Decompose the Growth Loop

The general principle

Eric Glyman describes how Paribus, his company before Ramp, was actually grown as the marketing instance of a broader rule: an output cannot be managed directly, only its inputs.¹ A coach cannot tell a team that today they practice to win the Super Bowl; the team lifts, sprints, and drills plays, and the score takes care of itself. Applied to growth, the method is to write the desired output as an equation, break it into the small number of discrete moments a user actually passes through, treat each moment as a coefficient, and optimize the coefficients one at a time.

The enabling constraint

Paribus had almost no user experience to speak of: a user would sign up, link their email, and that was effectively the entire product. Glyman calls this an amazing, if accidental, constraint, because with so little surface area the entire business collapsed into a small number of moments that could each be isolated and improved independently.

The four moments

The first moment is the homepage: an app nobody has heard of, doing something with no existing category, whose job is to articulate a compelling idea in as few words as possible and create desire. The second is the screen where a user links their Gmail account, letting software read their email and send mail as them; the only thing that matters at this stage is instilling trust, and everything else on the screen gets deleted in service of that one goal. The third is onboarding, where the referral ask lives, because intent is highest before any value has actually been delivered. The fourth is the savings moment, roughly a week later, when one hundred dollars arrives unprompted for a television the user had already bought; the design question there is how to make that moment so joyous that people want to tell everyone about it immediately.

The loop then closes on itself: what does a referred friend see, and how do they land back on the homepage. Each moment carries its own coefficient, the product of the coefficients is the resulting k-factor, and the entire growth effort for a summer reduced to two questions: how to get press and coverage, and how to make every user produce more than one additional user.¹

Testing when there is no traffic

The obvious objection is that a business cannot A/B test with ten users. Glyman's answer is to manufacture a discrete external event that produces a burst of arrivals, a Reddit post or a local news segment, treated as a shock wave of two hundred to five hundred users at once. The team would measure how many new users that specific cohort generated, in the early days roughly two hundred users produced about twenty new ones, change one variable, and fire the next wave.¹ This converts a continuous, too-noisy metric into a series of discrete cohort experiments, and it uses press as an instrument for testing rather than only as a growth channel in its own right. The target moved from twenty referred users per two hundred, to forty, to sixty, to eighty.

The best-known single result to come out of this program is the finding that loss aversion beats gain framing in referral copy.

Why it is a general method

Glyman's own summary: understanding the output a business is trying to drive, but really decomposing the inputs at every step, allows much more interesting questions to be asked. The decomposition is what makes a question answerable. Asking how to grow faster has no experiment attached to it; asking whether a friend converts better when a referral removes their fee or adds to their savings does. Ramp's later practice of reducing the business to a single variable, purchase volume, and optimizing that one number is the same instinct applied at company scale.