A Deep Learning Approach to Automatic Call Routing Rajiv Shah Director of Solution Architect and Professional Services 2019
About GigaSpaces 300+ Direct customers We deliver the fastest big data 50+ / 500+ analytics processing platform to Fortune / Organizations run your analytics & machine learning in production, at scale 5,000+ Large installations in production (OEM) 25+ ISVs
GigaSpaces Select Customers OEMs / ISVs / Partners
ABOUT THE USE CASE This use case shows how to modernize existing software architecture for an efficient call center routing workflow USE CASE BENEFITS: Enhance Customer Experience with automatic routing that prevents customers from being buried in a hierarchical menu Reduce Average Handle Time for optimized efficiency
BUSINESS CHALLENGES Impr prove Custo ustomer r Exp Experi rience Reduce Cos Redu osts ts: : low lower r AHT T Enhanced Syst Enh ystem Agili gility Faster call resolution: Higher agility when adding Faster call routing to the Faster routing new categories or correct agent means a more + satisfied customer departments Routing to correct agent
TECHNICAL CHALLENGES Simplification Continuous ML Training Performance Leveraging existing Continuous model training Event Driven Architecture opensource frameworks such based on previous based on prediction criteria is as BigDL in a unified platform transcribed calls + automatic required for optimal simplifies architectural training of alternative models performance supporting peak complexity ensure models with higher events scoring
Stop Pressing 0 Or * Automatic routing to the right agent for the perfect personalized experience Route to the I have a NLP Processing MAC expert windows MAC problem training, prediction, and tuning Browser InsightEdge event Spark job listens Controller writes BiGDL writes User speaks converts speech processor listens on Kafka topic data to Prediction to using web to text and for Prediction data and using BigDL InsightEdge and InsightEdge and routes call model, creates interface sends to to Kafka topic data grid session prediction controller
Operationalizing AI Example – Automatic Call Routing
LIVE DEMO: Instant Insight to Action • Run Deep Learning with BigDL on transactional data in real-time for instant insights • Trigger transactional workflows based on prediction criteria and scoring for real-time business impact • Simplify architecture, eliminate GPU requirements & reduce component and cluster sprawl for optimal performance & TCO
Automatic Call Routing
RESULTS Per erform rmance Acc ccura racy Con ontinious Tr Trainin ing Background Between Average processing and ~50ms 75%-85% training of ~10 routing accuracy minutes to create a new model
GigaSpaces Coverage
GigaSpaces Competitive Edge SPEED Any Data Live, Transactional & Historical Data Deploy Anywhere ANALYTICS SCALE
Data Analytics: Undeniable Value to your Business Dynamic Pricing Predictive Maintenance Helps grow sales by 30% annually Reduces maintenance costs by up to 75% per mile (transportation example) Optimized Operations Saves $100sK in annual savings Personalized Recommendation (banking example) Increases conversions by up to 20X for brick & mortar stores via location-based promotions Risk Analysis Reduces loan losses by 10-30% Fraud Analytics Reduces losses by 3 to 5% in mature environments and by over 30% in evolving contexts Call Center Automation Increases efficiency by over 90%
The Velocity of Business “A typical e -commerce “To prevent fraud, “A call center receives website will experience anomaly detection 450,0 ,000 ca calls lls/day, each 40% bounce if it loads in needs to happen call needs to be routed in more than 3 s seconds, against 500,000 less than 60 millis illiseconds ” including txn/sec in less than personalization offers” 200 millis illiseconds ” ECOMMERCE TELCO FINANCIAL SERVICES
