The Anti-Data Playbook: What Growth Leaders at Meta, Twitter, and Instacart Know About When to Ignore the Numbers

Every Growth Experiment From Gut Instinct Has Failed

Casey Winters has a confession that should make every data-skeptic uncomfortable. As CPO at Eventbrite and former growth lead at Pinterest, he's recommended countless experiments based on intuition honed across multiple hypergrowth companies. His track record when ignoring the data? Zero for everything.

The first thing I tried at Grubhub, failure. It was trying to set up some affiliate marketing relationships that I built out in the past. Didn't work. When I started at Pinterest, I tried to do some internal linking work for SEO. Didn't move the needle at all.

— Casey Winters

This isn't false humility. Winters articulates a counterintuitive principle that separates true growth operators from optimization theater: data isn't just useful for finding ideas—it's the core of finding good ideas. When Pinterest shifted its acquisition model to accommodate late-majority users by opening up content previews before signup, the team made a strategic bet without experimental validation. It failed. The pattern held across companies, across verticals, across his entire career.

The lesson cuts against the mythology of the visionary growth leader who "just knows." Ed Baker, former head of growth at Uber and part of the early Facebook growth team, learned the inverse lesson: the experiments that seem too small to matter often change everything. At Uber, his team secretly ran a test in Hyderabad allowing riders to book without entering a credit card—directly defying CEO Travis Kalanick's explicit instructions. The result? Conversion to first paid trip doubled. When Baker shared the data, Kalanick's response captured the tension perfectly: "I hate you guys, but I love you guys."

Cultural Translation Beats Feature Iteration

The Japan problem at Facebook looked insurmountable. The local competitor Mixi dominated. Conversion data showed a massive drop-off at the invite step. The country manager explained it simply: inviting friends to a new platform is considered rude in Japanese culture. Most growth teams would have moved on, accepted the cultural barrier, focused elsewhere.

Baker's team made one change. They stopped calling them "invites" and started calling them "announcements." The new framing: let Facebook announce to your friends that you're now on the platform and this is where they can find you. Japan went from one of Facebook's slowest-growing countries to one of its fastest. They surpassed Mixi shortly after.

Just by changing the wording, it was more culturally accepted. And Japan went from one of our slowest growing countries to one of our fastest growing countries.

— Ed Baker

This wasn't A/B testing button colors. It was understanding that growth constraints live in the space between product intent and cultural reception. Andy Johns encountered the same dynamic at Twitter, where username selection created expanding friction as the namespace shrank. Every new user faced an increasing probability of error messages: name taken, try again, name taken, try again, exit.

Johns assembled a team of four—one designer, one 19-year-old engineering intern, one frontend engineer, and himself. They implemented a hacky auto-generation solution without asking permission, knowing the ideological battles inside Twitter would stall any formal process. The result: 70,000 additional signups per week, representing a 20-30% lift. The lesson wasn't about moving fast and breaking things. It was about recognizing when internal ideology blocks obvious user value.

Retention Engineering Is the Actual Growth Lever

The Facebook growth team's obsession surprised Baker when he joined. He expected a relentless focus on new user acquisition. Instead, the team spent vastly more time on retention and engagement, particularly as the platform scaled. The math becomes obvious in retrospect: at massive scale, retention rate determines growth trajectory more than acquisition volume.

Bangaly Kaba saw this play out at Instagram, where he watched retention improve from roughly 20% to over 50%. The driver wasn't acquisition channel optimization or onboarding tricks. It was understanding what type of connections actually provided value—what made people find magic in the product and keep coming back.

World-class companies, especially those that have tremendous growth, have incredible focus. They focus on one thing at a time that is working really, really well. They make sure they understand it, they know why it's working, how to increase it, and then they build an unfair advantage around it.

— Bangaly Kaba

Instagram's unfair advantage became its account graph. The team identified the specific threshold of high-quality connections posting regularly that triggered the magic moment. Then they engineered the entire experience around hitting that threshold. Kaba's framework inverts the typical startup approach: don't choose a metric and optimize it. Start with the qualitative magic—when does a user experience something they can't get elsewhere?—then translate that into quantitative measurement.

At Instacart, the same principle manifested differently. The company built an incredible muscle around partnerships and operations, creating value through efficient service and expanding product availability. The North Star wasn't growth-team innovation. It was executing one core loop better than anyone else possibly could.

