Last autumn, in a meeting, an e-commerce manager turned their laptop toward us. On the ad dashboard the budget had doubled in six months, and the clicks had doubled with it. The sales curve sat exactly where it had always sat. Their question was one sentence: "How much more traffic do I need?"
It was the wrong question. The right one came in three parts: what share of arriving visitors buy, where exactly do the rest give up, and which of those give-ups can be fixed? The work that chases those three questions is called CRO. I invented the meeting for this article; I did not invent the question — we have heard that same sentence from different industries many times over the past few years.
What is CRO?
CRO — conversion rate optimisation — is the discipline of measurement, hypothesis and testing that turns a larger share of the visitors a site already has into customers. It does not buy traffic; it raises the yield of the traffic in hand. The method has three parts: find where the visitor stops using data, build a testable hypothesis about why they stopped, then examine that hypothesis with a controlled experiment. Its output is not a new design but a verified piece of knowledge.
What this definition leaves out matters as much as what it includes. CRO is not the business of making a page prettier — aesthetics can be a by-product, never the goal. Nor is it one person saying "I think that button is better"; one person's opinion is a hypothesis, and a hypothesis is not knowledge until data tests it. And it is certainly not a one-off project: as visitor behaviour, competition and price bands shift, the answer that held last quarter stops holding, so the work never finishes.
What is a conversion rate and how is it calculated?
A conversion rate is the number of visitors who complete the targeted action in a given period, divided by the total number of visitors. The formula is plain: conversions ÷ sessions × 100. If three out of a hundred sessions place an order, your website conversion rate is three percent. The difficulty is not the formula but filling in both sides of it honestly.
What you put in the denominator changes the answer. Count unique users instead of sessions and the rate climbs; leave bots and internal traffic in and it sinks. When mobile and desktop are not calculated separately on the same site, a healthy desktop experience can hide a broken mobile one for weeks. This is why a single e-commerce conversion rate cannot be the headline of a report; a rate without a breakdown is a problem hiding behind an average.
Then there is the choice of action. A purchase is the macro conversion, but not the only one: add-to-cart, form submissions, quote requests, newsletter sign-ups and account creation are all measured too, and these are called micro conversions. In businesses with long sales cycles, micro conversions are the only early signal — if you wait for the signature alone in a three-month quoting process, you will never see the effect of anything you improved in between.
What counts as a good conversion rate?
The honest answer: it depends on the sector, the price band, where the traffic comes from and how the measurement was set up; there is no single "good" number that holds for everyone. When you see an industry average circulating online, ask one thing: how many stores was it collected from, in which country, at which price band, with what traffic mix? If there is no answer, there is no number.
Picture two stores in the same category. One takes most of its traffic from brand searches, the other from broadly targeted display ads. The second one's rate comes out structurally lower, and that is not a failure but a different mix of intent. Putting a store selling thousand-lira items next to a manufacturer selling hundred-thousand-lira machines is the same error — the longer the decision takes, the lower the rate, because the same person visits five times before buying.
The only usable benchmark is your own history. Measure your baseline properly, split it into its parts, then set out to beat your own number. That presumes the measurement actually works: at SOYLU AVM the pixel setup was incomplete, and neither the source of traffic nor the path to conversion could be traced. Because the rate on the screen was wrong, trying to improve it meant nothing.
Which is cheaper: more traffic or better conversion?
You can get the same lift in sales two ways — by doubling traffic or by doubling the conversion rate. The difference lies in the shape of the cost: traffic bills you again every month, while a conversion improvement is paid for once and then applies to all the traffic that follows. That is why the return on CRO compounds while the return on ads stays linear.
Two metrics make that difference visible. CAC — customer acquisition cost — is the total you spend to win one customer, and it falls when the conversion rate rises without a single extra lira of spend. ROAS — return on ad spend — shows the revenue earned for each unit spent, and it moves up for the same reason. CRO is therefore not one line in the marketing budget but the multiplier on all the others.
Raising traffic means pouring water into the bucket faster. Raising conversion means closing the holes in the bucket — and while the holes are open, the speed you pour at hardly matters.
