Reducing Human-Agent Escalations with AI Self-Service Across Text and Voice
Client
One of the largest providers of energy, internet, television, and mobile services in the Netherlands.
Business Case at a Glance
- Operating constraint: Routine requests consumed capacity that the support team also needed for complex or sensitive conversations.
- Intervention: AI self-service across text and voice, grounded in the provider’s knowledge base and backed by human escalation.
- Client-reported operational outcome: The client reported a 30% decrease in the percentage of customer conversations escalated to human agents.
- Business relevance: Escalation percentage tracks the share of support demand routed to the human team rather than chatbot usage alone.
Business Challenge
The provider supports customers across four service lines. Routine questions and requests that require human judgment competed for the same support-team capacity, limiting the time specialists could give to complex or sensitive cases.
The business needed to reduce routine demand on human agents without restricting customers to a single interaction channel or removing access to personal support. Because the assistant would communicate directly with customers, its response quality also needed to be measurable.
What DevRain Delivered
DevRain built an AI support assistant for text and voice interactions in Dutch and English. It gives customers a self-service path for routine requests and allows conversations that need personal assistance to escalate to a human agent.
Customer-Facing AI Controls
Customer-facing AI introduces service and brand risk if its answers cannot be evaluated or customers cannot reach a person when needed. The solution was designed to mitigate those risks through:
- Knowledge grounding: The assistant uses the provider’s own knowledge base to keep responses aligned with company information.
- Human escalation: Customers can escalate the conversation to a human agent when personal assistance is required.
- Response-quality evaluation: The team built an AI evaluation framework that measures response quality and provides a consistent basis for improvement decisions.
Client-Reported Operational Outcome
The client reported a 30% decrease in the percentage of customer conversations escalated to human agents.
Escalation percentage is an operational workload indicator: it measures what share of customer conversations reaches human support through escalation. Tracking it tied the assistant to assisted-support demand rather than chatbot usage alone.
Technologies
- Azure Foundry
- .NET
- Microsoft Agent Framework
- Microsoft Bot Framework