The ABWatcher blog

CRO Fundamentals: The 5 Elements Every Program Needs

A no-fluff guide to the 5 pillars of a real CRO program — user research, funnel mapping, baselines, testing discipline, and iteration — with live A/B test examples.

Diego Hernandez

Growth Marketer · · Updated

TL;DR

  • A real CRO program needs 5 pillars: user research, funnel mapping, baseline metrics, testing discipline, iteration.
  • Deel was caught running a hero image A/B test (photo vs. product UI illustration) detected 30 Jun 2026.
  • Notion Workspace ran a pricing table experiment with asterisked plan names and extra feature rows, detected 5 Aug 2026.
  • Mailmodo tested a fully different page layout by browser, detected 30 Jun 2026 — shorter Safari version vs. longer Chrome/Firefox/Edge version.
  • Quince tested nav copy ("Everyday Steals" vs. "Everyday Picks"), detected 29 May 2026, confirmed across two browser engines.
  • Without a hypothesis and a measurable success metric, you're redesigning — not optimizing.

Why do most CRO programs fail before they start?

Most fail because teams run random tests with no research behind them and call it "optimization."

You've seen this. Someone in a meeting says "let's just test a green button" and six weeks later nobody remembers why. That's not CRO. That's decoration with extra steps.

Real CRO is a system. Five parts, in order. Skip one and the whole thing wobbles. Here's the breakdown, with live examples of brands actually running structured tests right now — not hypotheticals.

What is user research and why does it come first?

User research means watching what real users actually do before you guess what to change.

Session recordings, heatmaps, support tickets, sales call notes — all of it tells you where the friction lives. Skip this step and you're optimizing based on vibes.

Here's the thing about Quince's nav test, detected 29 May 2026: Safari visitors saw "Everyday Steals" as the first nav item, while Chrome and Firefox visitors saw "Everyday Picks." That's not an accident of rendering — it showed up consistently across two separate browser engines, which is the signature of a real controlled experiment.

Why does copy like this even get tested? Because somebody looked at behavioral data and found that "Steals" might trigger a different psychological trigger than "Picks" — urgency and bargain-hunting versus curation and taste. You don't get to that hypothesis without watching real user behavior first.

Takeaway: before you write a single test hypothesis, spend a week in your session recordings. The friction is usually obvious once you look.

How do you map the funnel before you test anything?

Funnel mapping means laying out every step a user takes from first visit to converted customer, and marking exactly where they drop off.

You can't fix what you can't see. Most teams jump straight to "let's test the CTA button" without knowing if the CTA is even the leak. Pull your funnel data first: landing page → product page → cart/signup → checkout/activation. Find the biggest percentage drop. That's where you test.

Mailmodo's layout test, detected 30 Jun 2026, is a great example of funnel-stage thinking taken seriously. Visit 1 on Safari showed a short page — hero, one feature section, basic footer. Visits 2 through 5 on Chrome, Firefox, and Edge showed a much longer page with testimonials, additional feature breakdowns, and more proof sections stacked in.

That's a top-of-funnel test asking a blunt question: does more content build enough trust to move someone forward, or does it just add friction and scroll fatigue? You only run a test that dramatic if you've already mapped where visitors bail — right after the hero, probably — and you're testing two entirely different theories about why.

Takeaway: map your funnel in a spreadsheet this week. Rank stages by drop-off percentage, not by what's easiest to test.

What baseline metrics do you actually need before testing?

You need a stable conversion rate benchmark, per funnel stage, over at least 2-4 weeks, before you run a single experiment.

No baseline, no way to know if a result is real or noise. This sounds obvious. Teams skip it constantly because baselines are boring and tests feel like progress.

Minimum baseline stack:

Metric Why it matters Typical minimum window
Overall site conversion rate Your north star, sanity-checks every test 2-4 weeks
Funnel-stage conversion (page-to-page) Shows where the real leak is 2-4 weeks
Average order value / deal size Catches tests that win on rate but lose on revenue 4 weeks
Traffic volume by segment Determines if you even have enough sample for significance Ongoing
Existing test velocity Tells you if "testing discipline" below is real or aspirational Rolling 90 days

Without these numbers, a 3% lift on a pricing page test is meaningless — you don't know if that's signal or normal weekly variance.

Takeaway: don't launch your first test until you can say your baseline conversion rate with a straight face and a confidence interval.

What does real testing discipline look like day to day?

Testing discipline means every test has a written hypothesis, a defined success metric, and a pre-set sample size before it ships — no exceptions.

Notion Workspace's pricing page test, detected 5 Aug 2026, shows what a disciplined, high-stakes test looks like. One variant added asterisks next to "Plus" and "Enterprise" plan names, a footnote reading "Plans may not be available for every workspace," extra rows for Premium support and a Customer success manager, and solid-blue CTA buttons — while six other sessions saw the standard, unmarked pricing table.

