Growth as Experimentation

Growth is science

George Bonaci, who runs growth at Eric Glyman's company Ramp, states the claim bluntly: "growth is just science, and most marketers are bad at science."¹ The goal of growth, in his account, is to figure out how to grow the business, usually top-of-funnel early, and the honest answer to how is that no one really knows, because every business is different. Because playbooks do not copy from one company to another, he prescribes the posture of a scientist rather than a practitioner: come in with a blank slate, form a hypothesis, run a batch of experiments, and run enough of them that something works.

The failure mode he identifies is thinking in terms of "what do I know and how can I apply it here" instead of "what is my hypothesis and how will I measure it." He attributes the difference partly to profile, suggesting the chemist or engineer brain scopes experiments in a way the writer or PR brain does not, and partly to habit.

Velocity over rigor, with a floor

Since most bets are assumed to fail, Bonaci argues throughput matters more than polish: "velocity is probably more important than getting things perfect." But he draws a hard floor, because pure sloppiness teaches nothing. "If you're just doing a bunch of sloppy things, it doesn't matter how many you run, you're not going to learn anything." His cautionary example is a declining top-channel webpage under deadline, where the team ran many changes at once on instinct rather than clean one-variable tests. They tripled conversion in weeks and hit the number, but never knew what actually worked and had to undo many of the changes later once they could test them properly. The reconciliation, he says, belongs to the growth portfolio of bets: optimize velocity when under the gun, then return and optimize rigor once you are not.

The real skill is upstream

Bonaci locates the discipline before the launch, in experimental design. The questions are what can actually be measured, what will be measured, how long until a result, and whether it will be statistically significant. A marketer either does that math, finds a data-team partner, or honestly admits the thing cannot be measured and collects qualitative data knowing the read might be wrong. The sin, in his framing, is not imprecision but failing to define the imprecision upfront. He extends this to pre-mortems and post-mortems: write out the many ways a bet could fail rather than a token three, and reserve post-mortems for genuinely unanticipated failures, since a failure you already assigned a probability to is not interesting once it happens. For a high-confidence bet, the pre-mortem catches the actual cause more than ninety percent of the time; for the biggest swings there is always a Black Swan, in his words, "I couldn't predict COVID."¹ This connects to segment-level analysis, where the post-mortem mechanics live.

No attachment to the experiment

A cultural corollary follows: no one should be attached to an experiment. If a person is emotionally tied to a tactic they have nurtured and it gets cut at three months, Bonaci treats that as a cultural problem, since "if you're not failing, you're probably not doing your job well." Attachment is only defensible when a bet was explicitly prioritized as high-confidence, in which case the honest question becomes whether the confidence itself was misjudged.

The framework is the operating system the other growth concepts plug into, from seeking alpha in growth on what to test to saturating the winning channel on what to do when one wins. Its own strain is the permanent tension between velocity and learnability, resolved only if the team actually circles back, and the fact that statistical rigor presupposes sample size and clean attribution, which the slowest, biggest bets rarely have.