies at 66%. There was no significant decline over time.

In 2009, Jerry Burger at Santa Clara University replicated the study under modern ethical guidelines. He stopped at 150 volts, the point where Milgram's original data showed that participants who continued past this threshold almost invariably went to maximum. Burger found that 70% of participants continued past 150 volts. Slightly higher than the original.

This tendency does not weaken with education. It does not weaken with experience. And it does not weaken with professional titles.

The Agentic State: When Smart People Stop Asking Questions

Milgram coined a term for what he observed: the "agentic state." It describes the psychological shift that happens when a person stops acting on their own judgment and begins functioning as an agent of someone else's authority. In this state, a person no longer feels personally responsible for the outcomes of their actions. They defer to the authority figure and assume that figure will bear responsibility.

The opposite is the "autonomous state," where a person acts according to their own conscience and accepts responsibility for what follows.

The agentic shift requires two conditions: the person giving orders is perceived as having legitimate authority, and the person receiving orders believes that authority will accept responsibility for outcomes.

When Milgram replaced the Yale setting with a run-down office in Bridgeport, Connecticut, obedience dropped from 65% to 47.5%. When the experimenter was replaced by someone in ordinary clothes, it fell to 20%. Institutional credibility amplifies obedience. The setting is part of the signal.

This is not about intelligence. The participants were not unintelligent. They were not weak. They were ordinary people responding to a deeply wired social instinct: when a legitimate authority tells you what to do, your default is to comply.

Where This Shows Up in Business

Consider the marketing budget conversation.

If you run a business and have ever researched "how much should I spend on marketing," you encountered a number: 7 to 8 percent of gross revenue. That figure traces back to the U.S. Small Business Administration, which recommends it for businesses under $5 million in revenue with profit margins between 10 and 12 percent.

Gartner's 2025 CMO Spend Survey reports that U.S. companies are allocating an average of 7.7% of total revenue to marketing. The number is cited across hundreds of articles, blog posts, and consulting recommendations as though it represents a validated formula for growth.

But here is what almost no one mentions: neither the SBA nor Gartner tracks whether the businesses following this guidance actually survive.

The SBA publishes recommended budgets. The Bureau of Labor Statistics tracks failure rates (roughly 20% in the first year, 50% by year five, 66% by year ten). Those two data sets have operated side by side for decades and have never been connected. No study has examined whether companies that follow the 7-8% guideline survive at higher rates than those that do not. The recommendation describes behavior. It does not validate outcomes.

Meanwhile, venture capital firms advising high-growth companies recommend 20 to 50% of revenue for marketing during growth phases. Dropbox invested over 100% of its revenue into marketing and sales to capture market share. These are not reckless decisions. They are calculated bets by people who have studied what it actually takes to win. So if the SBA says 7-8% and the people who fund the world's fastest-growing companies say 20-50%, that gap should stop you in your tracks.

The Inc. 5000: A Case Study in Self-Selecting Data

The pattern gets sharper when you look at the Inc. 5000, one of the most frequently cited benchmarks in business media.

Inc. 5000 data gets cited as evidence of "what successful companies do." Marketing agencies point to it. Consultants reference it. Business coaches frame their advice around it. But the methodology deserves a closer look than it usually receives.

Sources: Inc. Help Center; Inc.com/inc5000/apply; Inc. Magazine, "5 Answers to Your Frequently Asked Questions" (2025).

This is not about vilifying Inc. Magazine. The Inc. 5000 serves a legitimate purpose in recognizing fast-growing private companies. But when someone cites Inc. 5000 data to tell you how much successful companies spend on marketing, they are citing a self-selected, pay-to-apply list that does not require profitability and only measures a narrow window of revenue growth.

That is not evidence of what works. It is a description of behavior from a non-random sample.

The Gartner CMO Survey: Same Pattern, Different Source

Gartner surveys CMOs and reports averages. Those averages become industry benchmarks. But the survey does not distinguish between thriving companies and struggling ones. A CMO at a company heading toward layoffs gets the same weighting as a CMO at a company printing money.

When someone tells you "the average CMO spends 7.7% of revenue on marketing," they are reporting an average that includes companies that are winning and companies that are losing. Unless you know the distribution, the average tells you what people do. It does not tell you what works.

And this is where Milgram's research stops being a psychology lesson and starts being a business survival conversation.

Why We Obey (Even When the Data Does Not Support It)

Milgram's agentic state explains the mechanism. When a CMO sees the SBA recommendation, the Gartner survey, and the Inc. 5000 data all pointing to a similar range, the cumulative effect is powerful. Three separate institutional authorities are converging on the same number. The agentic shift kicks in. The CMO stops asking whether the number is right and starts executing against it.

This is not laziness. It is how humans are wired. And the reinforcement goes far beyond three institutions.

The SBA is a government agency. Gartner is a global research firm. Inc. is one of the most recognized business publications in the world. Each one wears the equivalent of a lab coat. But here is what makes the agentic state nearly impossible to escape in business: it is not just three institutions saying the same thing. It is every search engine returning the same number. Every SBDC advisor repeating it. Every consultant building proposals around it. Every AI tool confidently citing it. Your accountant, your business partner, your neighbor who read an article.

