How do you prioritize growth experiments?

Prioritize growth experiments by their distance from revenue, cost to run, and speed of feedback. The experiments most likely to move you toward profitability—or reveal why you won't get there—belong at the top of your queue. Everything else is a distraction that burns runway.

Revenue proximity is the primary filter

The first question to ask about any growth experiment is: if this works, how quickly does it convert into dollars? An experiment that might improve top-of-funnel brand awareness six months from now is categorically different from an experiment that tests whether a new pricing page converts 15% better this week. Early-stage startups should almost always bias toward the second type.

Paul Graham's analysis of fundraising phases makes a relevant structural point: investors in later rounds need to see that your initial experiment—your whole company—has actually worked, and that proof is most often denominator-level profitability or a credible trajectory toward it. The same logic applies internally. When you run growth experiments, the ones that test your path to profitability deserve priority over the ones that test awareness or virality loops that might eventually contribute to revenue quarters from now.

In practice, this means mapping every experiment candidate to a specific metric: conversion rate on the checkout page, activation rate on day one, average contract value on outbound sequences. If you can't draw a straight line from the experiment result to a revenue number within 30–60 days, deprioritize it until your core monetization is proven.

Urgency is a forcing function, not a feeling

One of the most underappreciated failure modes for growth teams is the drift into comfortable experimentation—running tests that feel productive without creating meaningful urgency around the result. Paul Graham's observation that startups fail between funding phases partly because they let unprofitability become habitual is directly applicable here: when there's no deadline attached to a growth experiment, teams naturally underweight the cost of a null result.

To counteract this, attach a time-to-decision constraint to every experiment before you run it. Not a deadline for the experiment itself, but a deadline for the decision it is supposed to inform. If your paid acquisition test doesn't produce a CAC:LTV signal within three weeks, what changes? If nothing changes operationally based on the result, the experiment should be questioned before resources are committed.

This urgency framing also helps you identify experiments that are actually decisions in disguise. Sometimes founders run A/B tests on landing page copy when what they really need to decide is whether this channel works at all. The experiment with the faster decision payoff wins, even if its statistical confidence is lower.

Score experiments on a three-axis grid before committing

A simple scoring method beats elaborate prioritization frameworks because it forces you to compare unlike things on common ground. Before greenlit an experiment, rate it on three axes: expected impact on a key metric (high/medium/low), cost to execute including engineer time and distraction cost (high/medium/low), and speed of feedback—how many days until you have a usable signal (fast/medium/slow).

The highest-priority experiments score high on impact, low on cost, and fast on feedback. In reality, no experiment scores perfectly on all three, which is where founder judgment earns its keep. A high-cost, high-impact experiment might be worth running if it answers a question that unlocks your next funding round. A fast, cheap experiment with modest impact still belongs early in the queue because it trains your team's experimental muscle and costs almost nothing if it fails.

The scoring process also surfaces hidden costs that often get ignored: what does this experiment break if it wins? A successful experiment that you're not ready to scale is almost as wasteful as a failed one. A pricing experiment that triples conversion but requires rebuilding your billing infrastructure shouldn't be run until you know you can absorb a win.

Sequence experiments to build on each other, not compete

Most teams run experiments in parallel when they should run them in sequence. Parallelism feels efficient but produces noise: when three variables change at once, you don't know which one moved the metric. More importantly, sequential experiments compound—each result informs the design of the next, and over time you build a model of your users that guides prioritization instinctively.

Start with experiments that establish baselines for your most critical funnel stage. Before testing acquisition channels, know your activation rate. Before testing activation copy, know which user actions actually predict retention. The sequence matters because growth levers upstream are useless if the downstream stage is broken. Sending more traffic to a funnel with a 2% activation rate produces a lot of expensive unactivated users.

A useful sequencing heuristic: run experiments on the stage of the funnel with the largest absolute drop-off first. That's almost always where the highest leverage lives, and fixing it makes every other experiment more interpretable. Once you've established that users who complete a specific action retain at 3x the rate of those who don't, you can design five acquisition experiments around getting users to that action faster.

Kill experiments faster than you start them

The discipline that separates productive growth teams from busy ones is aggressive experiment killing. Most teams have a bias toward letting experiments run longer—'we need more data'—when the real issue is that the early signal is disappointing and no one wants to call it. This is expensive. Every week an inconclusive or failing experiment runs is a week your team's attention is split and your roadmap is blocked.

Set a minimum detectable effect before you start. If your experiment can't move the metric by at least X, it's not worth running. Calculate the sample size that would give you confidence in that effect, then calculate how long it will take to accumulate that sample at your current traffic. If the answer is longer than four weeks, either the experiment isn't the right one right now, or you need to find a version of the question you can test on a smaller cohort.

The broader principle is that growth experiments are only valuable as inputs to decisions. A test you run but don't act on—because the results were ambiguous, because the team changed direction, because the winning variant required too much engineering—was a waste. Before you start any experiment, write down in one sentence what you will do if the result is positive and what you will do if it is negative. If you can't write both sentences, you're not ready to run it.

“The next time you raise money, the experiment has to have worked. You have to be on a trajectory that leads to going public.”

— Paul Graham, source

The one thing to do

Before you run any growth experiment, write down exactly what decision you will make if it succeeds and if it fails—if you can't write both sentences, deprioritize it.

Frequently asked questions

How many growth experiments should a startup run at once?

Most early-stage teams should run one to two experiments at a time. Parallelism creates interpretation noise and splits the team's attention. Sequential experiments compound into a real model of your users; parallel ones mostly create ambiguity.

Should you prioritize acquisition or retention experiments first?

Retention almost always comes first. If users don't stick, acquisition experiments just fill a leaky bucket faster. Fix the stage of your funnel with the largest absolute drop-off before spending on top-of-funnel growth.

What's the minimum viable experiment for a cash-constrained startup?

Any test you can run with existing traffic, no engineering changes, and a decision point within two weeks. Concierge tests, manual outreach sequences, and pricing page copy changes often qualify. If it requires a sprint to instrument, it's not minimum viable.

How do you prioritize when everything seems equally urgent?

Map every candidate experiment to its revenue proximity—how many days between a positive result and incremental dollars. The experiment with the shortest path wins. If two tie, pick the one that answers a question your next fundraise will require you to have answered.

Sources

More playbook answers · Growth Prophet home