Use Cases Spanning Industries Benefit from Near Real-time AI Decision Support Systems Built on GigaSpaces • • • Fraud Usage based Personal • insurance Credit risk scoring recommendations • • • Customer 360 Customer 360 Intelligent inventory mgmt. • • • Customer churn Customer churn Customer 360 • • FINANCIAL FI Claims management RETAIL RET IL Locations-based INSUR URANCE SER ERVIC ICES ECO ECOMMERC RCE promotions • • • Predictive maintenance Inventory planning Customer 360 (incl. churn) • • • Intelligent call center routing Customer 360 Fleet management • • Data Center Infrastructure • Predictive maintenance Customer 360 Monitoring (DCIM) • Predictive maintenance INDUS USTRI RIAL MED EDIA IA/ TRANSPORTATIO TRA ION IOT OT TEL TELCO
InsightEdge: Unifying Real-Time Analytics, AI and Transactional Processing in One Platform • Rich ML & DL support • Extreme performance • Fully Transactional Machine Learning • ACID Compliance & Deep Learning • Enterprise-grade (Security, High Availability) In-Memory KEY-VALUE GEO SPATIAL DOCUMENT • Co-located Apps and Services Multi Model Store TABLE COLUMNAR STREAMING • Seamless integration with Big Data Intelligent Multi-tier Storage Management ecosystem • Data sources (Kafka/Nifi/Talend/etc.) STORAGE • Data lakes (S3/Hadoop/etc.) ORCHESTRATION • BI tools (Tableau/Looker/etc.) CLOUD/HYBRID/ ON-PREMISE
Traditional vs. Unified “Translytical” Processing TRANSACTIONAL/ANALYTICAL TRANSACTIONAL/ANALYTICAL TRADITIONAL UNIFIED PROCESSING PROCESSING TRANSACTIONAL PROCESSING TRANSACTIONAL PROCESSING SLOW FAST IN-MEMORY FEEDBACK IMPACTS DATA REPLICATION FEEDBACK DATA GRID LOOP Real-time analytics LOOP Greater situation awareness Simplified architecture ANALYTICS ANALYTICS
UNIFYING Analytics and Transactional Processing at SCALE & SPEED BI TOOLS DATA LAKE DATABASE & DATA WAREHOUSE APPLICATIONS MOBILE WEB IOT ANALYTICS, MACHINE & DEEP LEARNING APPS & MICROSERVICES BI & VISUALIZATION SECURITY AND AUDITING MANAGEMENT AND MONITORING REST ORCHESTRATION EVENT MICROSERVICES DEEP MICROSERVICES EVENT MACHINE SPARK JOBS SQL/JDBC NOTEBOOK STREAMING (REST) PROCESSING PROCESSING LEARNING LEARNING (REST) RPC & MAP/REDUCE CDC Engine CORE CR8 RPC & MAP/REDUCE MemoryXtend - MULTI-TIERED STORAGE MULTI MODEL STORE DATA OBJECTS, JSON, KEY-VALUE, TABLES, TEXT, SSD IN-MEMORY RAM LAKE GEO SPATIAL, GRAPH DATA GRID EVENT PERSISTENT WAN GW - MULTI SITE REPLICATION PROCESSING MEMORY WAN GATEWAY CLUSTER MANAGEMENT & SERVICE DISCOVERY ON-PREMISE HYBRID CLOUD
AnalyticsXtreme: Accelerating Your Data Lake by 100X for Real-time Analytics Your Yo r data is is im immediately se searc rchable le, quer eryable, , and ava vaila lable fo for r analytics • Single logical view for hot, warm and cold data • Hot data resides on in-memory data grid and historical data on HDFS/Object Store • Hot data is muta table le and historical data is immutable le (parquet) Fast Access • Fast access to frequently used historical data STREAMING HOT & WARM DATA DATA Access any data through a unified layer INGEST • COLD & ARCHIVED Analytics (Spark ML) DATA • Query (Spark SQL) Automatic lifecycle management • Automatically handles the underlying data movement, optimization and deletion
Ultra-low latency and high throughput transactional processing IMDG Partitioned In-Memory Grid Shared-nothing, linear scalability, MOBILE WEB IOT elastic capacity Co-Location of Data and Business Logic Co-located ops, event-driven, ANALYTICS & BIG DATA APPS & MICROSERVICES SEARCH, BI & QUERY fast indexing SECURITY AND AUDITING MANAGENENT AND MONITORING MANAGENENT AND MONITORING EVENT MACHINE MICROSERVICES MICROSERVICES EVENT SPARKL SQL .NET SQL/JDBC SEARCH STREAMING JAVA LEARNING (REST) PROCESSING PROCESSING Event-Driven Processing and (REST) Map/Reduce No Downtime Auto-healing, multi-data center RPC & RPC & DATA MODELS WEB CONTAINERS EVENT replication, fault tolerance MAP/REDUCE MAP/REDUCE (SPATIAL, POJO, JSON) PROCESSING IN-MEMORY Fast Indexing Multi-Data Model DATA GRID POJO, .NET, Document/JSON, RAM SSD SPERSISTENT DATA REPLICATION Geospatial, Time-series STORAGE MEMORY & PERSISTENCE Seamless Integration with CLUSTER MANAGEMENT & SERVICE DISCOVERY Java/Scala ecosystem Cloud, Kubernetes, Docker Native ON-PREMISE HYBRID CLOUD
Recommend
More recommend
