Sobot Unites Generative and Multi-Faceted AI for 90% Accuracy and 88% Independent Resolution in Customer Contact
Sobot's "Five-AI" mix of Generative and Multi-Faceted AI posts 90%+ accuracy and 88% independent resolution. RAG-based answers plus role-based tools lift CSAT, AHT, onboarding.
Generative + Multi-Faceted AI: Practical Gains for Customer Support
Sobot released new results for its updated Sobot AI after testing with 100+ customers. The system delivered an average response accuracy above 90% and an independent resolution rate of 88%.
The approach behind these gains is simple and effective: combine Generative AI for precision answers with Multi-Faceted AI for role-specific workflows. Sobot calls it the "Five-AI" system-anchored by Generative AI and Multi-Faceted AI-to deliver human-quality service at scale.
What's new in Sobot Generative AI
Generative AI in customer service is only as good as its grounding and guardrails. Sobot's stack pairs Retrieval-Augmented Generation (RAG) with advanced large language models (LLMs) and targeted Small Language Models (SLMs) to keep answers accurate and brand-aligned.
- Intelligent chunking: Semantic and structure-aware splits preserve meaning, improving retrieval quality.
- Precise retrieval: Automatic query rewriting, recall thresholds, and re-ranking raise hit rates on the right content.
- Enhanced generation: Integrations with models from Claude, OpenAI, Amazon Bedrock, DeepSeek, and others produce fluent, natural responses while reducing hallucinations through style controls and brand tone.
- SLMs for targeted tasks: Focused models tackle repeat use cases. "SLMs are designed for solving specific problems in various industries... order tracking, recommendations, returns, refunds," said Yi Xu, CEO of Sobot.
If you want a quick primer on RAG, see this overview from Microsoft here.
Multi-Faceted AI for day-to-day operations
- AI Agent: Handles chat, voice, email, and social. Configurable tone, style, and length. Multilingual support for global teams. Aims to resolve more inquiries without handoffs.
- AI Copilot: Assists human agents with conversation summaries, content expansion and polishing, and one-click ticket fields-helping reduce handle time and improve consistency.
- AI Insight: A single dashboard with reports, analytics, intelligent QA, and VOC. Over 300 indicators to spot gaps fast and make informed adjustments.
What this means for your KPIs
- Independent resolution/FCR: More queries closed without escalation.
- AHT: Faster retrieval, better drafts, and smarter routing cut time per case.
- CSAT and QA: Consistent voice, fewer errors, and clearer explanations.
- Cost per contact: Automation where it fits, assist where it doesn't.
- Onboarding: Copilot guidance shortens the ramp for new agents.
How to implement with low risk
- Prioritize intents: Start with the top 15-20 high-volume, low-variance topics.
- Ground your knowledge: Centralize policies, SOPs, and product docs; remove duplicates and stale content.
- Tune retrieval: Set recall thresholds and test re-ranking on real tickets before going live.
- Brand controls: Lock tone, banned phrases, and escalation rules. Require citations for sensitive answers.
- Human-in-the-loop: Clear triggers for handoff (risk, sentiment, PII, high value accounts).
- Measure weekly: Track independent resolution, AHT, CSAT, deflection rate, and QA outcomes by intent.
- Compliance and security: Define data retention, redaction, and access policies up front.
Where Sobot fits
If you're running multi-channel support and want higher resolution without growing headcount, this mix of Generative AI + Multi-Faceted AI is built for that job. The reported 90%+ accuracy and 88% independent resolution signal real progress in day-to-day metrics-not just demos.
As Xu put it, "Sobot AI is not just the stacking of functions, but the deep integration with applications and users."
