New ResearchCorpBenchSee the results ↗We tested AI models on real marketing tasks.
Tryozi
TelehealthMarketing

Self-optimizing landing page agents

Tryozi learns which landing-page experiences work for each traffic segment.

More visitors start intake as agents segment traffic and personalize landing pages within approved boundaries. Each experiment builds a knowledge base of what works and what does not, guiding the next variant and traffic allocation.

Summarize with AI

The impact

23%

higher conversion to intake starts

Self-optimizing landing page agents

Case study facts

Team
Marketing
Systems connected
4
Time to live
3 weeks
Engagement
AI implementation

The impact

Results for the team

higher conversion to intake starts
23%
higher conversion to intake starts
More visitors start intake from landing pages that use approved messaging matched to their campaign context.
more page variants tested
5×
more page variants tested
The marketing team tests more page variants within approved claim and offer limits.
less time preparing variants
72%
less time preparing variants
Approved page components and messages reduce the work needed to assemble and check each new variant.

Situation

Different intentions. The same page.

One page has to serve visitors arriving with different expectations. Manually creating campaign variants makes learning slow, while unconstrained optimization risks changing claims and offers the business has not approved.


Challenge

More variants needed clearer boundaries.

Personalization had to reflect arrival context while preserving approved claims and offers. The team also needed dependable experiment assignment and measurement before treating a change as an improvement.

What we built

Personalized pages that learn from every test.

The optimization agent segments traffic using campaign and consented session context, then assembles personalized pages from approved content components. Experiments compare variants against a defined conversion event and guardrail measures.

Each test adds to a knowledge base of which messages and layouts work for each segment, including what underperforms. The agent uses that learning to choose the next variant and adapt traffic allocation autonomously within approved experiment rules. Material claim or offer changes still return to human review.

AI tech stack

  • Cloudflare Agent CloudAgent infrastructure
  • OpenRouterModel access
  1. Segment arrival traffic

    Group visitors using campaign and consented session signals to personalize their experience, never guessing sensitive traits.

  2. Select approved content

    Use approved messages and layouts alongside learned segment preferences, staying within agreed offer boundaries.

  3. Assemble the page

    The agent composes a coherent variant matched to the visitor's intent.

  4. Validate before launch

    Claims, event tracking and usability are checked before any exposure.

  5. Run the experiment

    Variants compete against a control within agreed guardrails.

  6. Learn and adapt

    Record what works and underperforms for each segment, then use that knowledge to choose the next variant and adapt traffic allocation within approved rules.

Evals and guardrails

Test generated pages. Keep changes within bounds.

  • Page-generation evals

    Check generated pages against approved claims and layout rules. Repeat tests with missing or conflicting campaign information.

  • Experiment evals

    Test that visitors stay in the same test group and actions are recorded correctly. Check groups stay separate before changing traffic.

  • Guardrails

    Use only approved page elements and claims. Block guesses about sensitive personal traits and require review of new claims or offers.

Results

Page learning carries into the next test.

More visitors start intake from landing pages that use approved messaging matched to their campaign context. The marketing team tests more page variants within approved claim and offer limits.

Approved page components and messages reduce the work needed to assemble and check each new variant.

higher conversion to intake starts
23%higher conversion to intake starts
more page variants tested
5×more page variants tested
less time preparing variants
72%less time preparing variants

From Tryozi

“We had tried other partners and nobody could unblock it. Revensi came in, understood the problem quickly, and turned a bottleneck into something that just runs. Total game changer.”
Yazan SalehCEO at Tryozi

Build capacity to grow. Own your intelligence.

Bring the workflow slowing your pipeline, campaigns or renewals. We'll map the bottleneck and show you the AI agent concept.