How flaschenpost lifted sitewide conversion and reduced zero-result searches with Constructor
SUMMARY
Executive Summary
Learn how flaschenpost, one of Germany’s fastest-growing online grocery and beverage delivery services, transformed search and discovery across its full assortment:
-
Upgraded from a legacy keyword search engine to an AI-first platform aligned with revenue, margin, and conversion KPIs.
-
Eliminated manual long-tail maintenance through automatic handling of misspellings, complex queries, and new products.
-
Validated performance through a rigorous AA/BB test that delivered a 4.14% lift in sitewide conversion.
ABOUT FLASCHENPOST
flaschenpost is a leading German on-demand delivery service offering beverages, groceries, and household essentials with two-hour delivery across 200+ cities.
Founded in 2016 and now part of the Oetker Group, the company has expanded from beverages into a full online supermarket, powered by a sophisticated logistics network and tech-driven operations.
Founded in 2016 and now part of the Oetker Group, the company has expanded from beverages into a full online supermarket, powered by a sophisticated logistics network and tech-driven operations.
The Challenge
A Keyword System Misaligned With Business KPIs
flaschenpost’s legacy search engine was fast, but it was built around technical relevance (literal keyword matching) rather than the business outcomes that mattered most for their full-assortment e-grocery model. It couldn’t evaluate the “attractiveness” of search results, or in other words, how likely they were to satisfy the shopper’s true intent and drive a purchase.
“We weren’t just trying to fix relevance. We needed a system that understood what ‘good’ actually means in a supermarket context — not just what matches a keyword.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
As a result, teams spent significant time manually steering results toward what customers actually wanted, rather than what the system could determine automatically.
The Challenge
Long-Tail Complexity and High-Maintenance Operations
flaschenpost’s assortment is broad, constantly changing, and filled with grocery-specific descriptors that a keyword engine struggled to interpret. High-intent searches like “gestückelte tomaten dose” (canned diced tomatoes) returned zero results, even though the correct products were in the catalog. Queries like “Dubai” — which shoppers used to find a specific type of chocolate — also produced empty pages. And simple misspellings (13% of all searches) such as “Sagt” instead of Saft (juice) sent shoppers to irrelevant iced tea products.
flaschenpost’s assortment is broad, constantly changing, and filled with grocery-specific descriptors that a keyword engine struggled to interpret. High-intent searches like “gestückelte tomaten dose” (canned diced tomatoes) returned zero results, even though the correct products were in the catalog. Queries like “Dubai” — which shoppers used to find a specific type of chocolate — also produced empty pages. And simple misspellings (13% of all searches) such as “Sagt” instead of Saft (juice) sent shoppers to irrelevant iced tea products.
The Challenge
A Reactive, Resource-Intensive Workflow
To maintain results quality, flaschenpost’s merchandising and Data Science teams spent considerable time creating synonyms, redirects, and slotting rules to compensate for gaps. Many fixes were reactive, triggered only after failures surfaced in dashboards or internal testing.
“So much of our work required dedicated manual effort. We tuned essential queries. But that’s not optimization — and it’s not scalable for a full supermarket assortment.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost.
Recognizing the Need for Change
flaschenpost had begun its digital transformation with a full rebuild of its PIM and underlying e-commerce platform. Once those were in place, they turned their attention to search. They reviewed several modern semantic search approaches, including an upgrade path from their existing provider, but nothing meaningfully addressed the core issues around long-tail interpretation, misspellings, and KPI alignment.
Constructor became a serious contender once the team entered a Proof Schedule — a lightweight POC where Constructor collected behavioral data from flaschenpost’s site for a short period and used it to project how its algorithms would optimize their results. These projected results are presented in an interactive sandbox environment, which provided flaschenpost with a clear view of how their assortment and queries would perform on Constructor without requiring a full integration.
Testing their hardest grocery queries in the sandbox exposed how differently an AI-first system could interpret semantics, handle descriptive and misspelled searches, and surface attractive products automatically.
The Proof Schedule reinforced those early impressions with measurable results and hands-on collaboration, giving the team confidence that Constructor could scale with their business.
“Seeing our assortment perform inside the Playground made it obvious the technology understood the structure and semantics of groceries in a way our old search never could.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
Recognizing the Need for Change
flaschenpost had begun its digital transformation with a full rebuild of its PIM and underlying e-commerce platform. Once those were in place, they turned their attention to search. They reviewed several modern semantic search approaches, including an upgrade path from their existing provider, but nothing meaningfully addressed the core issues around long-tail interpretation, misspellings, and KPI alignment.
Constructor became a serious contender once the team entered a Proof Schedule — a lightweight POC where Constructor collected behavioral data from flaschenpost’s site for a short period and used it to project how its algorithms would optimize their results. These projected results are presented in an interactive sandbox environment, which provided flaschenpost with a clear view of how their assortment and queries would perform on Constructor without requiring a full integration.
Testing their hardest grocery queries in the sandbox exposed how differently an AI-first system could interpret semantics, handle descriptive and misspelled searches, and surface attractive products automatically.
The Proof Schedule reinforced those early impressions with measurable results and hands-on collaboration, giving the team confidence that Constructor could scale with their business.
“Seeing our assortment perform inside the Playground made it obvious the technology understood the structure and semantics of groceries in a way our old search never could.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
The Solution
An AI-First Platform Optimized for Grocery Commerce
flaschenpost chose Constructor because it offered an AI-first approach purpose-built for ecommerce rather than a keyword system with semantic layers bolted on. Real-time personalization and KPI-based ranking enabled the platform to automatically align results with conversion, revenue, and margin targets — a capability that their previous solution could not achieve.
