brianletort.ai
← Library

Vodafone Germany

Mixed evidenceExceptions onlyHuman quality loop

TOBi to SuperTOBi customer-service triage

A service bot does not need to follow a fixed decision tree and escalate nearly every complex request; generative AI can resolve more requests while people handle exceptions and review response quality.

Customer service · Germany

An editorial scene for Vodafone Germany contrasts handle simple navigation questions and requests that followed fixed decision processes. with interpret complex or vague requests and generate answers from vodafone web and internal product information. in the tobi to supertobi customer-service triage workflow.

Executive brief

The operating-model shift, in one view.

The reusable pattern is a three-part operating model: automation resolves more requests, humans own exceptions, and experienced staff continuously audit response quality.

AI value · First-contact issue resolution

Company-reported

Vodafone reported that SuperTOBi resolved 25% more issues at first contact and independently resolved 74% of all inquiries in under ten seconds.

Vodafone-reported production figures without disclosed cohort methodology; '25% more' and the move from more than 50% to 74% may not use identical measurement definitions.

Before

Handle simple navigation questions and requests that followed fixed decision processes. → Receive almost every second request that the chatbot could not resolve.

After

Interpret complex or vague requests and generate answers from Vodafone web and internal product information. → Focus on complex cases that remain unresolved. → Review answers and dialogues daily for quality and continuous improvement.

Human boundary

SuperTOBi may answer and resolve routine requests; customer advisers own escalated cases, and human reviewers oversee response quality.

Why it matters

A service bot does not need to follow a fixed decision tree and escalate nearly every complex request.

How the work changed

Before

How the work ran before the change.

  1. Step 1 of 2

    TOBi rules-based chatbot

    Handle simple navigation questions and requests that followed fixed decision processes.

    ControlFixed decision trees constrained available answers.

  2. Step 2 of 2

    Customer adviser

    Receive almost every second request that the chatbot could not resolve.

    ControlAdviser resolves complex cases.

What changed

A service bot does not need to follow a fixed decision tree and escalate nearly every complex request.

Decision rightAI handles the default; humans own exceptions

After

How the same work runs now.

  1. Step 1 of 3

    SuperTOBi

    Interpret complex or vague requests and generate answers from Vodafone web and internal product information.

    ControlThe bot escalates requests it cannot resolve.

  2. Step 2 of 3

    Customer adviser

    Focus on complex cases that remain unresolved.

    ControlHuman advisers retain exception resolution.

  3. Step 3 of 3

    Former customer advisers

    Review answers and dialogues daily for quality and continuous improvement.

    ControlDaily human quality review.

Process model built from the published workflow evidence for Vodafone Germany. Every step, actor, and control appears in full below.
Every step, actor, and control

Exception path

Requests the bot cannot resolve are transferred to a customer adviser; response quality is reviewed daily by former customer advisers.

Work removed

  • A reported 25% of first-contact escalations relative to prior TOBi performance
  • Manual handling of additional routine and moderately complex questions
  • Customer waiting for many chatbot-to-adviser transfers

Decision authority

SuperTOBi may answer and resolve routine requests; customer advisers own escalated cases, and human reviewers oversee response quality.

Before

  1. 01

    TOBi rules-based chatbot

    Handle simple navigation questions and requests that followed fixed decision processes.

    Control: Fixed decision trees constrained available answers.

  2. 02

    Customer adviser

    Receive almost every second request that the chatbot could not resolve.

    Control: Adviser resolves complex cases.

After

  1. 01

    SuperTOBi

    Interpret complex or vague requests and generate answers from Vodafone web and internal product information.

    Control: The bot escalates requests it cannot resolve.

  2. 02

    Customer adviser

    Focus on complex cases that remain unresolved.

    Control: Human advisers retain exception resolution.

  3. 03

    Former customer advisers

    Review answers and dialogues daily for quality and continuous improvement.

    Control: Daily human quality review.

Work that left the path

  • A reported 25% of first-contact escalations relative to prior TOBi performance
  • Manual handling of additional routine and moderately complex questions
  • Customer waiting for many chatbot-to-adviser transfers

Human role before

Advisers absorbed nearly half of chatbot conversations because the rule-based system could handle only simpler, structured requests.

Human role after

Advisers focus on unresolved complex cases, while former advisers perform daily quality review of automated responses.

AI roleUnderstand less structured requests, retrieve approved Vodafone information, generate responses, and resolve more customer issues without a live handoff.

Outcomes

First-contact issue resolution

Company-reported

Prior TOBi resolved more than 50% of customer issues at first contact and forwarded almost every second issue to advisers.→Vodafone reported that SuperTOBi resolved 25% more issues at first contact and independently resolved 74% of all inquiries in under ten seconds.

Production status reported on 2026-04-27 after the German generative-AI upgrade. · Approximately 600,000 German customer-service chatbot conversations per month.

Vodafone-reported production figures without disclosed cohort methodology; '25% more' and the move from more than 50% to 74% may not use identical measurement definitions.

What leaders can reuse

Anti-pattern

Combining country- and journey-specific metrics into a global success rate or removing human escalation because headline containment is high.

Questions

  1. 01Which requests may the assistant resolve without human approval?
  2. 02What escalation and customer-recovery path exists for wrong answers?
  3. 03How are quality-review findings converted into controlled system changes?

Portability conditions

  • A bounded, maintained source of approved customer-service knowledge
  • Reliable escalation to live advisers
  • Daily or risk-based human review of generated responses
  • Journey-specific outcome measurement rather than global metric blending

Reputation risk

medium: customer-service chatbot results are company-reported and should remain market- and journey-specific.

Evidence and authority

What the public record supports.

Current · updated

2 independent, 2 primary; publication outcomes are verified and reported.

Bundle 1.0.0 · reviewed 2026-08-23 · stable ID b88b0d637576b1da

Related transformations

More in Customer service →

Ocado Group / Ocado Retail · Exceptions only

On-Grid Robotic Pick at Luton CFC

Robotic arms can pick more grocery items on-grid while people handle exceptions and fulfillment operations.

NHS England stroke networks · Exceptions only

Brainomix e-Stroke AI decision support in the thrombectomy pathway

AI can surface suspected stroke-transfer cases before sequential specialist communication finishes.

Maccabi Healthcare Services · Exceptions only

Prostate Pathology Quality Control

Second-read quality control is too labor-intensive to perform on 100% of cases.

Waymo · Exceptions only

Fully autonomous Waymo One driving

Within a defined domain, automated driving can control the vehicle while remote assistance handles exceptions.

Sources

Read the evidence, freshness, caveat, and version policy.