AI Phone Assistant at Deep Impact: First Steps to AI First, incl. Business Case

Deep Impact AG from Winterthur is taking another step towards "AI First": For three weeks, we have been successfully using our own AI-based telephone assistant. This article documents our initial experiences, the business case, and the most important learnings.

English edition — originally published in German as KI-Telefonassistentin bei Deep Impact: Erste Schritte zu KI First, inkl. Business Case.

The Business Case: Why We Implemented AI Telephony

The answer was clear: We wanted to be available 24/7 without someone constantly sitting by the phone.

The Planned ROI:

• Reduction of unanswered calls by 70%
• Time savings for reception of approx. 10 hours per week
• Improved customer satisfaction through quick initial response
• Automated documentation (call summaries)

Implementation: From Concept to Go-Live

Phase 1: Selection and Configuration

1. Recordings of our best-practice conversations for training
2. Configuring AI agents for frequently asked questions
3. Integration with Slack for live notifications
4. Defining fallback scenarios

Phase 2: Pilot Phase and Improvements

What we improved:

• Training with local accents and speech patterns
• Incorporated incorrect detections as training feedback
• Setting escalation triggers
• Defining call quality metrics

The First Results

Concrete figures from the first 3 weeks:

• 847 calls processed
• 712 of these fully resolved by AI (84%)
• 135 escalated to human agents (16%)
• Average call duration: 45 seconds

Call Summaries in Slack

The AI automatically creates a summary of each call and posts it to a dedicated Slack channel.

Example: > «Call from Monika Fischer: Question about Design Sprint offerings. Wants a quote for a 4-day project. Interested in AI integration. Follow-up needed.»

The Challenges

• Name recognition is tricky: Swiss names are often transcribed incorrectly
• Dialects and accents: Swiss German is difficult for AI
• Complex inquiries: For very specific technical questions, the AI still escalates too often

Learnings for Other Companies

1. Start small
2. Improve iteratively
3. Transparency – inform callers
4. Plan for escalation
5. Measure, measure, measure

Conclusion: AI Telephony Pays Off

After three weeks, we can say: The AI telephone assistant was the right decision. It saves time, improves accessibility, and provides valuable data.