Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Project Fido:Big Data, BI and B2B Analytics – Yeah, Let’s do that!
Jacob PellockSr. Director Software Engineering
August 17, 2016
The Question…
We sell sophisticated analytic solutions and services to our customers, but why don’t we use the same technology and techniques to manage our own business?
Big Data, Business Intelligence and B2B Analytics2 #HPCCMeetup
Agenda
1. Pilot
2. Implementation
3. Results
Big Data, Business Intelligence and B2B Analytics3 #HPCCMeetup
In short, the goal of Project Fido is to enable LexisNexis to answer basic business questions within each of our key sources of growth. For the initial pilot, we selected one of these questions around attrition, to test our capabilities
Big Data, Business Intelligence and B2B Analytics4
Growth Discipline
FrameworkBusiness Questions Analysis Type Capability
What products are our customers buying together? Cross sell Model Enhancement
Who are our most valuable customers? How do we keep them? Loyalty Segmentation New
What is the best segmentation of our customers to identify unmet customer needs? Behavior Segmentation New
Which training approaches increase usage and retention? Test vs. Control New
What pricing meets the needs and behaviors of our customers? Pricing Models New
Which customer insights should we provide our customer service reps? Profiling Analysis New
Which customers are most likely to leave and when? Attrition Model New
Which customers do we want back and are most likely to come back? Win back Model New
What is the most relevant messaging for each segment of prospects? Targeting Models Enhancement
Which leads are most likely to become Wins and have high LTV? Lead Scoring Model Enhancement
What is the best customized onboarding experience for fast uptake of products? Onboard Testing Enhancement
Which customers are most likely to buy new products? New Product Analysis Enhancement
Which new LN product will fill a void in the marketplace? White Space Analysis New
Base Business
New Customers
New Initiatives
#HPCCMeetup
What data do we have in-house that can be leveraged for this project?
What data could be used in analyzing our customers?
• Revenue / Volume • Pricing and Contracts
• Surveys (CES, NPS) • Firmographics
• Marketing Campaigns • Training
• Dunning • Credits
• Security Audits • Leads
• Customer Service Calls • Sales Contracts
• Request for Proposals • User Activity
Big Data, Business Intelligence and B2B Analytics5 #HPCCMeetup
…but also clearly highlighted our data constraints that make such efforts unsustainable on a regular basis
Current Environment:
Big Data, Business Intelligence and B2B Analytics6 #HPCCMeetup
Our Fido pilot in Financial Services, which predicted attrition for online products, was successful in demonstrating the power of analytics…
Hypothesis: We can identify customers who are about to attrite and we can prevent them from leaving through a focused Sales Call Blitz
Scope:
1. Financial Services online customers
2.Test linking capabilities via HPCC
3.Create working attrition model and output alerts to Sales for follow-up
Results:
1. The Attrition Model and Sales Call Blitz enabled the sales team to effectively act upon 13% of companies about to attrite
2.Lack of customer engagement and reduced # of active users are key indicators of attrition
Big Data, Business Intelligence and B2B Analytics7 #HPCCMeetup
Key Findings of Current Environment
Key Findings of Current Environment
We can do it but it’s hard. The pilot demonstrated that we have the internal capability to produce a productive model, but the process was manual, resource intensive, and non-repeatable under normal circumstances. To produce the model again would require months of data gathering and customer linking to prepare the data for model building. The model could not be automated for biweekly use by the business.
Internal enterprise data quality is inconsistent. Some data simply does not exist (e.g., historical transactions) while others can vary by business/function. Enterprise data exists in multiple systems and the knowledge and/or rules needed to use the data is not always documented or available, creating knowledge gaps and affecting the quality of the outputs.
Linking customers across systems is difficult. Lack of a unique identifier makes it difficult to link data across internal systems to facilitate analytics.
Lack of data documentation and governance increases potential misuse. Access to data is largely managed amongst a small group of reporting teams. Things like access request, data change requests and quality control are discussed in committees with limited defined authority or approvals, creating environments where data use can be inconsistent or inappropriate. As data usage increases, privacy, compliance and/or reputational risks may become more prevalent.
Efforts are duplicated across multiple teams as we all strive to get smarter. Multiple teams exist across Risk, each using unique tools and systems to manage similar data for analytical and reporting efforts, creating inefficient use of resources, and disparate fiefdoms.
