What Happens When You Measure Journalism by Data

Opening

Dear subscriber, something interesting happened on March 13th. Jeff Bezos invited Washington Post executives to his home in Washington, D.C. Next to the coffee station sat a broken lock from the Watergate break-in, on display. Two 90-minute sessions, with lunch in between. So far, this sounds like a typical media owner’s town hall.

But the way the meeting started was unusual. Every attendee sat in silence reading a distributed memo—a detailed document about the Post’s business trajectory and its use of data. If you’ve ever worked at Amazon, this scene will feel familiar. It’s Amazon’s “six-page memo” meeting culture, transplanted whole.

Bezos isn’t simply trying to cut costs at a newspaper. He’s transplanting the operating philosophy that built Amazon—data-driven decision-making, relentless efficiency, a culture of “prove it with metrics”—into a 148-year-old news organization. And two concepts have emerged from that process: “cost per story unit” and “Audience Value Score.” Today I want to talk about why these two metrics matter, and what happens when you apply them to journalism.

Where $300 Million Disappeared in Three Years

Let’s start with the numbers. The Washington Post ran a loss every year from 2022 through 2025. In 2025 alone, losses exceeded $100 million, bringing the three-year cumulative loss to roughly $300 million (about ₩400 billion). Bezos paid $250 million to acquire the paper in 2013—meaning the Post has now lost more money than it cost to buy.

The structural problem can be compressed into one sentence: costs went up while output went down.

Jeff Diomofrio, the former CFO now serving as acting CEO, laid out these figures for staff at the town hall:

• Article output down 42% compared to 2020
• Newsroom costs up 16% over the same period
• Output per reporter down 36%
• Total pageviews for news and opinion down 48%
• In some areas, thousands of dollars spent to publish a single article

Diomofrio summed it up in a line that could have come straight out of an Amazon slide deck: “Costs went up, output went down. Cost per story unit has doubled since 2020.”

It’s worth pausing on the phrase “story unit.” The moment you call an article a “unit,” journalism enters the vocabulary of manufacturing. It means tracking the production cost of a single article the way a factory tracks the cost of producing one product.

Transplanting the Amazon Operating Philosophy

What Bezos is trying to apply to the Washington Post isn’t simple cost-cutting. He’s trying to move the entire operating system that made Amazon the world’s largest company.

The evidence shows up in several places. First, the “silent reading” session at the Washington home meeting. At Amazon, PowerPoint is banned, and meetings run on six-page narrative memos¹. Everyone reads the memo silently for 20 to 30 minutes before discussion begins. Bezos brought this practice directly into Washington Post executive meetings.

Second, the rejection of proposals not backed by data. When former CEO Will Lewis brought a plan in November 2024 to cut 200 newsroom jobs, Bezos sent it back, saying “there isn’t enough data to support this.” Lewis’s team then assembled a small task force to build a data model that would satisfy Bezos—and no newsroom staff were included in that group.

Third, what emerged from that process was the “Audience Value Score.” Scored 0 to 100, it combines metrics like time spent on an article, share counts, new sign-ups, and subscription conversions. Managing editor for content strategy Brian Flaherty explained at the town hall that “a score above 70 is considered very good.”

This is essentially identical to how Amazon evaluates product performance: customer data → performance metrics → resource allocation decisions. Apply this framework to journalism, and it’s no longer an editor’s news judgment that determines a story’s worth—it’s reader data.

What Can Be Measured vs. What Should Be Measured

In practice, Murray (the editor-in-chief) and his team decided which jobs to cut based on customer data, following Bezos’s instructions. They compared which sections drew the most readers against how much it cost to produce that coverage.

Here’s what resulted:

• Sports section: eliminated (despite the tradition of famous sports columnists)
• Books section: eliminated
• Metro (local news) section: sharply reduced
• Foreign correspondents: mostly laid off (Middle East, Ukraine, China, South Asia, and more)
• Areas preserved: investigative reporting, politics and national security coverage

On paper, this looks like a rational call. Foreign coverage is expensive, and sports is hard to compete in against specialized outlets like ESPN or The Athletic. But there’s something the data doesn’t capture here. A foreign correspondent network is costly, but its very existence constitutes part of the Washington Post’s brand value. The international team was a 2025 Pulitzer finalist for its Gaza coverage. And the news of the US-Israel strikes on Iran broke just weeks after the Middle East correspondents were laid off.

