NOW PLAYING: AI, ML AND FINANCE IN REAL LIFE
Transcript of NOW PLAYING: AI, ML AND FINANCE IN REAL LIFE
NOW PLAYING:
AI, ML AND FINANCE IN REAL LIFE
October 22, 2019
Agenda#1: Churn Prediction—
The AI & FP&A Combo
#2: Transforming the Expense Process
Panel Discussion
Churn Prediction:
The AI & FP&A Combo
Part 1:
AI & CPM: Looking Into the Future
➢ About ServiceMaster & Terminix
➢ About Jedox
➢ Challenge: Churn Prediction
➢ Roadmap of Digitalization
➢ What We Learned
About Us
➢ A leading cloud EPM solution
➢ More than 2,500 customers in 140 countries
➢ 250+ certified business partners globally
➢ Delivers:
➢ Self-service budgeting
➢ Unified planning and forecasting
➢ Reporting & analytics & dashboards
➢ Seamless data integration
➢ Recognized as a leader by independent
analyst firms
➢ Home & commercial pest control
services
➢ Subsidiary of ServiceMaster
➢ Fortune 1000 company
➢ Based in Memphis, Tennessee
➢ Largest brand
➢ ~ 80 % of ServiceMaster revenue
➢ More than $5 billion in sales
➢ Significant impact on recurring revenue & growth
➢ Churn rate is a key driver of customer retention
➢ Implement effective product and pricing strategies
➢ Influence on Financial Planning:
➢ Customer care costs
➢ Sales commissions (new vs. renewal)
➢ Sales & marketing costs
➢ The accuracy of churn is critical for accurate FP&A
Churn Prediction for FP&A
➢ Multiple product offerings
➢ Diverse geographic footprint
(Seasonal & weather effects)
➢ Diverse demographic markets
➢ Massive amounts of data & variety of customer and deal
characteristics
➢ Recording, analyzing and predicting customer churn in timely
manner becomes quite difficult
Challenge: Churn Prediction
The Approach:
Digitalization of FP&A – RoadmapArea Foundation Automation Transformation Digitalization
Data Entry Excel Bottom-Up
+Integration
+Top-Down +AI
Business
Logic
Aggregations Allocations
Driver-Based
+Simulations
Process Workflow +Alerts +Prescriptive
UI Excel, PPT Excel & Web +Mobile
Roadmap of Digitalization at Service Master
Area Foundation Automation Transformation Digitalization
Year 2016 2017 2017-18 2019
Business • Manual work
• Excel-based
• Data extracts
from multiple
sources
• Limited version
control
• Move to the
EPM AI tool
• Data integration
with source
systems
• Centralized data
source
• Reporting
• Business Modeling
(Drivers)
• Faster decision-
making
• Web UI
• Unified planning
• Improved analytics
• AI
• Continuous
improvements
• Maintenance &
learning
• Ongoing training
Organization HQ – Corp FP&A FP&A
& BI and IT
Expand to subsidiaries Functional owners
➢ Solid FP&A is in place for budgeting and forecasting
➢ System already connected to data sources
➢ Internal teams (BI & FPA) well trained on
➢ Data integration
➢ Report creation
➢ System maintenance
➢ AI capabilities came “Out of the Box” with the FP&A platform
➢ Time saving (System selection, learning curve)
Leverage Current FP&A Solution
Mission Statement
Analyze and predict customer churn using data from
ServiceMaster’s FP&A platform and in-house data warehouse.
