AI assistants that convert browsing into basket growth

Client

Hilal, TwinFusion

Retail brands deployed grounded AI assistants for product questions and recommendations.

Project

Tasks

• Catalog and policy grounding via RAG over product and FAQ corpora
• Recommendation flows tied to live inventory and promotions
• Human escalation when confidence or sentiment thresholds trip
• Analytics dashboard for conversation quality and conversion

KPIs

1. 1. 70%+ AI resolution — without human agent for tier-1 product questions
2. 2. Measurable basket lift — on assisted sessions vs. unassisted browse
3. 3. Sub-3s response — p95 latency during campaign traffic
4. 4. Zero policy violations — grounded answers only from approved sources

Project journey

1. 1. PoC sprint

   Two-week grounded assistant on a subset of SKUs with manual eval rubric.

2. 2. Production integration

   Connected to commerce APIs, CRM, and escalation queues.

3. 3. Optimize & scale

   Prompt tuning, caching, and A/B tests on recommendation strategies.

Challenges

Hilal, TwinFusion needed systems that could scale with operational complexity — replacing manual coordination and disconnected tools that slowed decision-making.

• Manual workflows consumed team capacity and introduced errors at scale.
• Legacy tooling could not support modern analytics or automation requirements.
• Integration gaps between departments created data silos and delayed responses.

Results

Testimonial

“Customers get answers instantly — and we finally see which questions drive revenue.”