We brought in Viston to personalize merchandising, improve search, and forecast demand across festive seasons for our mid-market retail brand.
Business challenge
- Low search conversion and high stockouts in Tier-2/3 cities.
- Limited personalization across languages and channels (app, web, WhatsApp).
Our approach and solution
- Built a multilingual retrieval-augmented search and recommendations layer.
- Implemented demand forecasting by micro-region with supply planning triggers.
AI applications and benefits delivered
- Personalized recommendations, similar items, and bundles.
- Multilingual semantic search (English, Hindi, Tamil) with query rewrite and RAG.
- SKU-level demand forecasting feeding replenishment and markdown optimization.
Cost of implementation
- USD 39,000 including cloud and data work.
Tools and technologies used
- GCP BigQuery, Vertex AI, Recommendations AI, LangChain, FAISS, Cloud Run, Dataflow, dbt, Looker, Redis, Cloud Armor, GitHub Actions.
Quantitative outcomes
- +0.9 percentage point increase in overall conversion.
- +14% average order value.
- -18% customer acquisition cost.
- -22% stockouts on top 1,000 SKUs.
- 2.1x improvement in search click-through.
Key performance indicators (KPIs) tracked
- Conversion rate, AOV, CAC, CTR, add-to-cart rate, stockout rate, forecast MAPE, latency p95.
Pre- and post-implementation metrics
- Conversion: 2.4% → 3.3%.
- AOV: INR 1,480 → INR 1,689.
- Stockout rate: 13.5% → 10.5%.
- Search CTR: 8.2% → 17.3%.
- Forecast MAPE (top SKUs): 28% → 14%.
Stakeholder quotes or testimonials
- “Personalization finally works in Hindi and Tamil—our Tier-2 growth unlocked.” — Chief Growth Officer.
- “Planners trust the forecasts; we buy smarter pre-Diwali.” — Head of Merchandising.
Regulatory or compliance considerations
- India DPDP Act 2023, consent-based profiling, opt-out mechanisms, PII tokenization, audit logging, model fairness checks for language bias.
