How to transform a complex data environment of a global marketplace into a robust infrastructure ready for advanced AI optimisation, laying the groundwork for scalable revenue growth
Building the Foundations of AI-Driven Pricing at Groupon
From Spreadsheets to AI
Groupon partnered with FLO to transition its pricing strategy from traditional, manual spreadsheet-based hypotheses to an automated, machine-learning (AI)-driven approach. The initial challenge centred on navigating a highly intricate marketplace model across a massive enterprise scale. Groupon required a structural modernisation of its infrastructure to establish a single source of truth, eliminate data inconsistencies, and mitigate the strategic risks associated with semi-automated experiment evaluations.
About Groupon
Groupon is a global e-commerce leader operating a massive marketplace built on a complex discount-based model. This business model is highly specific due to multi-level margins, permanent discount promotions, and the dynamic participation of local partners (vendors) who share discount costs. From a technological standpoint, this requires processing a massive number of variables in real time.
Fixing Data Layers
To unlock automated pricing, we executed an infrastructure-first transformation focused on fixing foundational data layers. We built custom pipelines to address a critical machine learning gap in logging all price details for each offer and its variations. This involved a 6-month data engineering effort to resolve tracking anomalies and unify session identity across mobile and desktop platforms. Simultaneously, we spent 3 to 4 months replacing legacy spreadsheet evaluation with an automated, database-integrated testing framework to ensure code-based validation.
"FLO helped us to build needed data infrastructure and foundations of Machine Learning based pricing optimization. Early results are showing significant uplifts in revenue per user session."
Quentin Neuffert, Pricing Manager, Groupon
Database-Integrated ML
The architecture was designed to shift Groupon from static analysis to a rigorous, ML/AI-driven discipline. The core backend engine utilises robust data processing pipelines to automate data cleansing and enforce multi-layer validation checks directly within the database. For advanced business intelligence and granular segmentation (deep dives by category and price groups), we implemented customised dashboards. This high-quality data layer laid the groundwork for custom ML/AI models optimising for Gross Profit based on elasticity and seasonality.
Results in Numbers
50k+
SKUs automatically priced every day through Machine Learning algorithm.
3
operational dashboards to allow transparent analysis of Machine Learning decision making.
1
automated data pipeline built to entirely eliminate manual spreadsheet errors.
6%
Up to 6% increase in total revenue and conversion rates achieved during initial test batches.
Project outcomes
AI/ML Infrastructure Readiness: We successfully resolved the collection of data on prices at which items did not sell, creating the necessary, clean data foundation required for reliable algorithmic pricing.
Automated A/B Evaluation: We transitioned the experimentation framework from manual Excel processing to automated data pipelines, completely eliminating human error and drastically shortening reporting cycles.
Unified Tracking and Visibility: Resolving cross-platform session tracking, allowing business leaders to utilise custom dashboards for highly accurate, granular performance deep-dives instead of relying on binary "worked/failed" metrics.
"Dynamic pricing on an enterprise scale requires a long-term vision and strategic commitment. By prioritising a thorough modernisation of the data infrastructure over a rushed algorithm deployment, we built a permanent, scale-ready springboard for AI."
Viktor Šohájek, AI Lead, FLO
Technical Setup
Platform: Custom-built data pipeline integrated directly with Tableau for advanced business intelligence and visualisation.
Data: Comprehensive logging of all price variations (including unsold items) and unified cross-platform session tracking, resolving critical Session ID parsing bugs.
Analytics: Fully automated A/B testing framework replacing manual spreadsheets, enabling granular performance deep-dives by specific categories and price groups.
Architecture: Scalable Machine Learning foundation tailored for a highly complex marketplace, supporting dynamic discounts and multi-level shared margins between the platform and local vendors.
How AI Initiatives Force a Systemic Data Cleanup
Implementing complex machine learning models in large-scale retail environments requires a realistic strategic horizon, often taking 12 to 24 months to full statistical validation. A key takeaway from the field is that an ambitious AI initiative acts as an excellent corporate health check. By demanding total data accuracy, the implementation process naturally exposes hidden business vulnerabilities and forces a systemic cleanup that elevates the company's overall operational health.
This project wouldn't be possible without the help of these very important guys:
Viktor Šohájek – AI Lead
Václav Čepelák – Senior Data Scientist
Kateřina Husičková – Data Scientist
Are the limits of your current architecture holding back your AI ambitions?
We specialise in helping enterprise companies modernise legacy systems, eliminate technical debt, and build robust data pipelines that drive real business growth.
