🚀 My impact

By experimenting and optimising the PDP entry experiences starting in April 2025, the LPO initiative delivered a measurable uplift across key business KPIs by January 2026. Starting from the baseline of 47.1% BR and 1.7% CvR.

OP25 delivered:
• BR: -5%
• CVR: +4%
• GMV: 9M EUR

For 2026, the initiative expanded with a more comprehensive strategy that introduces personalisation models and deeper intent-aligned experiences.

Projected to deliver:
• BR: -8%
• CVR: +4%
• GMV: 45M EUR

Background

Zalando data revealed a misalignment at PDP and Catalog entry points: users arriving from Performance Marketing (Meta / SEA) and SEO often bounced immediately, with PDP bounce rates around 50% and Meta traffic exceeding 56%. Around half of PDP entry sessions came from Performance Marketing, with most users logged out and landing on pages that did not reflect their search intent.

The Challenge

Align PDP landings with logged-out paid traffic, removing friction to reduce bounce and increase CVR and GMV.

Role

AI-driven Lead Product Designer.

I led the bundle strategy and CRO experimentation at Zalando, applying a data-driven product design approach to reduce friction and improve conversion at scale for paid traffic landing on PDP (SEA and Meta) on mobile web, while laying the foundation for personalised design architecture in 2026.

• AI-Powered UX Designer

• Systems Thinking Designer

Deliverables

These are all the deliverables:

• Consolidated experiment Roadmap LPO initiatives 2025 and 2026 (influenced its definition)

• UX strategy initiatives 2026 (influenced its definition)

• Strategic Design Bundles for the PDP

• EDD (Experiment Design Document) for each bundle

• Defining KPIs per bundle and setting up control vs. variant experiments

• Analysis results and learnings
 
• Creation of the holdout group approach (influenced its definition)

• A11y documentation: Dark mode, Landscape view, 200 % Dynamic text, Screenreader annotations.

• Bounce behaviour - analytic document (influenced its definition)

• Workshops with Meta and internal workshops on bundle prioritisation and ideation

• UAT and QA documentation evaluating functional performance, design, error-handling logic, and A11y compliance

• Creation of the holdout group approach (influenced its definition)

• Rollout Guidelines for the LPO (influenced its definition)

• Handoff Documentation to engineers

Tools

Figma, Figma AI, Figma Make, Miro AI, Chat GPT 5.1, Gemini 2.5, Claude, Attention insight, Github, Eppo to run experiments, BigQuery data from Curated Base Table (CBT), Synapse internal data tool.

Final UI

This UI reflects the consolidated rollout of Bundles 1, 2, and 3, aligned with the broader Zalando brand refresh.

AI tools Design thinking phases

I used AI tools to streamline parts of the design thinking process, treating them as decision-support rather than answers in themselves. I selected tools based on their model strengths and used them only where they added clear value.

✅ UX Strategy

When I joined Zalando in April 2025, the vision for the Performance Marketing Channel — including the why and what — as part of the strategy had already been defined through Google/Meta workshops Baymard audit and and internal prioritisation.

With information scattered across different PDFs, I used ✦ Claude Opus 4.5 to generate a simplified matrix showing Goals, Vision, and Objectives for the LPO project

Goals LPO project

So I translated that in a understandable structure after reviewing it with the PM and the Head of Design.

We had to re-prioritise along the way. After reviewing initial KPI estimates with Analytics, we recalibrated the end of 2026 goals to –8% BR, +4% CVR, and ~€45M GMV. Due to a potential cloaking risk, SEO traffic was excluded from experiments, which removed Catalog tests and shifted focus to PDP experiments using paid traffic only. Goals were recalculated accordingly.

• Goals to initiatives 2025

For 2025, targets were BR −5%, CVR +4%, and ~€10M GMV, adjusted after cancelling SEO catalog tests. The roadmap evolved from three initial bundle hypotheses, iterating on learnings and constraints, while future bundles were designed to enable 2026 personalisation. Impact was estimated using historical CRO benchmarks, traffic data, and Baymard insights.

The how and when were still to be defined, meaning the next step was to lay out the execution plan and timeline for delivering the strategy. With financial pressure to deliver impact quickly, I accelerated Bundles 1–3 by prioritising bounce-rate reduction, recognising that lowering early drop-off was the necessary first step to unlock conversion and GMV uplift.

• Goals to initiatives 2026

So around October 2025 we started to create the OP planning 2026.
I co-led a workshop with a PM focused on hypothesis definition and initiative grouping for end-to-end customer journey OP planning, helping structure early thinking and opportunity areas using insights from Google and Meta workshops and 2025 data.

Around November 2025, we consolidated the 2026 roadmap by carrying over approved but unshipped Bundles (4–7), prioritised by BR and CVR impact and kept as a flexible backlog, while assessing their applicability to the native app.

