Vodafone's TOBi and SuperTOBi Handle Up to 70% of Customer Interactions at Less Than One-Third the Cost of Live Chat
Vodafone's virtual assistant TOBi, now evolving into the GenAI-powered SuperTOBi across 13 European and African markets, resolves up to 70% of all customer inquiries autonomously — at less than one-third the cost of live agent chat — while SuperTOBi's first-time resolution rate jumped from 15% to 60% in Portugal.
Background
Vodafone's scale — 300 million mobile customers across dozens of markets — makes AI-driven customer service automation both a strategic necessity and an economic imperative. Human agent capacity cannot scale proportionally with customer base growth. TOBi was introduced to handle the high-volume, well-defined inquiry types (billing, troubleshooting, account management) that represent the majority of incoming contacts.
What Was Implemented
- TOBi: IBM watsonx Assistant–powered virtual assistant in 13 countries, ~14 languages; handles billing, technical support, account management
- SuperTOBi: Microsoft Azure OpenAI–powered next-generation assistant; interprets full sentences/phrases; initial rollout Italy, Portugal, Germany, Turkey (July 2024); €140M investment
- SuperAgent: Microsoft Azure OpenAI Copilot-based human-agent assist tool; provides conversation summary at transfer; piloted in Ireland
- SuperSearch: enhanced search facility for customer-facing websites
- IBM watsonx.ai pilot: GenAI-enhanced journey testing and gap analysis for TOBi conversational design (99% turnaround improvement; gap analysis from N/A to <5 min per journey)
Results
TOBi (across 13 markets): - Up to 70% of customer inquiries resolved autonomously - Cost per automated interaction: less than one-third the cost of live agent chat SuperTOBi (Portugal, early results): - First-time resolution rate: 60% (from 15%) - Online NPS: improved by 14 points to 64 (above 50-point "strong" threshold) IBM watsonx.ai pilot (journey testing): - Journey testing turnaround: 99% improvement (6.5 hours → <1 minute per journey per persona) - Gap analysis: from no viable process → <5 minutes per journey
Lessons
- Combining a high-volume self-service chatbot (TOBi/SuperTOBi) with a human-agent assist tool (SuperAgent) is more effective than deploying either in isolation
- First-time resolution rate is a more meaningful KPI than deflection rate for measuring chatbot quality in telecom contexts
- Natural language understanding that interprets full sentences rather than keywords produces measurably better first-time resolution (15% → 60% in Portugal)
- Investing in AI-assisted journey testing and gap analysis (the IBM pilot) accelerates chatbot development quality and speed, reducing time-to-improvement for customer-facing flows