How a Swiss SME Used AI to Optimise Customer Query Handling

Meeting the Challenge of Customer Service in a Swiss SME

Facing rising expectations for fast, accurate customer support, a Swiss mid-sized retailer turned to AI-powered automation to streamline its customer query process. By deploying a large language model (LLM)-based system, they reduced manual workload, improved response speed, and enhanced customer satisfaction—all while ensuring compliance with Swiss and EU regulations.

The Problem: Slow and Costly Manual Query Handling

A Zurich-based SME in the consumer goods sector struggled with high volumes of incoming customer emails and messages. The small support team was overwhelmed, leading to:

• Delayed response times (often 2–3 business days)
• Human errors in query categorisation and prioritisation
• High staff workload and rising wage costs
• Occasional customer churn due to dissatisfaction

With competition intensifying and a recent Deloitte Switzerland survey confirming that nearly half of Swiss SMEs now see AI as a way to reduce wage costs, the leadership knew that an upgrade was needed.

The Solution: AI-Driven Customer Query Triage

The SME opted to pilot an AI solution based on a sovereign, Switzerland-hosted LLM—leveraging the newly released Apertus 1.5 model from ETH Zurich and EPFL for its strong reasoning and multimodal capabilities. The system was designed to process incoming emails and chat messages, automatically categorise them (e.g., returns, product info, complaints), and assign urgency scores.

Crucially, the AI was configured to:

• Auto-tag and route queries to the right team member or department
• Draft suggested responses for common questions
• Escalate high-priority cases (like complaints) directly to managers
• Support German, French, Italian, and English
• Operate entirely within Swiss data privacy law and EU AI Act boundaries

Benefits Realised

In the first three months, the SME reported:

• Response times halved: Median response time dropped from 48 hours to under 24 hours
• Labour time savings: Staff spent 40% less time triaging emails
• Fewer errors: Accurate routing reduced internal escalations by 60%
• Improved customer satisfaction: Net Promoter Score (NPS) improved by 17 points
• Lower operational costs: Overtime and temporary staffing needs decreased

Step-by-Step Implementation

1. Internal Needs Assessment

   • Mapped out pain points (delays, team workload, compliance needs)
   • Quantified query volumes by channel and language

2. Vendor and Model Selection

   • Chose a Swiss AI integrator with experience in secure, on-premises LLM deployment
   • Prioritised models with robust multilingual and multimodal support (Apertus 1.5)

3. Integration and Data Preparation

   • Integrated the AI system with email and live chat platforms via secure APIs
   • Fed anonymised historic query data to fine-tune categories and workflows

4. Testing and Human-in-the-Loop Review

   • Piloted with a subset of staff; all AI classifications reviewed by humans during initial phase
   • Adjusted routing logic and feedback loops for accuracy

5. Deployment and Training

   • Rolled out to the full support team
   • Provided staff training on supervising, correcting, and leveraging AI-generated drafts

6. Compliance and Ongoing Monitoring

   • Ensured regular audits of AI outputs for fairness, data privacy, and non-discrimination
   • Documented processes to satisfy regulatory (AI Act) traceability and transparency requirements

Lessons for Other Swiss SMEs

This use case demonstrates that AI-powered customer query triage is not just for large enterprises. With careful planning, the right local model (such as Apertus 1.5), and compliance-first integration, even mid-sized Swiss SMEs can unlock faster service and lower costs—without sacrificing trust or legal security. As AI adoption accelerates across Switzerland, practical, responsible pilots like this are setting the benchmark for what’s achievable right now.