We created this maturity-phase roadmap, moving from discovery → optimisation → AI-driven scale.

✅ Empathise

• Data analysis and quantitative research

At Zalando, we didn’t have Hotjar or similar behaviour-tracking tools, but we did have internal analytics. I collaborated closely with the analytics team to analyse the data in the sheets

To prepare for ideation, we compiled the key insights regarding the BR PDP into a Google Slides deck so the team could align on what mattered most

So I analysed the interactions before the user bounce across the key areas of the PDP.

• Qualitative Research

We already had solid research from Google’s mobile lab and prior interviews. I focused on extracting key insights—understanding user intent when landing on the PDP and identifying the main problems. Some proposals emerged from this workshop, but I re-evaluated them when I joined the company.

So I used ✦ Claude Opus 4.5 to extract all this information from a PDF report that Google created, Key insights, user behaviour, main problems to solve...just to have it more simplified and clear.

Afterwards I placed it into Miro, and using ✦ AI I mapped it out with post its an clustered them by key words...
Although we had an action plan from Google, we started to narrow down the opportunities to quick wins for PDP (O1) based on impact and effort

With SEO removed from the experimentation roadmap (Catalog O2), I focused on mapping the key sub-problems for PDP (O1) extracting insights from the workshop documentation.

Then, simplified everything into two pivotal problem for the PDP landing experience for users arriving from SEA. As a team, we focused on O1(PDP), holding O3 (Meta/facebook) until we could evaluate the results from O1.

✅ Define

User: Users landing from Performance Marketing channels (SEA, Meta) arrive with an intent to either explore or purchase, expecting relevant products, easy discovery and a fast path to decision.

Needs: They need landing experiences that match their search intent, surface similar alternatives effortlessly, provide enough information to evaluate options, and minimise friction when continuing the journey or deciding to buy. 

Insights: Research and analytics show high bounce rates, repeated return to Google/Meta, and friction in discovery flows — indicating that current PDP page often fail to meet intent quickly or support confident product exploration, resulting in drop-off before conversion.

☝Before ideation…

• Ways of working

As the team was new and I was working remotely, I established structured workflows with the Head of Product and PM to ensure clear iteration steps, cross-team alignment, and smooth delivery from exploration to launch.

• ZDS (Zalando Design system) and Brand Refresh

At Zalando, we worked with a large, mature design system containing all tokens, tiles, variables, and core components. When exploring hypotheses for Bundle experimentation, applying this system was mandatory.

An added complexity was the ongoing brand refresh delivered through continuous test-and-release. Close alignment with the ZDS team was required, but by using existing design system components and tokens, updates would automatically inherit the new styling when components were refreshed.

• Accessibility compliance

Following EAA (WCAG 2.2) requirements, each bundle included dark mode, landscape, 200% dynamic text, and screen-reader specs, reviewed weekly.

• Cross-team experimentation visibility

The LPO project ran experiments alongside other PDP initiatives, including QPE and TFC, each with different goals and audiences. To avoid overlap, manage dependencies, and maintain a holistic view of the PDP, we aligned closely across teams with a tracking artifact

• Holdout group set up and approach

We created a holdout group to maintain a clean baseline and accurately measure the incremental impact of our experiments.   Holdout setup & measurement approach

• Holdout group created to preserve a clean BaU baseline
• Holdout traffic excluded from LPO experimentation and feature exposure
• Purpose: measure true incremental impact over time

✅ Ideation Bundle 1-3 strategy

With financial pressure to deliver impact by August 2025, the approach was to explore Bundles 1, 2, and 3 in parallel to validate their potential impact and accelerate the creation of the roadmap—clarifying which bundle should be prioritised and run first based on expected impact.

• Bundle Strategy framework

I optimised the PDP as a system, assessing scalable solutions for Performance Marketing and their potential to scale to the wider Zalando audience.

• Rollout Governance & Strategy

We created this document that defines a clear and consistent process for rolling out successfully tested features. It ensures decisions are evidence-based, aligned across stakeholders, and low-risk. The goal was to enable faster and more confident rollouts.

• Hypothesis generation

While the Google workshop helped identify quick wins and high-impact opportunities, my focus was on validating hypotheses through early feedback.

I used ✦ Figma Make as a thinking accelerator to explore directions rapidly—prioritising speed over polish to initiate design conversations.

After rapid hypothesis generation in Figma make, I aligned with the PM and lead engineer to prioritise PDP improvements by impact and feasibility, informed by Baymard report and Google research—sequencing Sticky CTA, Breadcrumbs, and Similar Items into Bundles 1–3 on the roadmap.

Bundle 1 (Sticky CTA)

For Bundle 1, I focused on a clear learning objective: understanding how quickly we could move users through the conversion funnel when they land on the PDP from SEA and Meta channels.

• Context

When viewing a specific product, users struggle to quickly access key information, while banners and pop-ups distract from the path to checkout, causing hesitation.