“Most tools promise AI, but Constructor showed us what AI looks like when it’s tied to business outcomes. Ranking decisions finally made sense in the context of our KPIs.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
Instead of relying on manual overrides, the system continuously learned from shopper behavior, assortment changes, and intent patterns across the full journey.
flaschenpost chose Constructor because it offered an AI-first approach purpose-built for ecommerce rather than a keyword system with semantic layers bolted on. Real-time personalization and KPI-based ranking enabled the platform to automatically align results with conversion, revenue, and margin targets — a capability that their previous solution could not achieve.
“Most tools promise AI, but Constructor showed us what AI looks like when it’s tied to business outcomes. Ranking decisions finally made sense in the context of our KPIs.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
Instead of relying on manual overrides, the system continuously learned from shopper behavior, assortment changes, and intent patterns across the full journey.
Transparency and Control Through Glassbox AI
Constructor’s Glassbox AI framework gave flaschenpost visibility into how ranking decisions were made. Instead of debugging blind or relying on intuition, the team could see the exact signals and weighting that influenced results — enabling intentional, high-leverage interventions rather than broad, reactive fixes.
“We didn’t want a black box. With Constructor, we could see the reasoning behind the rankings and adjust intentionally instead of reacting blindly.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
This level of transparency was a key differentiator during the evaluation and remains a core part of the merchandising workflow today.
Continuous Improvement Through Reinforcement Learning
A major advantage for flaschenpost was Constructor’s reinforcement-learning architecture, where Search, Browse, Recommendations, and other discovery touchpoints continuously learn from real shopper behavior. Instead of operating as separate tools, each part of the experience reinforces the others — improving understanding of intent, product attractiveness, and what leads to successful outcomes.
This created a self-improving loop: every click, view, and purchase helped strengthen future rankings across the entire shopping journey. For a high-change grocery assortment, this adaptive intelligence was critical.
Our Solution
Validation through the Proof Schedule™
Constructor's Proof Schedule™ process allowed Cromwell to see concrete evidence of improved search results before making a commitment. By implementing Constructor's tracking beacon during the RFP process, Cromwell gained:
-
Clear visibility into how Constructor's results differed from their current solution
-
Concrete metrics to build a compelling business case
-
Before-and-after results demonstrating the enhanced customer experience
-
Clear visibility into how Constructor's results differed from their current solution
-
Concrete metrics to build a compelling business case
-
Before-and-after results demonstrating the enhanced customer experience
"We could already see how the results were different from what we were currently presenting. It was enough for us to see that we were tapping into something that would make an immediate impact on how our search worked," Powell explains.
Beyond the technical improvements, Cromwell appreciated Constructor's user-friendly interface: "We could see from the Constructor interface that it was going to make it easier for our merchandising and marketing people to understand how search is being used and what tools they have available to them to try out new strategies... just easier to get non-technical people on board."
The Results
flaschenpost validated clear, measurable improvements after adopting Constructor
The implementation process reinforced Cromwell's confidence in Constructor. With comprehensive documentation and direct access to Constructor's engineering team, Cromwell's technical team was able to integrate the solution efficiently.
"Constructor quickly became an extension of our internal team," Powell notes. The composable, API-first approach enabled Cromwell to implement incrementally without a disruptive "big bang" switchover, allowing for gradual rollout and testing.
The impact was immediate and substantial:
-
4.14% relative lift in sitewide search conversion (AA/BB test), exceeding their original goal of 3.2%
-
Significant reduction in zero-result queries, especially across long-tail and misspelled searches
-
Automatic handling of the 13% of searches containing misspellings, eliminating the need for manual correction
-
Sharp decrease in manual merchandising work, including synonyms, redirects, and boosts
-
Higher-quality result relevance leading to stronger add-to-cart and purchase behavior across core grocery missions
What's Next
A True Partner in Innovation
flaschenpost emphasized that Constructor’s collaboration stood out from the first conversations through ongoing optimization. The team highlighted the consistency of support across sales engineering, data science, and customer success, particularly during the Proof Schedule and early rollout phases.
Constructor’s responsiveness, analytical rigor, and willingness to engage deeply with flaschenpost’s KPI goals helped establish trust quickly.
“It never felt like a handoff from one team to another. The consistency of support made it clear we were working with a partner who understood our goals and shared accountability for the outcomes.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
This partnership mindset continues as flaschenpost scales into new categories and evolves its broader digital experience.
flaschenpost emphasized that Constructor’s collaboration stood out from the first conversations through ongoing optimization. The team highlighted the consistency of support across sales engineering, data science, and customer success, particularly during the Proof Schedule and early rollout phases.
Constructor’s responsiveness, analytical rigor, and willingness to engage deeply with flaschenpost’s KPI goals helped establish trust quickly.
“It never felt like a handoff from one team to another. The consistency of support made it clear we were working with a partner who understood our goals and shared accountability for the outcomes.”
— Julian Neumann, Head of Product - E-Commerce, flaschenpost
This partnership mindset continues as flaschenpost scales into new categories and evolves its broader digital experience.
What's Next
flaschenpost’s roadmap focuses on building an even faster, more intuitive shopping experience powered by ongoing optimization and next-generation AI capabilities:
-
Richer search suggestions and category cues within the flyout to help shoppers complete missions faster
-
Advanced category and browse experiences with personalized sort logic and adaptive filters
-
AI-driven search expansions that interpret broader shopper missions (e.g., full meal components, complementary products)
-
Continuous Optimization Program (COP) to incorporate contribution margin data directly into ranking for stronger profitability outcomes
With Constructor, these initiatives are designed to compound over time, strengthening both customer experience and business performance.