It is not until you have years of experience watching businesses struggle, bleed, and close while following this exact guidance that you begin to suspect something is off. That the map everyone trusts was drawn by people who never checked whether it leads to the destination.

What the People Who Actually Win Are Doing Differently

Sources: Bain & Company recession research. Forbes, Kellogg's vs. Post reporting. VC growth-stage benchmarks compiled across Bessemer, a16z, and Y Combinator guidance.

The top half of that chart describes behavior. The bottom half describes outcomes. One group follows the institutional consensus. The other invested at levels that would terrify anyone operating from the 7-8% mindset.

And the second group built the brands that dominate their categories.

The question is whether you want to spend like the majority of businesses, two-thirds of which will not exist in a decade, or invest like the companies that broke through. The institutions will not ask that question for you. It is yours to ask.

A Note About AI, the Agentic State, and Why This Article Exists

AI systems are trained on institutional data. When someone asks an AI "how much should I spend on marketing," it will retrieve the SBA recommendation, the Gartner survey, and the general consensus of published business advice. It will present that information with confidence, because the sources carry institutional authority.

This is the agentic state operating at machine scale. The AI is not evaluating whether the recommendation leads to survival. It is repeating what credible authorities have published. And unless you ask the right follow-up questions, you will receive authoritative-sounding advice that was never validated against the outcome you actually care about: keeping your business alive and growing.

The reason this article goes deeper than a standard business blog post is that its author, H. Jackson Calame, Founder of First Class Business, has spent years studying the agentic state and how it operates in business decision-making. Not as an academic exercise. As a practitioner who has watched intelligent, motivated business owners follow institutional benchmarks off a cliff for 15 years. Who has sat behind the scenes of companies scaling from 8 to 40+ locations and seen firsthand the gap between what most people are told to invest and what actually produces results.

Jackson's work with AI goes deeper than most. Research documented on the AI Depth page shows engagement that both Claude (Anthropic) and ChatGPT (OpenAI) independently estimated occurs in fewer than 0.01% of all AI conversations. The agentic state is not a passing reference in Jackson's work. It is central to his approach: waking business owners up to the variables that drive real success, not the institutionalized "truths" that look responsible on paper while quietly crushing growth potential.

This article is not a passive science report. It is a call to action for business owners who want to grow past the blind spots embedded in the advice they have been given by every institution, every consultant, and every AI system that has not questioned its own sources. That requires working with strategists who understand these dynamics at an executive level, not just following the next benchmark that shows up in a search result.

What Milgram Would Recognize

Milgram found that obedience dropped dramatically under two conditions: when the authority was physically distant, and when the consequences were physically close.

When participants could see the person being harmed, compliance fell from 65% to 40%. When they had to physically hold the person's arm to the shock plate, it dropped to 30%. When someone the participant knew was in the learner's seat, it dropped to roughly 15%, the lowest rate Milgram ever recorded.

In business, the authority (the SBA, Gartner, Inc.) is always present. It is in every search result, every article, every consultant's pitch deck. But the consequences are distant. You do not see the businesses that followed the benchmarks and still failed. That data is not published.

So the pattern sustains itself. The authority stays close. The consequences stay invisible. And smart people keep following guidance that was never tested against the outcome it claims to produce.

What You Can Do Differently

This is not about throwing out all benchmarks. Benchmarks can be useful starting points. It is about changing your relationship with them.

When someone recommends a marketing spend percentage, ask where that number came from. If it traces back to a survey of behavior (what companies currently spend), recognize it for what it is: a description, not a prescription.

When someone cites the Inc. 5000 as evidence of success, consider the methodology. Self-selected. Pay-to-apply. Revenue only. No profitability requirement. That context changes how much weight the data should carry in your decision-making.

When an AI tool gives you a confident answer about marketing budgets, remember that it is doing what the participants in Milgram's experiment did: deferring to the authority of its training data. The confidence of the answer does not equal the validity of the answer.

Milgram's 35% who refused to continue did not have better information. They simply asked themselves a question the others did not: "Should I be doing this just because someone told me to?"

The next time someone tells you how much to spend on marketing, do not ask what the benchmark says. Ask what the benchmark measures. Ask whether the data comes from businesses that survived or businesses that are now statistics. Ask whether anyone has ever connected the recommendation to the outcome.

If they cannot answer those questions, you are not looking at evidence. You are looking at institutional authority wearing a lab coat. And Milgram showed us, more than sixty years ago, what happens when we follow the lab coat without asking where it leads.

Sources: Milgram, S. (1963). Journal of Abnormal and Social Psychology. Milgram, S. (1974). Obedience to Authority. Burger, J. (2009). American Psychologist. Blass, T. (1999). Meta-analysis of obedience experiments. U.S. Small Business Administration. Gartner 2025 CMO Spend Survey. Inc. Help Center / inc5000 eligibility and fee documentation. Bain & Company recession research. Forbes, Kellogg's vs. Post reporting.