Founders reach for traffic because traffic feels like progress. Buy clicks, ship content, run a launch, watch the number climb. The catch is that traffic only magnifies what your page already does. If the page leaks trust or clarity, you are paying to discover the same weakness faster.
So improving conversion rates is an evidence problem before it is a traffic problem. The work is finding where people hesitate, naming what confuses them, and changing the one thing that removes the most friction.
The metric itself is trivial: conversion rate = (conversions ÷ visitors) × 100. Everything hard sits around that formula. A page can look polished and still fail because the message is fragmented, the proof arrives late, or the visitor has to choose between four buttons before they understand the offer.
More traffic will not fix a leaky page
Weak conversion usually gets diagnosed as a top-of-funnel problem. It rarely is. If visitors land and cannot work out what the product does, more visitors just means more people to lose — at a higher cost per loss. Treat the page as an experiment, not a billboard.
If you can't say why a change should convert better before you ship it, you're not optimizing. You're decorating.
This is where redesigns get dangerous. A redesign feels productive because it produces visible output, but a redesign without a hypothesis is a mood swing with a Figma file attached. The version you ship next should answer a question you wrote down first.
The loop: measure, analyse, hypothesise, test
The workflow is boring on purpose. Measure a baseline, analyse where people actually drop, predict what one change will do, test it, keep the winner. The value is in following the order instead of guessing your way to a win.
Pick one primary metric before you touch anything
One metric, chosen in advance. If you watch signup clicks, scroll depth, form starts and pricing visits at once, something will move and the team will call it a win. Signups for a self-serve product. Demo requests for sales-led. Nothing else gets a vote.
Then segment the baseline by device, traffic source and new versus returning. Not to build a dashboard — to see where the friction concentrates. A page can convert acceptably on average and be broken on mobile, or fine for warm traffic and hopeless for cold.
Read behaviour, not averages
Funnel data tells you where people leave. Session replays tell you how they behaved just before leaving. Form analytics tell you which field they stall on. Averages hide all three. Instinct is worse than any of them, because instinct chases the loudest complaint rather than the most expensive one.
Then write the hypothesis, and make it specific enough to be wrong. "Simplify the signup form" is a mood. "Cut the form from eight fields to four, because first-time visitors are abandoning at the company-size field" is a test. If you cannot check it cleanly, it does not belong in the backlog.
Find the bottleneck that blocks the most people
Not every page deserves attention. A win on a page nobody sees is a hobby. Look for the overlap between the biggest drop-off and the biggest audience — that is where a small improvement compounds. One to three bottlenecks is the whole list; anything longer is procrastination dressed as prioritisation.
| Where people drop | What it usually means | First thing to test |
|---|---|---|
| Above the fold, seconds in | The offer is not legible fast enough | Headline and the first visual |
| Mid-page, after scrolling | Proof arrives too late to matter | Move evidence up, cut the feature list |
| At the CTA | The ask feels bigger than the value shown | CTA wording and what sits next to it |
| Inside the form | You are asking for more than you have earned | Field count and field order |
| On mobile only | Layout or weight, not messaging | Mobile hero, asset sizes, tap targets |
Teams tend to start with the page they personally dislike. CRO is not a design critique. Rank friction points by how many real visitors hit them, fix the top one, then move.
Cut friction before you argue about button colour
Landing pages lose people for predictable reasons: they ask too much, say too little, load heavy, or offer too many exits. Give the visitor one path. A page with three competing asks forces a choice before the value has landed, which is backwards — make the value obvious, then give them one action.
Weight belongs in this conversation too. Compression, caching, a CDN and lighter assets are conversion work, not just technical housekeeping, because a hero that arrives slowly is a hero nobody read. The same goes for anything you embed: if a video on the page is a heavyweight file with a long runtime, it is costing you the thing it was supposed to win.
Most of the remaining leverage is in hierarchy rather than aesthetics. Headline, subhead, visual proof and CTA should all point at the same outcome. When they do not, visitors feel the disconnect without being able to name it, and they leave with a vague sense that the product is not for them. The three-second hook applies to a page as much as to a video: whatever loads first has to say what this is.
Visual proof is the piece founders under-invest in. Copy explains well and reassures badly. A short, silent teaser sitting near the primary CTA shows the product working before anyone commits to reading — which is a different job from the explainer most teams reach for. Placement, scripting and testing for that format are covered in the 15-second landing page video, and video content for startups covers where else the same asset earns its keep.
A/B testing when you do not have much traffic
Good tests are dull. One question in, one defensible answer out. Before you ship an experiment, run the list:
- One hypothesis, one sentence — the change, the reason, the expected effect.
- Sample size decided in advance — so the finish line is not moved by a good-looking Tuesday.
- One variable — headline, CTA copy, layout, or the presence of a teaser. Not all four.
- Full business cycles — a week is the floor; B2B traffic usually needs a month to shake off weekday and seasonality bias.
- One primary metric — everything else is a guardrail, not a victory lap.
- Written result — a win you cannot explain later is a win you will accidentally undo.
Our own bias, stated plainly: do not call a test until it clears roughly 95% confidence, and do not peek and stop early because the trend looks good. Early stopping is how teams ship changes that quietly stop working in production. If your traffic cannot reach significance in a month, stop A/B testing small elements and go test something big enough to see — a different hero, a different offer, a different first asset.
That is also the honest argument for testing a teaser rather than debating one. Whether visual proof above the fold helps your page is not a taste question, and it is cheap to answer. If it is worth trying, how to make promotional videos covers the format; then run it as variant B against the page you have.
Once the rhythm exists, conversion work stops being emotional. You argue less about opinions and learn more about what your audience actually responds to — which is the only durable way to improve conversion rates without thrashing the page every week.