The conclusion is not "stop advertising". You cannot run tests on a site with very little traffic, so investing in traffic up to a threshold is unavoidable. But when traffic already exists and sales are standing still, putting the next lira into ads is usually the most expensive option available.
The CRO process: measure, hypothesise, test, learn
The answer to "how do you increase a conversion rate" is not a list of tactics but an order. That is precisely what separates CRO from a list of tactics: when the order breaks — idea first, data later — what you are left with is a collection of guesses. The four steps run like this.
- Measurement. Verify the analytics setup, define the goals, split out the segments. The aim at this step is not improvement but certainty that the number on your screen is real.
- Diagnosis. Quantitative data tells you where you lose people, qualitative data tells you why. Session recordings, form abandonment analysis, support tickets and short interviews with five users are the raw material of this step.
- Hypothesis. Every hypothesis fits in one sentence: if we change this, that metric moves by this much, for this reason. A hypothesis whose reason you cannot write down teaches you nothing, whatever its result.
- Test and learn. Build the experiment, run it for the period agreed in advance, and record the outcome — won, lost or no difference. Losing tests are knowledge too; knowledge left unwritten gets retested six months later.
Each turn of this loop makes the next one cheaper, because diagnosis accumulates. If you are wondering how each element on the page itself should be built — headline, call to action, form length, speed — our landing page optimisation article walks through that layer one item at a time. This article gives you the concept; that one gives you the levers within reach. What the process looks like once it becomes a team discipline is described on our CRO service page.
What is an A/B test, and when does it give a meaningful result?
An A/B test is a controlled experiment that splits the same traffic randomly between two versions and measures which one converts more. It becomes meaningful at the moment the difference is too large to be explained by chance and the test has completed the duration agreed beforehand. The slash often drops away when people search for the term; "ab testing" describes exactly the same method.
The discipline has three rules. First, one variable: change the headline, the image and the button colour at once and you will know which version won but not why — and without the why, the learning does not travel to the next test. Second, decide in advance: the duration and the sample size needed are written down before the test starts. Third, whole weeks: the user who behaves a certain way on Tuesday is not the user who behaves on Saturday, so a test is never cut off mid-week.
The most expensive mistake is looking early. When a test is ahead after three days it is tempting to stop it and declare a winner; yet at small sample sizes the gap changes direction day by day, and tests stopped early make the wrong decision permanent. Looking at the dashboard is free; deciding on it is not.
What does a test tell you when the sample is too small?
Almost nothing. On a site with a few hundred conversions a month, reliably measuring a two or three percent difference takes months, which makes running that test a waste of time. This does not mean low-traffic sites cannot do CRO; it means the size of what they test has to change.
Three routes work when traffic is thin. First, test big changes: not the button colour but the offer itself, the shipping policy or the entire structure of the page. Large differences show up even in small samples. Second, measure micro conversions: when purchases are rare, add-to-cart and reaching the payment step give a signal far sooner. Third, put qualitative research first: watching five users hesitate as they fill in the order form can answer in one afternoon what a two-month test would have answered.
There is also an honest limit: not every gain comes out of a test. A broken mobile payment step, a shipping fee that never appears on screen or a page that takes five seconds to load is not tested, it is fixed. Testing is what you reach for when two reasonable options leave you undecided; testing an obvious defect only keeps it alive for another few weeks.
What does funnel analysis show?
Funnel analysis measures, in order, the steps a visitor passes through from the first page to the order, and shows how many people drop at each one. Its value lies in one thing: the overall conversion rate tells you a problem exists, funnel analysis tells you which step it lives in. This structure separates the chain that runs from category page to product page, from basket to payment step and on to the order.
How the analysis is read matters as much. The largest drop is not always the largest opportunity — a seventy percent drop between basket and payment is ordinary, the same figure between product page and basket is not. What you are looking for is not the absolute fall but the fall that is abnormal against comparable steps or against your own history. And every step has its own segments: a payment step collapsing on mobile can vanish entirely once merged with desktop data.
