Intelligent Customer Self Service. Dr. Joachim Stegmann Deutsche Telekom Laboratories
Contents. Motivation. Customer Concern Completion. Customer Insights and Quality Management. Multimodal Self Services. Content and Application Management. Enabling Technologies. Conclusion. Deutsche Telekom Laboratories 20.10.2008 2
Motivation. How to become best-service company? Human communication over the phone can be rather cumbersome! But the real challenge is: How to become best service company at lower cost? What customer wants Service costs How customer tells it to the operator What the customer really needed Best service with reduced budget? How the operator understood How the operator really solved it What the customer really needed Deutsche Telekom Laboratories 20.10.2008 3
Approach. Intelligent interaction for best service and cost control. Customer Concern Completion Customer insights & quality management Segment analysis and user research. Development of use cases. Quality management for process optimization, e.g. voice analytics. Multichannel & multimodal self services Unified processing of contacts by phone, email/sms, web, and mail. Voice self services. Visual communication and multimodal interaction. Content & application management tools Rapid prototyping and maintenance of customer selfservice applications. Evaluation, optimization, and enhancement of existing tools. Automated Application Building MMUI Design Content-to-Speech Enabling modules and solutions Multimodal Interaction Management Natural-Language Processing Audio Mining Semantic Domain SLM Modeling Computer Vision Deutsche Telekom Laboratories 20.10.2008 4
Customer Concern Completion. Avoid lost contacts and maximize immediate service. Example DTKS voice services Service options How options are used Immediately served Asynchronous service Lost Fully automated by IVR + enhanced technology Agent (after IVR classification) Automated Feedback (after VoiceBox) Agent callback (after VoiceBox) Lost Contacts (in dialog or waiting queue) Prepaid Customers Today Short-term goal Long-term goal 0% 25% 50% 75% 100% Loss rate reduces by: Asynchr. service Better technology 20% lost 95% automation Deutsche Telekom Laboratories 20.10.2008 5
Customer Insights & Quality Management. Evaluate use cases and ensure superior user experience. ISO 13407: Human centered design process for interactive systems Improved automation processes Following a user-centered development process to achieve the most appropriate procedure for automation. Meets requirements 1. Plan the Human centered process 2. Specify the context of use Define target group specific use cases. Draft interaction flows for e.g. visual communication. Iteration loops for refinements. 5. Evaluate designs against user requirements 3. Specify user & organizational requirements 4. Produce design solutions Deutsche Telekom Laboratories 20.10.2008 6
Customer Insights & Quality Management. Quality management in contact centers. Targeted at staffed call and contact centers. Automatically monitor and control customer interaction with agents (close to 100% of calls). Analysis of conversations for quality management and marketing purposes Prioritize calls for human monitoring (small percentage of calls). Technical Setup IVR Call statistics KPI Stress/Emotion Customer(s) Agent(s) Recording Transcription Keyword Spotting Data Mining/ Statistics Speech Analytics Performance Indication/ Action Items Two complementary, main drivers for positive business case. Improved efficiency for call center supervisors. Constant monitoring of KPIs, statistics suggesting improvements (cross-checks with marketing). In-domain knowledge and confidentiality needed for successful operation. Deutsche Telekom Laboratories 20.10.2008 7
T-Mobile Voice Portal Customer Insights & Quality Management. Online Statistics Tool. Caller fixed network Caller mobile network SM Contract AGENDER PK Online Statistics Tool Info- Modul EMO PK SM Prepaid LID PK Automatic age, gender, and language identification are performed once per call, emotion detection (anger) for each utterance. After classification, data is transmitted to a statistic server and combined with the call ID. Every 24 hours the results are sent to the Data Warehouse system. Currently, 25,000 calls are analyzed each day. Statistic Server Data Warehouse Evaluation for Customer Care and Marketing purposes Deutsche Telekom Laboratories 20.10.2008 8
Customer Insights & Quality Management. Example of results from Online Statistics Tool. Tariff options Which tariff options are used by the different customer segments? Example Roaming : 100% 80% 60% 40% 20% SENIORS MID F MID M YOUNG F YOUNG M KIDS Monitoring of angry callers Caller satisfaction in voice portal is constantly evaluated with an emotion thermometer. Enables short-term actions in case of decreasing customer satisfaction. Emo-Thermometer T-Mobile Deutschland 28.09.2008 0 % 100 % 0% w orldclass w eltw eit xtraroaming are delighted get relaxed mixed mitigation needed anger Xtraroaming is preferred by KIDS (Prepaid). YOUNG M and MID M use more worldclass (Postpaid). Deutsche Telekom Laboratories 20.10.2008 9
