When to Start CRO: 3 Stages of DTC Growth

WRITTEN BY

Gary Culaev, Founder and CEO, Weblics Consulting

When to Start CRO: The 3 Stages of DTC Growth (and When It's Too Early)

"Start CRO when you have enough traffic" is true and useless, because nobody says enough for what. Every DTC brand grows through three layers, and CRO means something different in each one. Here's where each layer starts and how small a win your traffic can actually see before you spend a dollar on testing.

Picture the Monday call.
‍
A store doing about $130,000 a month. Around 60,000 sessions, a 2% conversion rate. Six weeks ago the team launched a product page test, a new headline and reviews pulled above the fold. This morning the variant is ahead by 6%, in green. Next to the green sits a small grey badge. Not significant.
‍
Somebody says what the room is thinking. "So CRO doesn't work for us." The test gets switched off, and by Friday the budget for the next one has moved back to Meta.
‍
If I'd been on that call, I'd have said one thing. The test didn't fail. It went blind.
‍
Even if every session had entered the test, six weeks of that store's traffic is roughly 83,000 visitors split between two versions, enough to confirm a lift of about 14% or more. Confirming 6% needed around 430,000 sessions, more than seven months of traffic. The 6% may well have been real. The test was never going to be able to say.
‍
So if you're sure you don't have enough traffic for CRO, you're probably right about the testing part. A store that can't A/B test yet still has plenty of CRO to do, though, and which kind depends on the stage it's in.

In this guide

When does CRO make sense? Why the published answers disagree
How small a win can your traffic see?
Revenue tells you the stage, traffic tells you what you can test
Stage 1: product market fit, under $100K a month
Stage 2: growth, $100K to $500K a month
Stage 3: scale, $500K a month and up
Signs you've misread your stage

When Does CRO Make Sense? Why the Published Answers Disagree

Published thresholds for starting CRO sit between roughly 1,000 and 3,000 orders a month. They disagree because each one assumes a different size of win without saying so.
‍
• CXL's rule of thumb, first published in 2015 and updated in 2022, puts the line at roughly 1,000 transactions a month and says that below it, your effort is better spent elsewhere.
‍
• Blend Commerce recommends at least 50,000 monthly sessions, as its own guideline rather than a hard rule.
‍
• Splitbase sets the bar at 3,000 orders a month, below which it says you probably won't reach significance.
‍
At a 2% conversion rate, 50,000 sessions is 1,000 orders, so CXL and Blend are drawing the same line in different units. Splitbase draws it three times higher.
‍
None of them says where the number comes from. Here's a reading that fits the arithmetic: at 1,000 orders a month, four weeks of traffic can confirm lifts of about 18% and up, and at 3,000 orders, about 10%. Both lines are defensible. They assume you're hunting for different sizes of win.
‍
So how big is a typical win? The best public figure I know of comes from Analytics-Toolkit, which looked at 1,001 A/B tests run on its platform since late 2021. Among tests that produced a statistically significant win, the median lift was 7.5%, and about a third of all tests ended as significant wins. The analysis was run by Georgi Georgiev, the statistician who founded Analytics-Toolkit, wrote Statistical Methods in Online A/B Testing and teaches A/B testing statistics at CXL. He notes that the set leans toward experienced practitioners. It describes skilled testing programs. How well it maps onto the average store, the data doesn't say.
‍
7.5% is small. Most early stage stores can't see it at all.

How Small a Win Can Your Traffic See?

At a 2% conversion rate, a four-week test at 50,000 monthly sessions can confirm a lift of about 18%. A 7.5% winner would take the same store around six months to read.
‍
That 18% is the mesh of your net. Any win smaller than the holes swims straight through, and at 50,000 sessions a 7.5% win doesn't even touch the sides. The net comes up empty, and the crew decides the lake is dead.

Table: smallest lift a 4-week A/B test can confirm by monthly sessions, from 20K to 500K

Evan Miller's sample size calculator will rerun the table on your numbers.

Chart: smallest lift a 4-week A/B test can confirm by monthly sessions, vs the 7.5% median win

Where This Table Falls Short

Sequential methods, which about 90% of the Analytics-Toolkit tests used, can stop early when an effect is large. They don't make a small effect visible on small traffic. Testing on revenue per visitor needs even more traffic than conversion rate, because order values swing far more than a yes or no purchase.
‍
The table also assumes every session enters the test. A test that runs on product pages sees just the traffic that reaches them, so look up that number instead of your total.
‍
So what stage is a store with 60,000 sessions in? Depends what it sells.

Revenue Tells You the Stage. Traffic Tells You What You Can Test.

