Your Landing Page Doesn't Need a Redesign — It Needs a Diagnosis

The brief nearly always arrives in the same shape. The landing page is underperforming, so could I take a look at the design. By the time someone writes that sentence, they have already decided what the problem is — and they have decided it is visual, because visual is the only layer most people can see.

In most of the pages I audit, the design is not the problem. The layout is competent, the page loads, the buttons sit where thumbs expect them. What is missing is a reason to act that the visitor did not already have before arriving. A redesign does not supply one. It rehouses the absence in better typography and buys the team six weeks of feeling productive.

Three different failures, one identical symptom

A conversion rate is a ratio, and nearly every argument about it concerns the numerator. The denominator is where the problem usually hides. Before touching a single element, you need to know which of three failures you are actually looking at, because the fix for each one makes the other two harder to see.

A traffic-quality problem

The page is converting the people it deserves. The traffic is not buying-shaped — wrong intent, wrong stage, wrong country, or an ad that promised something the page does not sell. The tell is that the overall rate moved while every individual segment stayed flat. The mix changed, not the page. Pull your rate by campaign, by placement, by device, by landing query. If your best source converts four times better than your worst and that gap has been stable for months, you do not have a page problem. You have a sourcing problem, and it belongs in how the media is bought and targeted rather than in the hero section.

A message problem

People arrive, understand roughly what is on offer, and are not moved. Or worse, they cannot work out what is on offer at all, and leave politely rather than confused. Message problems do not produce a neat drop-off point. They produce long dwell times, deep scroll, a lot of reading and no action — people hunting for the sentence that makes it make sense and never finding it. This is the most common failure and the least visible one, because nothing in the analytics looks broken.

A friction problem

The visitor has decided, and the page makes acting expensive. The tell is unambiguous: an action started and not finished. Form abandonment at a specific field. Checkout collapse on mobile only. A validation error nobody can see on a small screen. Friction is the only one of the three that produces a locatable step, which is exactly why it gets over-diagnosed — it is the one failure a dashboard can point at.

The order you run the diagnosis in

The order matters more than the tooling. Each step rules out an explanation so that the next step is interpretable.

1. Segment before you theorise. Split the rate by source, by device, and by new versus returning. Half of all "the page broke" conversations end here, when it turns out paid social quietly tripled as a share of sessions.
2. Verify the measurement before you trust any of it. If the platform reports 340 conversions and the CRM holds 180, nothing downstream is worth debating. Fix attribution first — tracking you can reconcile against what finance actually banked is the precondition for every other decision on this list.
3. Watch hesitation, not exits. Session recordings are poor at telling you why someone left and excellent at showing you where they stopped moving. Look for the scroll that goes back up. Re-reading is a message signal, not a layout signal.
4. Read the page cold. Hide everything below the first screen and ask someone outside your industry what you sell, who it is for, and what happens if they click. If they cannot answer all three inside five seconds, you have found it.
5. Only now look at the mechanics. Fields, steps, load time, error states, tap targets.

Most teams run this backwards. They start at step five, because step five is the only step that converts neatly into a ticket an engineer will accept.

Form length is rarely the culprit — the size of the ask is

"Shorten the form" is the most reliably wrong advice in this discipline. People will hand over eleven fields for a mortgage quote and abandon after four for a newsletter. What they are weighing is not effort. It is exposure — what they believe happens after they press the button.

A phone-number field does not cost thirty seconds. It costs the expectation of an unwanted call on Tuesday afternoon. Remove it and submissions may rise while lead quality falls through the floor, which is a worse outcome dressed as a win. Three things are usually worth more than deleting fields:

• Say what happens next, in specifics. "I reply personally within one working day, by email, and there is no call unless you ask for one" will outperform any button treatment you can think of.
• Shrink the commitment, not the form. A page offering a demo and a page offering a teardown of your current setup can ask identical questions and perform nothing alike.
• Delete any field nobody reads. If you cannot name the person who uses the answer, that field is costing you conversions to decorate a spreadsheet.

When the ask itself is the problem, the fix sits upstream of the page. It becomes a question of what the market already believes about you and what it would take to be worth a reply at all, which is a marketing analysis question before it is a layout question.

Most pages cannot afford the test they want to run

Here is the arithmetic that quietly kills most testing programmes. A workable approximation for the sample you need per variant is 16 × p × (1 − p) ÷ d², where p is your current conversion rate and d is the absolute improvement you want to be able to detect.

Run it once and the room goes quiet. At a 2% conversion rate, detecting a 20% relative lift means detecting 0.4 percentage points — roughly 19,600 visitors per variant. Call it 40,000 sessions for a straight A/B. If the page sees 3,000 a month, that test runs for over a year, by which point your audience, your prices and your competitors have all moved. So you stop it early, read the noise as a result, and ship it.

Testing button colours while the headline is wrong is sanding a door that opens onto a wall.

Low-traffic pages are not exempt from improvement. They are exempt from split testing. Below that threshold, make large changes — a different promise, a different offer, a different order of argument — and judge them on direction across a defined window alongside qualitative evidence. The reason to make big swings is not courage. It is that only large effects are detectable at small sample sizes. Which also makes the colour test doubly wasteful: it is the change least likely to produce an effect you could ever measure, run on the page least able to measure it.

What to do this week

• Export the last 90 days segmented by source and device. Establish whether the rate fell or the mix shifted. Twenty minutes, and it decides everything that follows.
• Reconcile one full month: platform conversions against CRM records against invoices. Write the three numbers side by side. If the gap is over 15%, stop all other work until it is not.
• Send the page to five people outside your industry. Ask only what you sell, who it is for, and what happens if they click. Record the words they use, not a score.
• Count the monthly sessions on that page, run the sample-size arithmetic above, and decide honestly whether you are entitled to A/B test at all.
• Rewrite the headline and the sentence beneath the button against the single objection your sales replies answer most often. If you are under-powered, ship it as a change rather than a test, and say so in the log.

That sequence is most of the conversion rate optimisation work I take on, and the majority of the value usually lands in the first two steps, before anything on the page has moved. The redesign, on the occasions it is genuinely warranted, is far easier to brief once you know precisely what it has to fix.