Conversion Rate Optimization

When growth was measured by traffic alone, there was only one job to do: bring in more people. A month with 40,000 visits and 1,200 signups, facing a target of 2,400 signups next quarter, produced a plan to reach 80,000 visits. The answer always came out of the advertising budget.

That approach stalls the moment ad prices rise. When the cost of bringing in one person goes from 8,000 won last year to 14,000 won this year, holding signups at last year's level takes close to double the budget. So the meeting kept its shape. Grace would ask what it would take to double signups, Ravi would answer that at current prices it would take double the ad spend, and the only moves available in that room were raising the budget or lowering the target.

Meanwhile one number sat untouched. If 40,000 people arrive and 1,200 sign up, then 38,800 people left without doing anything. The share of visitors who take the intended action is the conversion rate, and here it is 3 percent. Look one layer in and the picture gets sharper: 5,000 people reached the signup screen and 1,200 finished. A single screen loses 3,800 people.

Touch that screen and the result changes without a single extra visitor. A signup form asking for name, company, role, team size and password can drop to email and password, with the rest asked later. The share of people who get through that screen moves from 24 percent to 34 percent. The same 5,000 arrivals now produce 1,700 signups instead of 1,200, and the 500 extra signups cost nothing in ad spend.

Conversion rate optimization broke off as its own job to cover that ground. It handles different material from the acquisition side: not ad creative or keywords, but the order of screens, the number of fields a form asks for, the sentence printed on a button, the seconds a page makes someone wait. It argues from different evidence too. Instead of how many people arrived this month, it works from a record of where people dropped out, and it picks the next thing to fix out of that record.

The terms filed under conversion rate optimization fall into four groups: finding where the blockage is (diagnosis), deciding which blockage to fix first (target), removing obstacles (methods), and nudging people who hesitate (persuasion).

Diagnosis is about locating the spot before touching anything. It splits in two, along the line between where the loss happens and why it happens.

1. Funnel leak — lay out the steps a user passes through in order, count how many remain at each one, and the step that loses an unusual share is the leak. In a chain of 40,000 visits, 5,000 arrivals at the signup screen, 1,200 completed signups and 700 first documents created, the pass-through rates are 12.5 percent, 24 percent and 58 percent. What you look for is not the lowest number but the number that is low relative to its neighbors. The 12.5 percent at the top is normal, since plenty of visitors came to look around. The 24 percent is not, because those are people who arrived intending to make an account and three out of four walked away.
2. Friction — what you find once you open up a leak. It names each individual obstacle that makes someone hesitate before the next step. A phone verification step, a screen asking for a card number before a free trial, a three-second wait, a field label nobody can parse: all of it is friction. A leak tells you where; friction tells you what. The two stay separate because one leak usually has several causes mixed together, so removing one of them often moves no numbers at all.

Target covers the place that receives this diagnosis most often. A signup flow is the sequence of screens someone passes through to create an account. A flow of three screens and nine fields and a flow of one screen and two fields lead to the same product but deliver different numbers of people to it. This spot gets called first because every user has to pass through it. One percentage point more through that gate lifts the count in every step behind it, and while it stays blocked, the screens further in are built for people who never arrive.

Methods cover what actually removes friction. Both terms here move in the direction of asking for less.

1. Social login — letting people in through an account they already hold, such as Google or Kakao. Four fields for email, password, name and company collapse into one button, and nobody has to invent and remember another password. It comes with a price. Users who forget which account they signed up with get stuck on their next visit, and the product may not collect information it wanted. That is why it usually ships alongside asking for the rest later.
2. Progressive profiling — collecting what you need over time instead of piling it onto the first screen. Rather than asking for team size, role and intended use at signup, the product takes an email, lets the person start, and asks about team size the first time they share a document. Nothing is given up; the question is only moved later. Someone who has already started using the product has a reason to answer, so the same question tends to get a higher response rate than it did at the door.

Persuasion pushes rather than removes. It splits in three, depending on whether the supporting evidence is borrowed from other people, borrowed from time, or produced by the product itself.

1. Social proof — signals that other people are already using this. A line saying 1,240 teams started this week, a row of customer names under the fold, a quote from an actual user. It gives someone who hesitates to hand company details to an unfamiliar product something to lean on. Small numbers backfire, though: "37 teams so far" reads as a warning rather than an endorsement.
2. Urgency — devices that make waiting feel expensive, like a trial counter showing three days left or a discount that ends this month. They pull people who would otherwise decide later into deciding today, and they break faster than anything else when the claim is not true. A discount that has ended every week for two months gets caught on the second visit, and after that nothing else on the page is believed either.
3. Value first — designing the flow so the result arrives before the request. Letting someone build one document with no account and asking who they are only when they try to save it is the standard shape. Where the first two persuade with words, this one persuades with an experience. The ask lands at the moment the work already done would be lost, and the same request gets a different answer there.

The four groups run in order. Diagnosis points at a spot, target picks one of those spots, and methods and persuasion supply the handles to turn there. Skip the order and the failure looks the same every time: a screen loaded with testimonials, countdowns and customer logos, while the step that was actually losing people sits untouched.

"The signup screen alone loses 3,800 people. Let's start there this time."

The leak table is being used to narrow this round of work down to one place.

"Let's pull two fields out first, and add social login only if that isn't enough."

Start with the smallest change you can undo, and check it before adding another. Change several things at once and there is no telling which one did the work.

"Signups are up but the number of teams that reach checkout hasn't moved."

The gain in one step did not carry into the next. The suspicion is that the change did not remove friction so much as wave through people who were never going to buy.

The two weeks Pagedeck spent on its signup flow is the ordinary version of this work. Ravi put the funnel table on screen with the remaining count next to each step, and the meeting's attention moved off the ad budget and onto the signup screen. Noa cut three screens down to one and pushed everything except email and password to a point after the first document, and Theo shipped it to half of new visitors.

Two weeks later the signup screen was passing 31 percent instead of 24, and signups had gone from 1,200 to 1,550. Mia called one more meeting because nobody yet knew who the extra 350 people were. In the same session Grace settled the question of what counted as success: not signups, but how many teams were still writing documents two weeks in. Only after that number held did the change go out to everyone.

The most common misreading is that this work means testing button colors and wording. Most of the stories that travel have that shape, so it is an easy conclusion to reach — a green button beat an orange one by some percentage is short, quotable, and carries well. The real sequence starts at the other end. Count where people are dropping out, find what is in their way at that step, and then build the change that removes it. Color is the last kind of material in that process, if it shows up at all. Reverse the order and a team runs twenty small variations across the site while the conversion rate goes from 3.0 percent to 3.2, and the fact that three of every four people abandon the signup screen never comes up.

Reading a higher conversion rate as automatically better is the other common mistake. It is the number being raised, so higher looks like progress. But the number also rises when people are pushed through. Add a countdown, shrink the cancel button, blur the billing terms, and the month's rate moves from 3 percent to 5. Those extra people were pressed rather than convinced, so refund requests and first-month cancellations rise with them. One step's gain gets paid for by the step behind it. Teams working on conversion therefore read the rate next to whatever follows it. If signups climbed and the count of teams still active two weeks later did not, that change is a candidate to roll back rather than a win.