That's not a cosmetic tweak. That's a hypothesis about trust and transparency on the highest-stakes page on the site. You don't run a pricing page experiment on a whim — one bad variant there costs real revenue in days, not weeks. This is the kind of test that requires a pre-registered success metric (probably revenue per visitor, not just click rate) and a hard stop-loss if it tanks.

Compare that to Deel's hero test, detected 30 Jun 2026. Control showed a photo of a woman working on a laptop. The variant showed a product UI dashboard illustration with dotted background elements — and the divergence was confirmed by a cookie (frontend.exp22-people-design) toggling true/false plus two distinct Optimizely/VWO experiment IDs running in parallel. That's textbook infrastructure: named experiment, tracked cookie, platform-level A/B split. Nothing ambiguous about what's being tested or why.

As Mouseflow's CRO guide puts it bluntly: without a defined hypothesis, a measurable success metric, and a structured experiment, "you're redesigning, not optimizing." Deel and Notion both show what the structured version looks like in production.

Takeaway: before your next test ships, write the hypothesis, the metric, and the kill criteria on one slide. If you can't, you're not ready to launch.

How do you turn one test into an iterative program?

Iteration means every test — win, loss, or flat — feeds the next hypothesis, instead of being filed away and forgotten.

This is the pillar teams skip most. They run a test, get a result, ship the winner (or don't), and move on. No documentation. No pattern recognition. Six months later they're testing the same button color again because nobody remembers the first time.

Build a simple test log: hypothesis, result, learning, next question. That's it. The Contentsquare CRO guide makes a point worth internalizing here — A/B testing is important, but it's not the whole program. Plenty of conversion wins come from fixing broken elements, confusing navigation, and form friction that never needed a split test in the first place. Iteration means folding those fixes into the same loop as your formal experiments.

Look at the four detections above as a single continuous program instead of four unrelated tests:

  • Deel is iterating on emotional vs. functional hero imagery.
  • Notion is iterating on pricing transparency and plan naming.
  • Mailmodo is iterating on page length and proof density.
  • Quince is iterating on nav microcopy tone.

None of these are one-off tests. They're threads. The next test at each of those companies almost certainly builds on this one.

Takeaway: this sprint, set up a shared test log doc. Every test — yours or a competitor's you're tracking — gets one row. In three months you'll have a pattern library, not a pile of disconnected experiments.

Common questions

What's the difference between CRO and just running A/B tests?

CRO is the full system — research, funnel mapping, baselines, and iteration — while A/B testing is just one tool inside it. You can run A/B tests forever without a CRO program if you're not grounding them in research or measuring against a real baseline first.

How long should a CRO program run before you see results?

Most teams need 60-90 days to get through one full baseline-test-iterate cycle with statistically meaningful results. Rushing this — testing on 3 weeks of noisy data or with too little traffic — produces false wins that don't hold up.

Do small companies need all 5 pillars, or can they skip some?

Small teams can run lightweight versions of each pillar, but skipping one entirely — usually baseline metrics or documentation — is the most common reason early CRO efforts collapse. Even a scrappy spreadsheet baseline beats no baseline at all.

Why do companies like Deel or Notion run tests on high-stakes pages like pricing and hero sections?

Because those pages carry the highest traffic and highest revenue impact per visitor, small percentage gains compound fast. A well-run test on a pricing page, like the one detected on Notion Workspace on 5 Aug 2026, can move revenue more than a dozen smaller tests elsewhere.

What tools do I need to start a CRO program from scratch?

You need session recording/heatmap software, an A/B testing platform, and a way to track baseline analytics — no single tool covers everything. As Conversion's CRO Primer notes, most mature programs stitch together multiple tools rather than relying on one platform for the full picture.

Your next move

Don't try to build all 5 pillars this week. Pick the one you're weakest on — probably baselines or documentation, be honest — and fix just that before you launch another test. A CRO program built in order beats five tests fired off in a random sprint, every time.

Read next

More from the ABWatcher blog

From the glossary

Vocabulary the article uses. See all terms →

  • Shrinkage Estimator— A statistical adjustment that pulls noisy per-variant or per-segment estimates toward an overall average to reduce overfitting to randomness.
  • Flicker Test— A diagnostic check to confirm whether users briefly see the original page before a variant renders in client-side experiments.
  • Pre-Experiment Bucketing Audit— A review process confirming users are being randomly and evenly assigned to experiment variants before launch.
  • Above-Fold Exclusion Rule— A test-design rule that deliberately prevents changes to above-the-fold elements so an experiment isolates below-the-fold effects.
  • Flicker-Free Testing— A technical setup that prevents users from briefly seeing the original page before a test variant loads.
  • Orthogonal Test Design— A test structure where multiple variables are changed independently so their individual effects can be measured without interference.

See more like this

ABWatcher catches A/B tests like this across 1,000+ brands.

Watch live experiments at 1,000+ high-converting brands, complete with hypothesis and takeaway. 7 days of Pro, free, no card, then Free forever for ten watched companies.