The Archie Abrams Playbook for Building Growth Teams That Think in Decades

Optimizing for Churn Sounds Insane Until You Understand Power Laws

Archie Abrams runs growth at a company that powers 10% of U.S. e-commerce outside Amazon and Walmart. Shopify merchants generated $235 billion in GMV in 2023—roughly the economy of Finland. And his team's job is not to keep merchants around. It's to make it absurdly easy for anyone to try entrepreneurship, knowing most will fail.

The logic is unorthodox but elegant. Most SaaS companies live and die by retention because their revenue model is subscription-only. Lose a customer, lose $29 a month forever. Shopify makes most of its money from payments tied to merchant success. The business model is closer to venture capital than software: a portfolio bet on a power law. Get thousands of people in the door cheaply, accept that most won't make it, and let the handful who scale into Allbirds or Figs fund the entire cohort.

The way we think about churn is really going back to Shopify as our mission and what we want to do, which is to increase the amount of entrepreneurship on the internet. Most businesses do ultimately fail. Can we lower the barriers to getting started and get as many people in the door trying their hand at entrepreneurship?

— Archie Abrams

What matters isn't retention rate. It's total GMV produced by a cohort over one, two, five years. Not per-merchant averages, but aggregate output. The winners don't just offset the losers—they make the system wildly profitable. Abrams calls it "the angel investing parallel." You don't need every bet to pay off. You need the outliers.

Why 30–40% of Your Winning Experiments Probably Did Nothing

Here's where it gets uncomfortable. Shopify runs long-term holdouts on every major experiment and automatically revisits them a year, two years, three years later. What they've learned is brutal: somewhere between 30% and 40% of experiments that show early lift produce zero incremental GMV a year out.

The mechanism is usually pull-forward. You goose a metric—say, number of users who make their first sale—and celebrate the lift. A year later, the GMV curve for that cohort looks identical to the control. You didn't create value. You accelerated a behavior that would have happened anyway, or you attracted a user segment that churns fast and contributes nothing long-term.

I would encourage everyone, if you can look at some of the experiments that you thought were your biggest winners, look at the downstream metrics for a year, two years on that experiment. I'll bet you'd be surprised how many times the metric is different than what you thought it would be.

— Archie Abrams

Shopify's experimentation infrastructure is built to surface this. They run two layers of holdouts: one that holds back 5% of users from all changes in a quarter, and another for new-merchant experiments where they split cohorts 50/50, ship the winner to 100%, but track the original assigned groups for years. The system forces humility. It trains the growth team to distrust short-term wins and to treat activation metrics as noisy proxies, not truth.

The flip side: occasionally an experiment that looked neutral or even slightly negative early reveals a pocket of high-value merchants downstream. The most common unlock is reducing monetary friction—offering a discount or trial extension that gives a bootstrapped entrepreneur just enough runway to find product-market fit. Standard wisdom says discounts attract low-quality users. Abrams has seen the opposite when the discount causally changes someone's ability to succeed.

The Danger of Letting Teams Own Funnel Stages

Abrams is skeptical of the way most organizations carve up ownership. When you assign teams to funnel stages—acquisition, activation, retention—you create a local optimization trap. Teams start treating their slice of the conversion rate as the north star. The easiest way to improve conversion at your stage is to make the prior stage harder, filtering for higher intent. You look good. The business suffers.

When you have teams naturally break up the world into different funnel stages or different points in the journey, it gets very seductive to look at my part of the funnel and what's my conversion rate through that part of the funnel. In practice, it's almost always easier to just make it harder to do the thing right before your step in the funnel to increase your conversion rate.

— Archie Abrams

The incentive becomes perverse fast. Instead of growing the absolute number of activated users, you shrink the top of the funnel to pump your percentage. Abrams pushes his team to think in absolute numbers and total cohort value, not rates. The best way to get more people activated is to get more people in the door, even if it wrecks your conversion rate.

Shopify's organizational structure reflects this philosophy. The company splits product development across three groups with radically different time horizons. Core product builds for the 100-year vision—what commerce needs a century from now, driven by CEO Tobi Lütke's taste and intuition, not KPIs. Merchant services builds payments, shipping, and other revenue-driving tools on a medium-term horizon. Growth owns the end-to-end customer journey and is the only group explicitly optimizing metrics, but even they measure success in multi-year GMV curves, not monthly dashboards.

What Happens When You Ban KPIs From Core Product Teams

Core product teams at Shopify don't have KPIs. Metrics are essentially banned. Decisions are made on taste, intuition, and alignment with the long-term vision. This sounds like chaos until you realize it's a deliberate firewall against short-termism.

The growth org—over 600 people across product, design, engineering, data, ops, and growth marketing—exists partly to absorb the accountability pressure that would otherwise warp core product's time horizon. Growth can run experiments, chase leading indicators, and move fast on conversion wins. Core product can ignore the noise and build the foundational stuff that takes years to pay off but defines the platform's durability.

This creates tension, but productive tension. Growth has to justify its work with data and long-term GMV impact. Core product has to justify its work with vision and conviction. Neither can colonize the other's territory. The system works because Shopify is willing to let different parts of the org operate under incompatible principles.

Shopify, we very purposely set up different parts of the org to think on very different time horizons and with very different ways of thinking about how to build product. Very different than a lot of companies that typically have one unified north star that the entire company is rallying around.

— Archie Abrams

The Practical Advice: Ship Fast, But Measure Long

For teams that can't wait years to validate experiments, Abrams offers pragmatic guidance. Don't let the perfect be the enemy of the good. If an experiment shows short-term lift, ship it. You're probably not hurting the business. But don't overclaim credit. Instrument as deeply into the funnel as possible and measure for as long as you can afford. Get rigorous about defining early signs of real success—behaviors that correlate with long-term value, not just activity.

The dangerous pattern isn't shipping too fast. It's mistaking motion for progress and building a growth culture that rewards vanity metrics. Abrams has seen neutral experiments turn positive after a year, but he's never seen a negative short-term result flip positive long-term. So velocity is fine. Just be honest about what you actually know.

What separates Shopify's approach isn't access to better data or a larger team. It's the willingness to build systems that tell uncomfortable truths. To run holdouts that make your wins look smaller. To organize so that short-term and long-term thinking don't cannibalize each other. To accept churn as a feature, not a bug, when your business model rewards outliers. Abrams isn't running a typical growth org. He's running a multi-decade experiment in what happens when you take the long view seriously.

There's a lot of long-term monitoring of experiments over very long time horizons to both inform what those input metrics are, and more importantly, hold ourselves accountable to: Did we actually move what we cared about, which is that long-term GMV, in the right way?

— Archie Abrams