Test systematically and apply learnings

After building segment-specific pages, continuous testing is how you improve them over time. But most companies test wrong: too many elements at once, insufficient traffic, random ideas instead of hypotheses, no documentation of learnings.

This chapter shows you how to prioritise pages for testing (highest traffic first), test one element at a time (isolate what works), set proper test parameters (sufficient duration and volume), and apply learnings systematically (don't re-test the same hypothesis everywhere).

Test highest-traffic pages first: Your compliance-driven page gets 1,500 visitors/month. Your breach-reactive page gets 300 visitors/month. Test compliance page first. Why? Results come faster (reach statistical significance in 2 weeks versus 8 weeks), and even small improvements have big impact (5% lift on 1,500 visitors = 75 additional leads/month versus 15 leads/month on low-traffic page).

Don't test pages with under 500 visitors/month. Takes too long to reach significance. Optimise those pages by applying learnings from high-traffic pages, but don't run dedicated tests on them.

Test pages close to targets first: If your goal is 4% conversion and one page converts at 3% whilst another converts at 1%, test the 3% page. Small improvements get you to target. The 1% page has fundamental issues (wrong traffic, wrong message, wrong segment), needs rebuilding not testing.

Test winning pages, not losing pages: Conventional wisdom says "fix what's broken". Wrong. Optimise what's working. If a page converts at 5% (above benchmark), test to push it to 6-7%. If a page converts at 1% (far below benchmark), rebuilding is better than testing. You can't test your way from 1% to 5%, you need structural changes.

Elements to test in priority order: (1) Headline (affects 50% of conversion, everyone reads it), (2) Proof type and placement (affects 30% of conversion, key to credibility), (3) CTA wording and placement (affects 15% of conversion, determines action taken), (4) Page length and structure (affects 5% of conversion, more complex to test). Start with headline tests, move down the list.

Only change one thing per test. If you change headline + proof + CTA simultaneously, you won't know which change drove results. Isolate variables.

Headline tests: Keep everything else identical (same proof, same CTA, same page structure). Only test headline variations. Example: control "Security training that reduces breach risk" versus variant "Reduce breach risk 47% with security training" (adds specificity). Run until statistical significance. If variant wins, make it the new control and test another variant.

Test 2-4 headline variations: outcome-focused versus pain-focused, specific versus generic, short versus long, question versus statement. Don't test 10 variations simultaneously (splits traffic too thin).

Proof tests: After headline is optimised, test proof elements. Control "230 financial services firms use our platform" versus variant "Used by compliance teams at Lloyds, HSBC, Barclays" (named customers versus count). Or test proof placement: above-the-fold versus mid-page. Or test proof type: testimonial quote versus stat versus case study.

CTA tests: After headline and proof are optimised, test CTA. Wording: "Book demo" versus "See platform demo" versus "Schedule your demo". Placement: above-the-fold versus below proof. Design: button colour, button size, button text.

Set proper controls: 50/50 traffic split between control and variant (not 80/20, you need sufficient traffic on both), random assignment (not time-based like "control Monday-Tuesday, variant Wednesday-Thursday"), consistent experience (if someone sees control once, they see control every time, no mixing).

Minimum test duration: 2 weeks (accounts for day-of-week variation, most B2B traffic has weekly patterns). Minimum conversions: 100 per variant (need sufficient sample size for statistical significance). If your page gets 1,000 visitors/month at 4% conversion, that's 40 conversions/month, need 2.5 months to reach 100 conversions per variant. Adjust expectations accordingly.

Don't just track blended conversion rate. Track by segment. A headline might improve conversion for compliance-driven but hurt conversion for proactive segment. If both segments use the same page, you need to know segment-specific impact.

Use UTM parameters or campaign tracking to identify segment. LinkedIn ads for compliance-driven tag traffic as segment=compliance. Content marketing for proactive tags as segment=proactive. Now you can see: headline test improved compliance-driven conversion 15% but decreased proactive conversion 8%. Net positive, but reveals the page is trying to serve two segments with different needs.

This data informs page splitting decisions. If every test shows compliance-driven and proactive responding differently, they need separate pages. If they respond similarly, they can continue sharing.

Statistical significance threshold: Use 95% confidence minimum. Don't declare a winner at 80% confidence (too likely to be random variation). Use a significance calculator (many free online). Input: control conversion rate, variant conversion rate, sample size. Output: confidence level. Wait for 95%+ before declaring winner.

Minimum detectable effect: Decide upfront: what's the smallest improvement worth caring about? If control converts at 4%, is 4.1% worth implementing (2.5% lift)? Probably not (too small to matter). Is 4.4% worth implementing (10% lift)? Yes (meaningful improvement). Set your threshold (typically 5-10% lift minimum) and don't bother with smaller wins.

Account for false positives: if you run 20 tests, one will show significance by pure chance (that's what 95% confidence means). Don't get excited about one winning test. Look for patterns across multiple tests. If outcome-focused headlines beat pain-focused headlines in 3 separate tests, that's a real pattern. If they win once and lose twice, it's noise.

After you've learned something on one page, apply it to similar pages without re-testing. Don't re-test the same hypothesis on every page.

Example: You test headlines on compliance-driven page. Outcome-focused headline ("Complete training in 30 minutes") beats pain-focused headline ("Stop wasting time on ineffective training") by 12%. This is a validated learning: compliance segment responds to outcome framing, not pain framing.

Now apply this learning: update your Google search ads for compliance segment to use outcome framing. Update your remarketing page headlines to use outcome framing. Update your email sequences to use outcome framing. All without re-testing. You've already proven outcome-focused messaging works for compliance-driven segment, apply that pattern everywhere.

Document learnings in a testing log: Date, page tested, element tested, hypothesis, control, variant, results, confidence level, segment-specific notes, next actions. This log becomes your institutional knowledge. When you launch a new campaign for compliance-driven segment, consult the log: we already know outcome headlines beat pain headlines, specific outcomes beat generic outcomes, hard CTAs beat soft CTAs. Start with these patterns.

Build a pattern library: After 10-20 tests, patterns emerge. For compliance-driven segment: outcome headlines beat pain headlines, speed proof beats behaviour change proof, hard CTAs beat soft CTAs, short pages beat long pages. For proactive segment: data headlines beat outcome headlines, behaviour metrics beat speed proof, soft CTAs beat hard CTAs, long pages beat short pages. These are your documented patterns for each segment.

When you build a new page for compliance-driven segment, use the pattern library as your starting point. Don't start from scratch. Build using proven patterns, then test refinements.

Refresh tests periodically: Patterns can change. What worked last year might not work this year. Re-test winning patterns annually to confirm they still hold. If outcome headlines still beat pain headlines for compliance segment after 2 years, the pattern is solid. If they suddenly stop working, something changed (market conditions, competition, segment beliefs shifted), investigate and adapt.

Prioritise testing on highest-traffic pages (results faster, bigger impact). Test pages close to target conversion rates and test winners not losers. Test elements in order: headline first (biggest impact), then proof, then CTA, then page structure.

Test one element at a time with proper controls. 50/50 traffic split, minimum 2 weeks duration, minimum 100 conversions per variant, 95% confidence threshold. Don't change multiple elements simultaneously.

Track conversion by segment, not just blended average. A change might help one segment and hurt another. Use segment-specific data to inform page splitting decisions.

Apply learnings across similar pages without re-testing everything. Build a testing log and pattern library documenting what works for each segment. When launching new campaigns, start with proven patterns then test refinements. Re-test patterns annually to confirm they still hold.

With landing pages optimised, you're ready to implement systematic experimentation. That's the next playbook.