PERFORMANCE MARKETING · PRACTICAL PROJECT
Make the next click count.
Learn how to compare two landing pages, measure meaningful conversions and turn an experiment into a portfolio case study.
19 November 2026 · ZapScholar
Imagine you are running ads for a Hyderabad meal subscription startup. People click your ad, but hardly anyone requests a trial. Your first instinct might be to redesign the entire page. Resist that temptation. If you change everything at once, you may never learn what actually helped.
This guide walks through a small experiment you can plan as a beginner, even before you have permission to spend on ads.
Version A is your control, usually the existing page. Version B is a variation. For example, A might ask visitors to book a consultation while B offers a trial menu preview. To isolate the effect of the offer, keep the traffic source, layout and other conditions consistent wherever possible. Assign visitors to a version at random and keep their assignment consistent on repeat visits.
A before-and-after comparison is not the same as a randomized test. A weekend promotion, a different audience or a change in ad spend can explain a difference between two weeks. Google Ads offers custom campaign experiments for supported campaign types; check the available options before deciding how to split traffic.
Start with the biggest plausible obstacle to a conversion. Read the ad, open the page on a phone and try completing the form yourself. A broken form is a bug to fix, not a hypothesis to test.
| Element | Example change | Why test it? |
|---|---|---|
| Headline | Generic healthy meals vs weekday meal delivery | Clarify the offer |
| Form | Six fields vs three essential fields | Reduce effort |
| Proof | Add genuine, permissioned customer reviews | Address trust |
| CTA | Contact us vs check delivery availability | Clarify the next step |
Do not add invented reviews, false scarcity or hidden fees. A test should improve the customer experience, not trick visitors into submitting a form.
For a subscription business, a valid trial request may be more useful than a button click. Define a qualified request before the test: for instance, a genuine contact from a serviceable delivery area. Record how you will deduplicate requests and handle spam.
Use this structure: because we observed X, changing Y should improve Z. Example: because the current form asks for details we do not need until onboarding, removing three optional fields may increase valid trial requests per visitor.
Duplicate the page, change the chosen element, and test both on mobile and desktop. Check the confirmation message, keyboard navigation, form labels, loading behavior and successful submission. Keep privacy notices and consent requirements consistent. Obtain approval before editing a live business website.
Record a version identifier and a successful submission event, rather than counting every click on the submit button. Use the same definition of a conversion in both versions. Test the event in your analytics setup and reconcile a sample against the actual enquiries received. Never put names, phone numbers or email addresses in tracking URLs.
For external campaign links, follow Google Analytics campaign URL guidance and use consistent source, medium and campaign parameters. UTMs identify traffic sources; they do not randomly assign visitors to variants. Preserve platform auto-tagging where appropriate, and use a separate experiment identifier for the page version.
Estimate the existing conversion rate, the smallest improvement worth acting on and the sample size needed for your chosen statistical method. Use a sample-size calculator for two proportions if your metric is a binary visitor conversion. Plan for normal weekday and weekend variation. Avoid stopping just because one version looks better on day two. If the traffic is too low, treat the work as a pilot rather than claiming a proven result.
Compare conversion rates, lead quality and cost. Check whether allocation was approximately as intended, tracking worked and there were no major outages or campaign changes. Use an appropriate confidence interval or test of proportions and document uncertainty. A result can be inconclusive; that is a legitimate outcome.
The numbers below are fictional and intended only to demonstrate the calculations. They are not ZapScholar student results or industry benchmarks.
| Metric | A: original | B: shorter form |
|---|---|---|
| Unique assigned visitors | 500 | 500 |
| Visitors submitting a valid enquiry | 20 | 30 |
| Conversion rate | 4% | 6% |
| Qualified leads | 12 | 12 |
| Illustrative allocated ad cost | ₹3,000 | ₹3,000 |
| Cost per qualified lead | ₹250 | ₹250 |
Conversion rate: converting visitors ÷ assigned visitors × 100.
Relative lift: (6% − 4%) ÷ 4% × 100 = 50%.
Absolute lift: 2 percentage points.
Cost per qualified lead: allocated spend ÷ qualified leads.
Version B generates more enquiries, but both versions deliver the same number of qualified leads at the same cost. Moreover, the observed conversion difference is not sufficient to establish a winner at a conventional two-sided 5% significance level using a simple two-proportion z-test (approximate p-value 0.15, assuming independent observations). The sensible next step is to investigate lead quality and collect more evidence according to the preplanned design, not announce a 50% business improvement.
A useful portfolio case study explains your decisions rather than displaying a screenshot of an ads dashboard. Create a short document containing the business brief, customer problem, hypothesis, annotated screenshots of A and B, measurement plan, anonymized results and your recommendation. Include what you personally did and what you would change next.
If you cannot run a live campaign, create two mock pages and test the submission workflow with dummy data. Label the project as a simulation. Do not invent traffic or sales. You can still demonstrate clear thinking, page design and measurement skills. Explore the performance marketing learning roadmap for related skills such as creative testing and budget decisions.
Business goal: ____
Observed problem: ____
Hypothesis: ____
Control and variation: ____
Primary metric and quality guardrail: ____
Assignment method and planned sample: ____
Tracking checks: ____
Results and uncertainty: ____
Decision and next experiment: ____
Yes. You can randomly split existing organic or referral traffic using a suitable experiment implementation. With no real traffic, you can practice setup and usability checks, but cannot measure a real conversion lift.
There is no universal duration. It depends on traffic, baseline conversion rate, the effect you want to detect and your planned analysis. Include representative business cycles and follow the stopping rule you chose before launch.
Yes, but comparing two pages with several changes measures the combined effect, not the individual contribution of each element. A single meaningful change is easier for a beginner to interpret.
GA4 can help measure behavior and conversions, but you still need a reliable method to assign visitors to versions and analyze the experiment. Analytics reporting by itself is not randomization.
Check tracking and experiment integrity first. If the result is credible, retain the control and record what you learned. A well-documented unsuccessful experiment can be valuable evidence of your judgment.
ZapScholar’s 18-week Practical Digital Marketing Course includes landing pages, paid campaigns, analytics and mentor-reviewed startup projects. Learners in Hyderabad can explore the live-online training option. Ask about current project availability and how campaign work is supervised.