At SOYLU AVM we began at exactly this point. Pixel and conversion tracking were rebuilt from scratch and traffic sources were segmented; only then did the campaign go live. By the sixth day of the campaign total revenue reached $1.5 million and total traffic rose by 150 percent. The lesson is in the order: a campaign launched before measurement is in place reports spending, not results.
Why is the cart abandonment rate so high?
Cart abandonment is the share of visitors who add an item to the basket and never complete the order, and it runs high in every store. Part of that height is behavioural — people use the basket as a shopping list, a price comparison tool or a bookmark — and that part cannot be closed. The part that can be closed is the surprises the buyer meets at the payment step.
- Cost that appears late. When the shipping fee, tax or service charge shows up for the first time at the payment step, the buyer leaves not because the price changed but because their trust did.
- Forced registration. Being unable to order as a guest is a needless barrier for a one-time buyer; the invitation to create an account can wait until after the order.
- A long form. Every field not required for delivery is another chance to abandon; you should be able to defend why each piece of information is being asked for.
- A missing payment option. When the method the buyer is used to is not on the list, a purchase already decided on dies at a technical obstacle.
- Unclear delivery and returns. If "when will it arrive" and "what if I don't like it" go unanswered on the page, the buyer postpones rather than takes on the risk.
What these items share is that none of them is a persuasion problem. The job at the payment step is not to sell but to remove the friction in front of someone who has already decided to buy. Emailing reminders for abandoned baskets works too, but it comes second: remove the cause of the abandonment first, then set up reminders for what remains.
How does social proof affect conversion?
Social proof lowers perceived risk by letting a buyer confirm their own decision against other people's. It works through reassurance rather than persuasion: a brand claiming its product is good is an interested party, a customer saying the same thing is not. This is why a review, a customer photo or a short experience video does the job faster than marketing copy making the same promise.
Placement decides as much as content. A testimonial section gathered on the homepage does less work than a single review sitting where the decision is made — on the product page, next to the price, immediately before the payment step. The same holds for form: an average rating is threshold information, while a three-sentence review describing the buyer's own constraint answers the objection directly.
In the GYMWOLVES case social proof was one of the load-bearing levers of the campaign. After the data flow was repaired and the conversion funnel rebuilt, video-led social proof was produced with athletes and influencers. The target was to double sales in 3 months; by the end of month three sales were up 12×, session duration had tripled and engagement had risen 8×. Engagement and duration climbing together is no coincidence: when a buyer finds a reason to stay on the page, conversion follows behind.
The six most common mistakes in CRO
Most of the mistakes we see in the field come from impatience rather than a lack of knowledge. Six of them repeat with some regularity.
- Starting before the measurement is verified. A badly built tracking setup hides the effect of an improvement and the effect of a mistake in exactly the same way.
- Copying a competitor. Their page works for their traffic, their pricing and their audience; without a hypothesis explaining why the same arrangement would work on yours, the copy is just a guess.
- Stopping a test early. The version leading on day three can be the loser by the second week.
- Changing too much at once. Knowing which version won is no substitute for knowing what made it win.
- Watching only the macro conversion. In businesses with long decision cycles the order count signals late; without micro conversions every improvement in between goes unmeasured.
- Not writing the learning down. A testing programme without records tests the same hypothesis three times in two years and starts from zero on each occasion.
What all six share is a sense of time. The return on CRO comes not from one test but from small learnings stacking on each other, which is why a programme reads over a year rather than a quarter. The difference between a team running four tests in three months and recording none, and a team running two and writing both down, shows up in the second year — by then the second team knows what works and why, and builds every new test more cheaply.
In closing: the one test you can run this week
This article's thesis fits in one sentence: the conversion rate is not a marketing metric but the multiplier on all your marketing spend — and raising the multiplier is almost always cheaper than raising the number being multiplied. Looking at the funnel before the traffic is not a preference; it is the order the work comes in.
There is something concrete you can do today: complete an order on your own site from start to finish on your phone, and write down how long each step takes. On which screen did the shipping fee first appear? How many fields did you fill in? Where did you wait? The longest wait on that list is this week's hypothesis — and testing it does not require buying a tool first.
If you are deciding whether to run this with your own team or bring in outside support, we have gathered what to look at when evaluating a CRO partner in a separate article.