Big Data, Business Intelligence and B2B Analytics8 #HPCCMeetup
Agenda
1. Pilot
2. Implementation
3. Results
Big Data, Business Intelligence and B2B Analytics9 #HPCCMeetup
Based on the pilot key findings, we identified key attributes that need to be provided regardless of the technology solution pursued
Big Data, Business Intelligence and B2B Analytics10
The way forward includes:
1. Central repository for gold-standard data scalable for future expansion
2. Customer data linked across all data sources
3. Data governance applied to manage how data is accessed, used and documented
4. Flexible enough for used by all analytical and reporting users of varying skill type
#HPCCMeetup
Our Solutions Are Powered by HPCC Systems at Their Core
Big Data, Business Intelligence and B2B Analytics11
• Grid computing
• Data-centric language (ECL)
• Integrated delivery system that offers data plus analytics
BigData
StructuredRecords
UnstructuredRecords
NewsArticles
ProprietaryData
PublicRecords
Unstructured and Structured Content High Performance Computing Cluster Platform (HPCC) Analysis Applications Key Capabilities
• Over 4 petabytes of content
• 50 billion records
• 10,000 sources
• 7.5 billion unique name and address combinations
• Multi-bureau/multi-source models and bureau roll-over support
• Extensive experience leveraging atomic level data, combining and leveraging disparate data
• Approximately 400 models deployed (custom and flagship)
• Data and analytics
• Identity verification and authentication
• Fraud detection and prevention
• Investigation
• Screening
• Receivables management
Fusion
Linking
Refinery
Open Source Components
Complex Analysis
Clustering Analysis
Link Analysis
Entity Resolution
Financial Services
Government
Health Care
Insurance
Legal
Retail
Scientific Technical Medical
Exhibitions
Full Complement of Big Data Capabilities All in One Platform
Big Data, Business Intelligence and B2B Analytics12
MPP Platform
Analytics Tools
Common Data Centric
Language
Data Science
Portal
Graph Analytics (KEL)
ETLEntity Disambiguation
(SALT)
Dashboard Creator Data Composition Builder
In Memory Ad-Hoc BI
Machine Learning
ECL
Delta Store Thor ROXIEStreaming
#HPCCMeetup
The Data Warehouse will streamline processes to produce a centralized data warehouse containing data from various areas of the business to enable analytical capabilities…
Big Data, Business Intelligence and B2B Analytics13
BEFORE AFTER
#HPCCMeetup
Obligatory Performance Improvement Timings
Big Data, Business Intelligence and B2B Analytics14
Process HPCC Timing(Minutes)
Legacy Timing(Minutes)
% Improvement
Revenue 1 Load 15 100 85%
Revenue 2 Load 15 40 63%
Batch Logs 10 150 93%
Batch Allocations 75 420 89%
Product Ops Load 2 95 98%
#HPCCMeetup
Monitored Population
Big Data, Business Intelligence and B2B Analytics15
Fin
anci
al S
ervi
ces
Monitored Population
Non-Monitored Population
Revenue not within defined limits Seasonal restrictions Certain Product restrictions New Accounts (<1 year old)
Lik
elih
ood t
o A
ttrite
Alerts
Contr
ol
Top 10% riskiest customers acted upon by Sales
No Alerts(predicted low risk)
High
Low
#HPCCMeetup
Fido Analytics – Sales Feedback Loop
Big Data, Business Intelligence and B2B Analytics16
Ensuring Relevant Alerts: FIDO will suppress an alert for five sequential months if sales has taken action and if the alert is for the exact same criteria/score.
Reports will include: Reason codes Contact info Revenue and usage trends Customer service call volumes Reports structure will allow
sellers to drill down to additional account information
Log calls and provide feedback of alerts to improve model predictability
1. Identification of likely attritors (monthly)
FIDO
2. E-mail alerts, reports, and dashboards
3. Sellers report their activities
Act
ivit
y Tr
acki
ng
Das
hb
oar
ds
Ale
rts
#HPCCMeetup
Agenda
1. Pilot
2. Implementation
3. Results
Big Data, Business Intelligence and B2B Analytics17 #HPCCMeetup
Organizational Changes
• Transformed and repurposed existing reporting and BI Teams
• Centralized development resources across core technologies
• Added Statisticians/Modelers
• A few incremental resources
Big Data, Business Intelligence and B2B Analytics18 #HPCCMeetup
What Worked Well
• Close collaboration between business and technology
• Executive buy-in and sponsorship throughout
• Marketing the Initiative
• Iterative development process to continually improve
• Flexibility – balancing business case with business need
• Cultural Change – thinking about customers differently
Big Data, Business Intelligence and B2B Analytics19 #HPCCMeetup
Challenges
• Determining what belongs in Fido – where to draw the line
• Project prioritization
• Resource constraints
• Measuring Results
• “Selling” the business case
• Model Pivots
Big Data, Business Intelligence and B2B Analytics20 #HPCCMeetup
Useful Links
Big Data, Business Intelligence and B2B Analytics21 #HPCCMeetup
• LexisNexis Open Source HPCC Systems Platform: http://hpccsystems.com
• Online Training: http://learn.lexisnexis.com/hpcc
• The HPCC Systems blog: http://hpccsystems.com/blog
• Other Case Studies: https://hpccsystems.com/resources/case-studies
• Wiki: https://wiki.hpccsystems.com/
• Our GitHub portal: https://github.com/hpcc-systems
• Community Forums: http://hpccsystems.com/bb
Thank you!