From page tests to journey tests
Trends say CRO is shifting from isolated pages to full journeys. Here is the practical how, with examples, constraints, and a playbook that works without big-tech infrastructure.
Andrea CorviWhy page level optimisation stops scaling
Page tests are not wrong. They are often the fastest place to start. The problem is the assumption that local lift equals global impact. Users do not convert in a straight line. They bounce, return later, switch devices, and arrive through different channels. A single landing page is one touchpoint in a longer sequence.
Journey testing is the practical response. You still run controlled experiments, but you design them around how conversion happens in reality. That changes which unit you randomise, which metric you optimise, and how you interpret sample size and attribution.
Local lift is not enough
Improving CTR on a page can shift who reaches downstream steps. The downstream mix changes, and the business outcome can move differently.
Channels interact
Email, paid search, and in-product nudges can amplify or cancel each other. Journey experiments force you to state which touchpoints are in scope.
Devices break assumptions
Session based assignment falls apart when a user starts on mobile and completes on desktop. The journey spans identities.
A beginner friendly way to define a journey
Start with a sentence. A journey is a start event, the critical steps you might change, and an end event that matches the decision. For many teams that looks like ad click, landing page, sign-up, onboarding touchpoint, and first value event.
Then decide three things. Your unit of analysis, your exposure definition, and your analysis window. These three choices decide whether results are interpretable and whether you can scale the programme.
A beginner friendly journey definition
Start simple. Write a start event, the key steps you will change, and a single end event you want to improve. Then decide the analysis window and whether you randomise by user, account, or session.
Choose a unit that matches the decision
- User: best default for most journeys, especially cross-device and return visits.
- Account: common in B2B SaaS when multiple users share the same organisation outcome.
- Session: only safe when you care about an immediate effect and repeat exposure is unlikely.
If you are not sure, default to user-level assignment. It is easier to explain and less likely to create accidental contamination.
Journey experiment types you can run today
You do not need a complex platform to start. You need a clear scope and a consistent template. The patterns below work for intermediate practitioners and small teams, as long as you are honest about constraints.
Multi-step funnel tests
Change messaging and friction across steps, not just the entry page. Evaluate impact on the end event, and use driver metrics per step to debug where lift comes from.
- •Landing headline plus sign-up form friction
- •Pricing message plus checkout reassurance
Onboarding flows
Optimise activation, not just sign-up. This is often the highest leverage journey test for SaaS, because it changes long-term retention through early experience.
- •Guided setup versus self-serve
- •First value checklists and progressive disclosure
Email plus on-site tests
Coordinate a lifecycle touchpoint with an on-site experience. The design challenge is to keep assignment consistent across channels and avoid partial exposure.
- •Onboarding email plus in-app nudge
- •Cart abandonment email plus checkout banners
The practical constraints you must plan for
Journey testing makes three problems more visible. Attribution, sample size, and metric selection. If you do not plan for them, you will ship faster for a month and then stall because results are hard to interpret.
Sample size grows down-funnel
The deeper the outcome, the rarer it is. That usually means a longer runtime or more traffic. If you cannot afford that, use a driver metric as your primary and treat the deeper outcome as directional.
Attribution is not causality
In journey testing you will see more cross-channel paths and delayed conversions. Attribution models can help you debug, but the experiment design is what makes the result causal. Define exposure, define windows, and stick to the plan.
Metric selection needs guardrails
Journey experiments are more likely to move intermediate metrics. Use guardrails to avoid accidental harm. If you are not already doing this, start with the guardrail metrics guide.
A practical stance for small teams
- Use one primary journey metric, and keep step metrics as diagnostics
- Plan sample size for the primary, not for every step
- If you look at many metrics, treat wins more conservatively
The winner's curse applies here too. When you celebrate only significant wins, you tend to overestimate lift. See the winner's curse article.
A practical playbook for 1 to 3 person teams
Journey testing feels advanced because it forces clarity. The good news is that the mechanics are simple once you have a stable brief template and a checklist.
Before you launch
- •Write the journey in one sentence, and list steps in scope
- •Choose the unit and define exposure, including re-entry rules
- •Set the primary metric, drivers, and 2 to 4 guardrails
- •Plan sample size and duration for the primary outcome
After you launch
- •Monitor SRM, exposure, and instrumentation before reading effects
- •Interpret step metrics as diagnostics, not as success criteria
- •Document a decision with confidence intervals, not just p-values
- •Feed learnings into the backlog, and keep the cadence consistent
References
- Kohavi, R., Tang, D., & Xu, Y. (2020). Trustworthy Online Controlled Experiments. Cambridge University Press. Cambridge
- Kohavi, R., et al. (2020). Online controlled experiments at scale: Lessons and extensions to medicine. Trials. Trials
- Deng, A., et al. (2018). Pitfalls of Long-Term Online Controlled Experiments. Microsoft Research. PDF
- Tang, D., et al. (2010). Overlapping Experiment Infrastructure: More, Better, Faster Experimentation. Google. PDF
Share this with your team
Help others design experiments that match how users actually convert.
Related Resources
Plan experiments with proper power analysis.
Protect experiments from hidden harm.
Structured checklist for trustworthy experiments.
Why significant A/B test wins overestimate impact.
Ship more tests with templates and guardrails.
Frequently Asked Questions
Make the next journey test easier
Use a standard template, define guardrails, and plan sample size for the metric you will actually ship on.