The Small Team That Shipped 70,000 Weekly Signups

Twitter's username problem illustrates why growth teams fail at scale. The friction was obvious. The solution was technically straightforward. But implementation stalled because of ideological conflict about what Twitter "is known for." Johns made a tactical decision: build the hacky version with a skeleton crew, launch it as an A/B test, and take personal responsibility if leadership objected.

The four-person team shipped in days. The data arrived shortly after. The solution worked. This wasn't cowboy engineering—it was understanding that conviction requires speed, and speed requires small teams with clear authorization to take risks.

I said, team, let's just throw in some hacky solution where we just automatically generate a username for that person. Let's just run it as an A/B test. Let's not ask for permission. If the axe falls because people are unhappy that we ran this A/B test, it was my choice.

— Andy Johns

Baker learned the same lesson at Uber. When India operations reported that credit card requirements blocked massive user populations, the team ran a test despite CEO opposition. The 2x conversion improvement made the case impossible to ignore. The pattern across these companies isn't about rogue operators—it's about leaders who understand the difference between experiments that need consensus and experiments that need data.

Instagram's multiple accounts discovery emerged from a similar investigative process. The team focused on "access churn"—users who signed out and couldn't get back in because they forgot passwords. Fixing that problem created unexpected downstream effects: content production increased, engagement rose, and the team had no idea why. They devoted an entire roadmap to understanding the phenomenon. The answer: people with second or third accounts were getting locked out of those additional profiles. That insight led Instagram to lean into multiple account support years before it became table stakes.

Growth Model Constraints Trump Growth Team Tactics

Winters articulates the most important responsibility of a growth leader: understanding the growth model of the business and identifying current constraints. Most of those constraints, particularly in network effects businesses, aren't solvable by growth teams. They require new product functionality or entirely new value propositions.

As CPO at Eventbrite, Winters manages product management, product design, growth marketing, and research. The growth model analysis consistently reveals that the binding constraints aren't growth-team problems. They're product problems. This creates a fundamental tension in how companies structure growth organizations. The one-size-fits-all approach to building growth teams fails because different businesses face different constraint types.

Growth is about connecting people to the value that's already been created by the business. So growth is not about building totally new features, totally new products. It's about once we have those and we've confirmed that they have product market fit, that we're doing the best job possible at connecting as many people as possible to those products and features.

— Casey Winters

Johns observed the same pattern across companies. Startups too quickly assume that optimization-based approaches—running experiments, analyzing data, A/B testing toward the future—will unlock the next growth phase. This assumption overwhelms their ability to take big risks and think from an innovation standpoint. Growth methodology becomes a crutch that prevents the product innovation actually required to break through plateaus.

The best growth operators know when their tools don't apply. They recognize constraint types. They understand that making it easier to find existing value or acquiring more people who already want the product works only when the product delivers differentiated value to a clear audience. When those conditions don't hold, no amount of conversion rate optimization will matter. Baker's stance is unequivocal: needle-moving experiments reveal themselves immediately with minimal data. Tests that require weeks to detect fractional improvements in button color probably aren't worth running when bigger opportunities exist.

The Only Unfair Advantage Is Knowing What You're Actually Good At

Kaba's observation about hypergrowth companies cuts through the noise of growth tactics and marketing playbooks. Instagram didn't try to be good at everything. They became world-class at understanding their account graph and the specific connection types that created value. Instacart didn't disperse effort across every possible growth channel. They built an unmatched capability in partnerships and operations, then ran that play relentlessly.

This level of focus requires saying no constantly. It means understanding your growth model well enough to know which constraints actually bind. It means recognizing when cultural friction blocks product adoption and finding the minimal translation layer. It means building small teams with clear ownership who can ship experiments before internal politics calcify.

The meta-lesson across these operators: growth isn't a bag of tricks you apply generically. It's a diagnostic discipline. You identify where value exists, where friction blocks access to that value, and what structural changes eliminate the friction. Sometimes that's a wording change in Japan. Sometimes it's allowing cash payments in India. Sometimes it's auto-generating usernames. Sometimes it's fixing password recovery to discover multiple account usage patterns.

And sometimes—maybe most times—the answer is that your growth team can't solve the actual constraint. The product needs to get better. The value proposition needs to sharpen. The market fit needs to deepen. The operators who built the growth engines at Facebook, Twitter, Instagram, Instacart, and Uber share an unexpected humility: they know exactly when their expertise doesn't apply. That knowledge might be the most valuable growth skill of all.