At a $100 average order and a 2% conversion rate, $100,000 a month is 1,000 orders and 50,000 sessions, right on the CXL and Blend line. $500,000 a month is 250,000 sessions, where a 7.5% winner reads in about five weeks.
‍
That's why the three stages sit where they do. It's our model, and we'd rather say so than dress it up as an industry standard. The boundaries are where published testing thresholds and the size of a typical win line up at a $100 order.
‍
Your order value moves them:

Table: monthly sessions at $100K and $500K revenue by average order value, with CRO stage lines

Stage 1: Product Market Fit, Under $100K a Month

CRO for early stage brands means fixing what's broken and listening to customers. Under about 50,000 monthly sessions, a four-week test can only confirm lifts of roughly 20% and up, several times the size of a typical win.
‍
If the number you wrote down is above 20%, you're here, whatever your revenue says.
‍
• Fix what's broken first. Checkout errors, slow mobile pages, variants that don't load, a discount field that eats orders.
‍
• Watch real people use the site. An hour of session recordings usually shows more than a month of dashboards.
‍
• Talk to buyers and to people who almost bought. Five calls is often enough to hear the same objection twice.
‍
• Question the offer before the page. If people don't want the product at this price, no layout will fix it.
‍
CXL's list for low traffic sites looks much the same.
‍
What's a waste here is anything built around testing. At this stage you don't need a CRO agency, and anyone selling you a testing program is selling tests your traffic can't read. Do you need a CRO agency yet? covers what to do instead, and the seven questions to ask before hiring one apply whenever you do.
‍
It's also the reason Weblics doesn't run testing programmes on stores below 50,000 monthly sessions. Below that line, a program can't see enough to earn its fee.

Reminder:

Too early for testing is the cheapest stage of CRO. Five customer calls and an honest list of broken things will usually teach you more than a quarter of inconclusive tests.

You're on your way out at roughly 1,000 orders a month or, more precisely, when your mesh drops below 20%.

Stage 2: Growth, $100K to $500K a Month

Between $100,000 and $500,000 a month you can A/B test, but only changes big enough to move conversion by double digits read inside a month. Across this band the mesh tightens from about 18% to about 8%.
‍
The store from the Monday call sat here, at the bottom of the band, running a test its traffic couldn't hold.
‍
What reads at this stage is big swings. A restructured product page or a new promise on the first screen can plausibly move conversion by 15% or more. A new button color can't. Template changes count double, because one change to a template serving 200 products gets the traffic of all 200.
‍
Everything else, ship on judgment. Blend Commerce's shorthand is test risk, fix certainty: if you're confident a change is right and the downside is small, ship it, track before and after, and save your traffic for the calls you can't make yourself.

Pro Tip:

Pick tests your mesh can hold. Before you launch, look at the number you wrote down and ask whether this change could realistically beat it. If the honest answer is no, ship it as a fix or park it.

Waiting has a price here. Say a store at $300,000 a month has a change worth 5% sitting somewhere on its product page. That's about $15,000 a month, around $180,000 a year. A quarter of tests that can't read costs roughly $45,000 of it, before counting the next change you never got to.
‍
If you want a starting list, 9 PDP Patterns We Test Before Spending Another Dollar on a Redesign was built for this stage: changes big enough to read and cheap enough to ship without a redesign.
‍
You're leaving growth when a 7.5% winner reads in about five weeks, around 250,000 monthly sessions at a 2% conversion rate. Past that line, the risk flips.

Stage 3: Scale, $500K a Month and Up

Above $500,000 a month, a 7.5% winning test reads in five weeks or less, so CRO can run as a continuous program. At 500,000 sessions it takes about two and a half weeks.
‍
At 500,000 sessions, even a template that sees half the traffic can finish a readable test roughly every five weeks. If about a third end as clear wins, as in the Analytics-Toolkit set, that's three or four confirmed wins a year from that one template. They stack.
‍
This is where a line I keep coming back to holds hardest: conversion rate is the only growth input whose cost doesn't scale with the growth it produces. Ad spend has to climb for revenue to climb. A conversion lift, once found, keeps paying on every extra visitor at no extra cost. At $500,000 a month, 5% is about $25,000 a month, around $300,000 a year. The full argument is in The DTC Scaling Trap.
‍
CRO here looks like a roadmap ranked by expected impact and effort, with a guardrail metric on every test. We compared the main scoring frameworks in ICE, PIE and RICE. The cadence matters as much as the ranking, and it shouldn't stop when one test loses.
‍
The risk runs the other way now. Being early costs you a few wasted tests. Being late costs you every month the mesh was fine and nobody was fishing.

Signs You've Misread Your Stage

The most expensive mistake isn't starting too early or too late. It's running the kind of CRO that belongs to a different stage.
‍
• Your tests keep ending inconclusive. Check your mesh before you blame the idea.
‍
• Someone reports significant wins on a store under 50,000 sessions a month. It can happen with a large effect. It happens far more often when a test is stopped the first day the number looks good.
‍
• Your revenue says growth and your sessions say stage one. High AOV stores land here first, and the sessions decide what you can test.
‍
• You're past $500,000 a month and still shipping on opinion. That's where a month without a program costs the most.
‍
Back to the Monday call. The team had three ways forward that morning. Switch the test off and decide CRO doesn't work, which is what they did. Keep running small tests for another seven months and hope one reads. Or check their mesh and pick the work it can see: fixes and research above 20%, big swings between 8% and 20%, a real program below 8%. Only the third gets an answer in weeks.
‍
Your number is already written down.
‍
The question was never whether you're big enough for CRO. It's whether your test can see what you're looking for.