• Data/Insights

Early research signals and insights from the initial Google workshop pointed to this behaviour...

• Competitive analysis

I used ✦ ChatGPT to analyse the pros and cons of introducing a sticky CTA, assessing its suitability for our specific use case, user intent, and the Performance Marketing channel.

• Let’s analyse banners and other distractions, and assess whether they can be removed.

The initial hypothesis was that removing non-essential elements would improve discoverability and focus user attention.

• Iteration

Highlighting the most representative iteration to speed up progression through the conversion funnel

We aligned on exploring how “the sticky add to bag ” could be leveraged to create greater impact. So I created a quick prototype in Figma to gather early feedback applying removed banners and AI assistant and hide search on page load...

Despite resistance from most departments, since we were changing the user behaviour, the change was justified for the LPO team to evaluate how fast users could be moved through the conversion funnel. So Based on feedback, I aligned with the content team to refine the copy, and redo the flow with “Add to bag”——> “Go to bag”

• Prototype

To create the Bundle 1 prototype, I worked directly in Figma using existing design system components, which was faster and more accurate than Figma Make, as it does not yet fully preserve ZDS components, tokens, and spacing. A high-fidelity prototype was required for EDD inclusion and design sign-off.

A key part of each bundle was aligning with the LPO engineering lead and Reference Store engineers to review scenarios, flows, and edge cases (logged-out vs. logged-in), assess technical constraints, identify blockers, and ensure alignment between production and design-system components.

• Design solution

So I ended up with this design solution...

• Trade-off

From an implementation standpoint, I worked closely with engineers to iterate on the auto-scrolling behaviour when users land on the PDP. We explored several technical approaches with a focus on keeping the solution lightweight and maintainable within the Reference Store. To compare options, we created a table outlining the trade-offs, and ultimately leaned toward Solution 1.

• Experiment / Hypothesis-led

I structured each bundle around 5 key elements: the customer problem, the underlying rationale or hypothesis, the target audience, the associated risks and learnings we want to understand.

• Bundle forecast impact

Representative example applied across all bundles

I used ChatGPT as an analytical aid to evaluate whether the proposed changes were likely to drive measurable impact

• Predictive attention

Representative example applied across all bundles

I also used the ✦ Attention Insight plugin to validate visual hierarchy, attention distribution, and CTA prominence before moving into experimentation Document.

• Handoff to engineers and A11y compliance

Representative example applied across all bundles

Accessibility compliance was mandatory at Zalando, so I focused on screen reader annotations and ensuring the required handoff aspects for engineering. Since we used design system components, dark mode and 200% dynamic text were already implemented, leaving mainly annotation order and guidance notes to refine.

• EDD (Experiment design document) Rollout decision and & KPIs

This document defines the experimentation framework for the Sticky “Add to Bag” CTA on mobile PDPs for Performance Marketing traffic. I worked closely with the PM and analyst on problem statement, hypothesis, KPI definition, guardrails, and decision criteria.

➡ I challenged the measurement strategy (new flow intentionally limited multi-size selection), by introducing GMV before cancellations and return rate alongside GMV after returns, ensuring that conversion gains were evaluated against order quality and long-term GMV impact.

• AI Analysis results

Representative example applied across all bundles

Although the learnings were documented in the EDD, I captured screenshots of the key EPPO metrics and used ✦ ChatGPT to extract preliminary insights as guidance before reviewing the results in depth with the analytics team.

• EPPO Results and Key insights

Analysis results experiment Bundle 1

• 🚀 Learnings and Rollout decision

A sticky Add-to-Bag CTA that opens the size picker helps users move faster into the purchase flow by making the primary action constantly accessible. Engagement improved and return rates showed no significant impact, validating rollout with GMV quality guardrails.

• BR: -0.71%
• CVR: CVR is treated as neutral / inconclusive.

Rollout decision: The hypothesis was validated and we rolled out this feature to PM channel.

☝ In the end, Bundle 1 was released after Bundle 2 and positioned accordingly in the roadmap, as it required more time to implement than initially expected.

• 🚩 Why the Conversion Rate remained neutral?

This forced login gating interrupts purchase intent for logged-out and directly suppresses mobile web CVR by blocking high-intent users at critical moments. In 2026, we plan to address this by exploring alternative solutions—such as guest checkout—. This change will affect all bundles and is expected to deliver a significant uplift in CVR.

• Foundational personalisation architecture

After Bundle 1, we began talks on 2026 model capabilities, shifting Performance Marketing toward dynamic, intent-aware, session-adaptive system

To ensure the PDP could scale beyond short-term optimisation in 2026, I mapped the capabilities of the underlying personalisation models—distinguishing between session-level intent, history-based signals, and what works reliably for logged-out PM users.

The effects of personalisation will not be fully felt or visible until the end of 2026