Every company ultimately wants to influence revenue. But depending on your monetization model and sales motion, revenue may be a lagging indicator that takes 6–12 months to show up after an experiment runs. Before you test anything, align your team on the metric that will actually tell you whether an experiment worked in a timeframe you can act on.

If you're a complex enterprise SaaS with a 6-month sales cycle, your experiment metric is probably new qualified meetings or 1-year ACV pipeline, not closed revenue. If you're a PLG product where users can upgrade within hours of signing up, you might measure free-to-paid conversion or trial activation rate within 72 hours. If you're somewhere in between, you might track demo requests or sign-up starts as your primary proxy.

There's no universal right answer. The right answer is the one your whole team agrees to measure before the experiment launches, and doesn't change mid-flight.

You cannot accurately measure what you haven't tracked. Before launching any experiment, audit your full attribution stack: UTM parameter coverage, conversion event configuration in GA4, CRM attribution logic, and ad platform pixel health.

The critical thing most teams skip is quantifying their expected data gap. No tracking setup is 100% accurate, and pretending it is leads to false conclusions. In most industries, UTM parameter coverage runs at 60–80% of actual traffic. But for some ICPs, cybersecurity professionals, developers, enterprise IT buyers, ad blocker rates can be 30–50%+, which means your measured traffic and conversions may represent significantly less than half of what's actually happening.

Spend time before your first experiment understanding what your coverage looks like. If your tracking captures 65% of conversions, a result that looks flat might actually be a 20% lift you're not seeing. Set thresholds accordingly.

Every experiment in this guide is more effective when it's aimed at a specific, well-defined audience. Before you start testing, make sure you have a clear, written definition of your Ideal Customer Profile, both at the individual level and the organizational level.

At the individual level: job title, seniority, department, and the specific pain they're trying to solve. At the organizational level: company size, industry verticals, tech stack signals, growth stage, and any other firmographic characteristics that predict a good fit.

The more tailored your experiment experience is to that specific person, the better it will perform. "Anyone at a 50–5,000 person company" is a universe, not an audience. "VP of Engineering at a Series B fintech company using AWS with a team of 20–100 engineers" is an audience you can write for, target precisely, and measure accurately.

Before testing a new variant, you need something to compare against. Capture current conversion rates, traffic volumes, open rates, CPL, or whatever metric your experiment will influence, for a minimum of 2 full weeks before you launch.

Without a baseline, a result is just a number. With a baseline, it becomes a signal. If your homepage is converting at 2.3% today and your new hero variant drives 3.1%, that's a 35% lift worth understanding and building on. If you didn't know the 2.3%, you have no context for the 3.1%.

This applies even to experiments where you're not running an A/B test, if you're launching a new nurture sequence, know your current lead-to-opportunity rate before you turn it on.

Ending an experiment early because it looks good (or bad) is one of the most common and costly mistakes in growth. Define your success criteria before the experiment starts, and commit to not acting on results until those criteria are met.

At minimum: set a minimum run time of 2 full business weeks to account for weekly seasonality. Calculate your required sample size upfront — if you only have 50 visitors per day, a 2-week test may not give you enough data to be conclusive, and you need to know that before you start.

On confidence thresholds: the industry standard is 95%, and that's the right bar when you have the traffic to reach it. But don't let perfect be the enemy of useful. If you're at an earlier stage with lower data volume, there's nothing wrong with acting on 80% or 90% confidence, especially when the cost of waiting for 95% is weeks of additional run time and the cost of being wrong is low. Be honest with yourself about which decisions warrant the higher bar and which don't. A homepage hero test on a high-traffic site should clear 95%. A headline test on a page with 30 visitors a day probably shouldn't wait that long.

Also define what you'll do with inconclusive results. "No significant difference" is a valid and useful result — it tells you that the variable you tested doesn't materially affect conversion, which is worth knowing.