Services / Websites & Conversion / CRO & A/B Testing
A/B tests that lead to decisions — not just spreadsheets.
CRO only works with hypothesis discipline and statistical cleanliness. I build programs with clear learning goals, valid setups and roll-out plans — so that wins get scaled, not just documented.
Hypothesis pipeline
Tests run along learning goals, not gut feeling. Every hypothesis has an expected effect.
Statistically valid
Clear sample size, frequentist or Bayesian methodology, no peeking trap. We don't stop tests too early.
Roll-out, not just a dashboard
Winning variants go into the main version. Tests are a tool for decisions, not a research project.
What you can book
Six building blocks, individually or as a package
We usually start with a CRO audit + hypothesis backlog. Test volume only comes once the setup is in place.
CRO audit
Heatmap analysis, session recordings, funnel drops, form abandonment. Output: problems map.
Hypothesis backlog
PIE-/ICE-prioritized list, each hypothesis with an expected effect, effort and learning goal.
Tool setup
VWO, Convert, Vercel edge splits or a custom setup. Tracking integration, QA routine.
Implementation
Building test variants — code, design, copy. Small changes in the tool, larger ones as a branch in the repo.
Statistics & analysis
Sample plan, significance check, subgroup analysis, side-effect documentation. Clear win/loss decision.
Roll-out + learning log
The winner goes live, the loser into the learning log. Monthly sync with backlog updates.
How do we differ?
Three ways to buy CRO
CRO market leaders deliver volume at day rates. Tool self-service is fast but thin. We couple strategy and implementation.
| CRO market leaders konversionsKRAFT, FELD M | Tool self-service VWO, Convert + in-house team | Truong Suarez Strategy + building hand | |
|---|---|---|---|
| Model | Retainer €8–20k+/month | Tool license + in-house hours | Fixed-price audit + modular test packages |
| Hypothesis source | Heuristic frameworks, best-practice library | In-house gut feeling + tool recommendations | Data-driven from heatmap + funnel analysis |
| Implementation | In-house dev team | Tool editor (limited) | Tool editor + code branch depending on complexity |
| Statistical discipline | Very strong, often Bayesian | Tool default values, peeking risk | A clear sample plan before every test |
| Best phase for you | Enterprise with a test program > 50/year | First test program with an own dev | 10–30 tests/year, growth with a system |
Comparison based on publicly available information, as of 2026. If your situation would be better served elsewhere, I'll tell you so in the intro call.
How we work
Five phases, one point of contact
Discipline in the setup, speed in the iteration. Every phase has an output you can keep.
CRO audit
Funnel analysis, heatmaps, session recordings, form drops. Output: problems map.
Hypotheses
PIE/ICE backlog, expected effects, test prioritization. Output: test plan.
Setup
Tool choice, tracking integration, QA routine. Output: test platform live.
Test loop
2–4 tests in parallel, clearly documented. Output: monthly learning log.
Review
Program review, methodology update, backlog refresh. Securing wins long-term.
Stack
What we work with
No black-box tools. Everything we use, you can run yourself — if you want to.
Test tools
- VWO
- Convert
- Vercel Edge Config
- Optimizely (Enterprise)
Analysis / Heatmap
- Hotjar
- Microsoft Clarity
- FullStory
- GA4 + BigQuery
Statistics
- Evan Miller A/B Sample Calc
- Bayesian A/B tooling
- BigQuery (subgroups)
Docs / Ops
- Notion (test database)
- Linear (tickets)
- Loom (sync)
- Looker Studio (dashboard)
Recommended entry point
Two paths, depending on where you stand
For running CRO programs, the tracking audit reveals setup weaknesses. For building a program, the Growth Sprint is the better choice.
For you if
You're already testing
You want to know whether tracking is clean, tests run in a statistically valid way, and how the program can be structured.
5–7 days · report + 30-min call
Starter Audit / Tracking
- – Tool setup check (VWO/Convert/Edge)
- – Statistics-hygiene assessment
- – Hypothesis-backlog review
- – Top 3 program recommendations
For you if
You're building the program from scratch
You're starting CRO as a systematic program and want to set up the hypothesis backlog, tool setup and statistical discipline correctly from the start.
2 weeks · report + workshop
Growth Sprint
- – Funnel & heatmap analysis
- – 20–30 hypothesis backlog (prioritized)
- – Tool & statistics setup
- – 3-month test roadmap
- – Half-day workshop with the team
Not sure? The symptom triage on the audits page helps you choose. The audit fee is credited toward a follow-up project.
When this becomes relevant
Typical starting points
Three recurring situations where CRO & A/B testing is the right tool.
Funnel plateau
Ads no longer scale
When it's not the ad account but the funnel after the click that's the problem — what CRO concretely does.
Tracking
Tracking despite consent loss
Why classic test tools have to be combined with server-side tracking so conversions stay valid.
Launch
Launch with landing pages
When new landing pages go live — how A/B tests find the conversion levers from the start.
FAQ
What clients often ask before the first collaboration
Do I need minimum traffic for CRO?
Rule of thumb: > 10,000 visitors/month on the test page and > 200 conversions/month. Below that, tests take very long (4–8 weeks) and significance is hard. With lower traffic, go for heuristic optimization rather than an A/B program.
VWO or Convert or a custom solution?
For marketing teams without dev resources: VWO or Convert (editor + statistics dashboard). For teams with Next.js + Vercel: Vercel Edge Config splits (much more performant, no flicker). For enterprise with BigQuery/data warehouse: Optimizely or custom splits + statistical analysis in dbt.
How long does a test have to run?
At least 2 full weeks (to capture weekday fluctuations) and until significance per the sample plan. Typically 2–6 weeks depending on traffic and expected effect. Peeking (stopping the test as soon as “p < 0.05”) leads to false wins — we stick to the plan.
What's a realistic success rate?
Industry average: 10–25% of tests are clear wins, 10–15% clear losses, the rest neutral. Whoever claims a 50% win rate either has peeking problems or cherry-picking. Learning from losses matters just as much — hence the clean learning log.
In-house or agency?
Hybrid works best: I build the system (audit + backlog + statistics routine), your team carries it forward. After 6–12 months the program is anchored internally, and I stay on for strategy reviews and tricky stats questions.
Let's talk
Three paths — depending on where you are.