What the “Audience Value Score” measures is readers’ present behavior—clicks, dwell time, shares, subscriptions. But a substantial part of journalism’s value lies in preparing for things that haven’t happened yet. When Watergate reporting began, the “Audience Value Score” of those articles was probably low. But the value that reporting brought to American democracy could never have been converted into any score.

Oz’s Lens

Honestly, I’m looking at this situation through two lenses at once.

Bezos’s approach is logical. For a business that has lost $300 million over three years, with a structure where output shrinks while costs rise, leaving it unaddressed would be irresponsible. The concept of “cost per story unit” is basic cost management. For any product, you can’t set prices or make investment decisions without knowing the unit production cost. What’s actually surprising is that the Washington Post didn’t already know this basic fact.

But looking through the lens of data, I’m worried about how the “Audience Value Score” is designed. Time spent, shares, subscription conversions—all of these metrics measure short-term, individualized behavior. But journalism’s public-good value²—watching power, sounding social alarms, shaping public discourse—isn’t captured by these metrics. This is exactly where making a screw in a factory differs from writing an article. The value of a screw is delivered directly to its user, but the value of a good article is delivered indirectly—even to people who never read it.

There’s a more fundamental problem here. Even as Bezos calls for listening to subscriber data, more than 60,000 subscribers left after his decision not to endorse a presidential candidate and after the opinion section’s ideological shift. If data is supposedly the core of management, then the most dramatic data signal of all—a mass wave of cancellations—was the one that got ignored. This isn’t “data-driven decision-making.” It’s closer to “using data to justify a decision already made.”

At Amazon, when customers dislike something, the product changes. At the Washington Post, when readers left, the direction didn’t change. The same person, applying the same data philosophy, arrived at a different conclusion. Why? Probably because, for Bezos, the Washington Post is no longer purely a business—it’s also a tool of political positioning. Getting a seat on stage at President Trump’s inauguration doesn’t seem unrelated to the business environment Amazon and Blue Origin operate in.

Closing

First, introducing cost management and productivity metrics into journalism is, in itself, necessary. But if “cost per story unit” replaces editorial judgment entirely, only what can be measured survives.

Second, Amazon’s operating philosophy doesn’t translate to every industry. E-commerce, where customer data feeds directly into product improvement, and journalism, where public-good value is central, have fundamentally different feedback loop structures.

Third, always check the intent of the decision-maker hiding behind the phrase “data-driven.” Data is a tool. The same data can support opposite decisions. In the end, what matters isn’t the data—it’s the motives of the person interpreting it.

Next time someone tells you “the data says so,” ask them back: “Who chose that data, and for what purpose?”

References & Further Reading

• Benjamin Mullin, Katie Robertson, “How Jeff Bezos Upended The Washington Post”, The New York Times, 2026.03.14. This is today’s primary source—a reconstruction of Bezos’s direct intervention drawn from testimony by 24 insiders.
• Alexandra Bruell, “Washington Post Lost More Than $100 Million in 2025”, The Wall Street Journal, 2026.02. Source for Diomofrio’s town hall remarks and the financial data—the 42% drop in articles and 16% rise in costs both come from here.
• Sarah Kaplan (Washington-Baltimore Newspaper Guild), “Actually, the Washington Post Layoffs Were a Bigger Bloodbath Than You Thought”, Washingtonian, 2026.02.09. A union analysis arguing the actual scale of layoffs was larger than reported—source for the figure that 44–47% of the newsroom was cut.
• Colin Bryar, Bill Carr, Working Backwards: Insights, Stories, and Secrets from Inside Amazon, St. Martin’s Press, 2021. Essential reading for understanding the origins of Amazon’s six-page memo culture and its “working backwards from the customer” method. Published in Korea under the title Working Backwards (Dasan Books).
• Hanaa’ Tameez, “Washington Post layoffs disproportionately affected union members of color”, Nieman Journalism Lab, 2026.02.13. An analysis of the structural imbalance in the layoffs, covering the impact from a diversity standpoint.

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis — business data management and business analytics — while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master’s from Korea University’s Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

1. Six-Page Memo: A meeting document format Amazon uses instead of PowerPoint. At the start of a meeting, everyone reads a six-page narrative document in silence before discussion begins. Bezos has said, “It is impossible to write a six-page narrative memo without thinking clearly.” ↩

2. Public Good: In economics, a good that is non-rivalrous (one person’s consumption doesn’t reduce another’s) and non-excludable (available even to those who haven’t paid for it). Information about wrongdoing revealed by investigative reporting benefits citizens who never read the article, which is why journalism carries a public-good character. ↩