Provide insights about:
➢ Churn drivers (reasons)
➢ Prediction of customer churn (agreement-level)
AI Working Cycle
Data
Collection
Evaluate
Drivers
Analyze
Results
Expansion
Sample data from a few branches (10%)
~10 features (out of 30) that are relevant to identify “churn pattern”
97.6% Accuracy to predict Churn Yes/No
Full data set
First Step: Churn Prediction
Learning & Configuration
Drivers
Expansion
Data Input
Testing & Model Results
Second Step: Churn Prediction
Run model on all branches
Data from ALL branches
~10 features (same as the first stage)
Accuracy: 95.3% Churn=“Yes” - 98.99% Churn=“No”
Actionable list of high-risk customers (time-based)
Drivers
Expansion
Data Input
Testing & Model Results
Third Step: Churn Prediction
"Near-future" churn
Additional data source; ~3 months after the first snapshot
8 features to predict “next quarter churn” (out of 40)
Accuracy: 70-85% (not enough “trained periods” and data changes)
“Focus list” – For preemptive actions by other departments
Drivers
Expansion
Data Input
Testing & Model Results
Next Steps
Corrective or proactive actions➢ Share information with the branches
➢ Measure success
Improve model to support increased prediction accuracy ➢ Frequent data loads (i.e. every 3 months)
➢ Add additional data sources → more features (focus on data that
changes over time)
➢ Share information with the branches
➢ Measure success
Challenges
➢ Define an accurate business question
➢ Skillset
➢ Collaboration with other departments
➢ Training & maintaining the model
The impact to the organization
➢ Better financial forecast → planning
➢ Create an inside team to transform business activities
Change to business and capabilities
➢ Preliminary results shows improvement in retention (goal: 5% decrease in churn)
What did we learn from the AI process?
➢ Find a small use case with discernable business impact (Forecast, driver analysis, data cleansing)
➢ Prepare known data (Additional data can be added later)
➢ Look for an available technology - Using a known, trusted, in-house tool saves implementation, selection time and resources
➢ Create a prototype
➢ Remember, AI is a learning process more than a project
Back at the office:
Summary & Recommendations
Part 2:
Transforming the Expense Process
Naveen Singh
CEO, Center
Rahim Shakoor
Controller, Docker
Transforming the Expense Process
Agenda:
• Problem
• Business Process
• Impact
• Back in the Office
State of Docker 2017: Challenges
60% Y-o-Y
employee growth
Robust travel
budget
Minimal policy
& analytics
The Status Quo Trajectory
2016 2017 2018 2019
Cumulative Change Relative to 2016
Processing Time
Expense Software Fees
Travel Costs
What takes so much time?
• Tracking down expenses
• Checking for coding
• Manual auditing
• Processing accruals
Docker Roadmap
Identify current
workflow issues
2017 2018 2019 2020
“Automation stops at
the back office.”
The Evolution of Expense Management
Foundation Digitization Automation Transformation
Spreadsheet with
physical receipts
Expense report +
receipt capture
Approval routing &
tracking
Excel Mobile phones Cloud
Better
outcomes
AI/ML
The Evolution of Expense Management
Foundation Digitization Automation Transformation
Spreadsheet with
physical receipts
Expense report +
receipt capture
Approval routing &
tracking
Excel Mobile phones Cloud
Better
outcomes
AI/ML
Docker Roadmap
Start initial
pilot
2017 2018 2019 2020
Respond to
user feedback
Predict expense type to increase accuracy
Data Input Card transactions
Model Flow of data from transactions to GL
Train/Test Employee use + finance review
Results: reduced review time
Expand Data Set Deploy broadly
Maintenance Periodic re-training and testing
Before:
After
Roadmap
2017 2018 2019 2020
Focus on
easy wins
Initial Results
2016 2017 2018 2019
Cumulative Change Relative to 2016
Processing Time
Expense Software Fees
Travel Costs
What’s Next?
Pay Process Audit Report Optimize
Real-time data AI and ML Analytics
Streamlined workflow
Better
decisions=
What’s Next?
Pay Process Audit Report Optimize
Real-time data AI and ML Analytics
Streamlined workflow
Better
decisions=
Highlighting Insights
Key Takeaways
• Next generation technology bring
processing costs down significantly
• AI and ML automate and analyze real-time
data to drive better outcomes
Back In The
Office:
• Get started on the journey from foundation to transformation
• Consider how to shift from a system of record approach to a system of intelligence
• Measure and understand your current process
• Research how your current technology providers are innovating around AI to streamline processes
Panel
Discussion Jamie Cousin
Liran Edelist
Rahim Shakoor,
Naveen Singh