Multimodal Self Services Overview. Multimodality improves attractiveness of customer service. Interactive Voice Response (IVR) turns out to Interactive Media Response. Increase degree of automation and first done rate in customer self care. Enabling Modules Media Response Layer + Video IVR + Multimodality + Pre-qualification Menu-driven functionality support Rich animated media support Intuitive dialog with life-like corporate head + Personalization + Analytics + Automation Voice Only + Self Care + QA System Quick Fix Interactive Service Communication One-to-many configuration Deutsche Telekom Laboratories 20.10.2008 10
Multimodal Self Services: Use Cases. Menu-driven functionality support. Customer heard about NaviGate She wants to use it on her mobile phone. Concern Multimodal Contact Center (MCC) Support Hello How can I use my navigation? Automated Process Installation and usage: Visual setup procedure. Trouble shooting: Clear step-by-step solution. That was easy! Now I would like to find Customer contacts MCC MCC sends solutions Customer resolves Customer Value Chain Get solutions everywhere Time saving Comprehensible Deutsche Telekom Laboratories 20.10.2008 11
Multimodal Self Services: Use Cases. Rich animated media support. Customer wants to setup her new IPTV connection and equipment. Concern I want to set up my IPTV? Multimodal Contact Center (MCC) Support Media clip is provided to customer s mobile phone where agent serves and offers step-by-step visual support. Customer can ask further questions or replay support clip at any stage. Any media-capable device Bridging the technology gap: Support for complex issues solution reproducible re-iterative guidance Deutsche Telekom Laboratories 20.10.2008 12
Content & Application Management Tools. Harmonizing tools towards a common ground. Realization of cost saving potentials by using common content and application management tools. Provide tool chain for complete application lifecycle Application Development Common workbench One content and application management repository Plattform / App 1 Plattform / App 2 Plattform / App 3 Multiple applications and platforms Nowadays Tools-Environment Next-Level Tools-Environment Deutsche Telekom Laboratories 20.10.2008 13
Optimization of technologies First done rate Enabling Modules and Solutions. How to increase service quality? Enhancements of technology Language models for Telekom domain. Combining statistical language models and grammar-based speech recognition. and modeling of domain knowledge Ontologies for Telekom domains. Automated problem solving. increase automation! First in asynchronous services. Later in customer self-service portals. Large-(Open)-domain speech recognition (SLM) Ontology-based self service Immediate Self Service More processes served automatically Async. Service (e.g. VoiceBox) Automated Feedback (Mail, Callback,..) Automation Deutsche Telekom Laboratories 20.10.2008 14
Enabling Modules and Solutions. Increase automatic coverage of customers concerns. Key technology: Large & Open domain Speech Recognition my Speedport v700 update! Closed-Domain ASR Open-Domain ASR my wife says our bill is too much this is bad! Good IVR-based service for a few customers with anticipated needs Using fixed grammars or statistical language model Can be automatically learned (cf. CopperCoin) For well-behaved domains (e.g. Intranet) Good speech-based service for all customers even with unanticipated needs Handling of out-of-domain speech Semi-automatic learning of broader models, even on unstructured domains (e.g. Internet) Semi-automatic tuning to specific domains Deutsche Telekom Laboratories 20.10.2008 15
Multi-Channel Contact Layer Enabling Modules and Solutions. Knowledge-based language engineering. Key technologies: Knowledge based language engineering, Domain Modeling Increasing service quality by introducing service intelligence for DTKS domain. Evaluation of existing systems for content and application modeling and maintenance (off-line processing). Applying natural language processing (on-line processing). User query Linguistic Processing Information Retrieval (un)-structured domain description Domain knowledge (e.g. ontology) System answer Response Generation Information Filtering To be developed. Existing technologies Deutsche Telekom Laboratories 20.10.2008 16
Conclusion. Intelligent customer self service is a solution to provide better service at lower cost. The main objective is to increase the rate of automated customer concern completion based on the following building blocks: Optimize relevant use cases by automatic customer insight generation. Improve usability by multichannel and multimodal self services. Enable rapid prototyping and maintenance with harmonized content & application management tools. The basic technology is available now, however Large & open-domain speech recognition, and Knowledge-based language engineering still have to be improved. Deutsche Telekom Laboratories 20.10.2008 17