1. Executive Summary & Overview
Eczacıbaşı Consumer Products (Eczacıbaşı Tüketim Ürünleri), one of Turkey's largest and most established industrial conglomerates, manufactures market-leading hygiene, paper, and personal care brands including Selpak, Solo, and Sanipak. As e-commerce marketplaces and online consumer reviews expanded exponentially, manually analyzing hundreds of thousands of unstructured customer reviews across fragmented retail platforms became impossible using traditional market research methodologies.
Eczacıbaşı partnered with Zeo's Generative AI and Data Science consultancy team to engineer a proprietary Natural Language Processing (NLP) and Artificial Intelligence Review Intelligence platform for the Sanipak brand. The enterprise AI system automated large-scale sentiment analysis, surfaced unmet customer expectations, and transformed raw consumer reviews into actionable product R&D roadmaps—winning "Best Use of AI for Data" at the European Search Awards 2025 and earning recognition in Fast Company Türkiye's "Top 50 Most Innovative Companies" 2024.
2. Challenge & Market Context
Large-scale e-commerce consumer review intelligence presents severe natural language processing hurdles:
- Massive Unstructured Customer Review Volumes: Hundreds of thousands of open-ended customer reviews distributed across major e-commerce marketplaces (Trendyol, Hepsiburada, Amazon Turkey) exceeded human analytical capacity.
- Complex Turkish NLP Morphology & Nuance: Turkish is an agglutinative language with rich morphological suffixation, slang, sarcastic feedback, and misspelled consumer phrasing that standard off-the-shelf NLP tools fail to parse accurately.
- Bridging Unstructured Feedback with Product R&D: Translating abstract consumer complaints ("ambalajı zor açılıyor", "emici gücü yetersiz", "kokusu ferahlatıcı") into precise engineering parameters for manufacturing teams.
- Real-Time Competitive Benchmarking: Continuously scraping, cleaning, and benchmarking competitor product reviews without data loss or classification bias.
3. Objectives & KPI Targets
The strategic artificial intelligence and data science initiative defined clear technological benchmarks:
- Proprietary Turkish NLP Sentiment Pipeline: Develop customized machine learning models capable of high-accuracy aspect-based sentiment analysis on Turkish consumer reviews.
- Automated Aspect-Level Categorization: Classify customer sentiment across key product dimensions (Packaging Usability, Absorption Quality, Scent & Texture, Price-Performance, Delivery Condition).
- R&D Product Development Intelligence: Deliver automated dashboard reports translating consumer sentiment shifts into concrete product improvement recommendations for Eczacıbaşı engineers.
- Third-Party Verification & Industry Recognition: Validate the innovation through rigorous international search and data award panels.
4. Strategy & Innovation
Zeo engineered a next-generation AI Review Intelligence architecture:
- Custom Aspect-Based Sentiment Analysis (ABSA) Architecture: Engineered deep learning transformer models fine-tuned on Turkish e-commerce vocabulary to separate multi-clause reviews containing both praise and complaints.
- Semantic Entity Extraction & Topic Modeling: Applied unsupervised clustering algorithms to discover emerging consumer pain points and unannounced product feature expectations before competitors.
- Executive R&D Dashboard & Alerting System: Built interactive BI dashboards providing real-time sentiment tracking, competitor benchmark matrices, and product flaw alerts for brand managers and product designers.
5. Execution & Implementation Breakdown
- Phase 1 — Data Ingestion & Preprocessing Pipeline:
- Aggregated and cleaned hundreds of thousands of historical and live reviews across major Turkish e-commerce marketplaces.
- Developed custom Turkish lemmatization, spell-correction, and stop-word filtering routines.
- Phase 2 — AI Model Training & Aspect Classification:
- Trained transformer-based models on product attribute categories, achieving high F1-score accuracy in aspect classification.
- Implemented multi-label classification separating product quality feedback from logistics/shipping complaints.
- Phase 3 — Enterprise Integration & Decision Workflows:
- Integrated structured sentiment feeds into Eczacıbaşı Tüketim Ürünleri's brand management and product development decision cycles.
6. Results & Quantifiable Business Impact
The AI Review Intelligence platform delivered transformative enterprise value:
- European Search Awards 2025 Winner: Won "Best Use of AI for Data", validating the project as Europe's premier search and data intelligence innovation.
- Fast Company Türkiye Top 50 Most Innovative Companies 2024: Recognized among Turkey's most pioneering corporate AI innovations.
- 100% Automated Consumer Feedback Mining: Replaced manual review sampling with real-time, comprehensive market-wide review intelligence.
- Direct Product R&D Enhancements: Consumer feedback directly informed physical packaging redesigns, absorbency formulation upgrades, and marketing messaging refinements.
7. Awards & Industry Recognition
The project stands as a benchmark in enterprise AI application and data engineering:
- European Search Awards 2025: WINNER — Best Use of AI for Data
- Fast Company Türkiye 2024: Ranked — Top 50 Most Innovative Companies
8. Methodology & Strategic Key Takeaways
- AI for Data Solves Real Business Bottlenecks: Deploying AI to structure massive unstructured datasets produces immediate, measurable corporate value beyond conversational chatbots.
- Aspect-Based Sentiment Outperforms Binary Classification: Dissecting customer reviews into distinct product attributes (packaging vs quality) provides precise R&D guidance.
- Language-Specific NLP Fine-Tuning Is Essential: Training custom models on regional language nuances is critical for accurate sentiment extraction in complex languages.
