Presentation delivered at Bank AI 2018 in Austin, TX. Banco Popular, the largest bank in Puerto Rico, is on a journey to leverage intelligent automation. The first building block was starting with Robotics Process Automation (RPA), setting up the proper governance and controls. From there, we've incorporated other innovations such as Natural Language Processing. This is what what we've learned, how we did it, what has worked for us, and what didn't.
2. About Banco Popular
• Headquartered in Puerto Rico
– Branches in NY, NJ, FL, USVI and British Virgin Islands
– 216 branches (25% in mainland USA)
• 8,000 employees (~800 in mainland USA)
• $48B in assets
6. 10
1
1
2
2
3
4
5
6
7
20
45
Others
Collections
Commercial Credit
Mortgage
Digital Strategy
Customer Contact
Individual Lending
Retail
Comptroller
Wealth Management
Technology
Operations
Bots
requested
by division
25+ enablement sessions
Multiple requests to become Citizen
Developer
1,100 saving in hours on
average for every bot
Request have been growing
exponentially since launch
RPA Results
5 Citizen Developers
fully dedicated and
growing
4 Weeks on average to
build a bot
• Bot Supervisor evolved to
become a new job
• 7 to 10 bots would need a
dedicated supervisor
8. to the Digital
WorkforceJourney
RPA Factory
Phase 1
Large automations for batch
processing
Initial Stage
Simple Macros and
complicated
programming
9. Bots were successful at large batch process…
However, smaller and more universal tasks
wouldn’t be suitable, yet there were so
inefficient!
Which employees will
be working next
Monday?
Who became
Personal AssistantsBots
1. Verify the work schedules
2. Verify approved leave requests
3. Verify pending approvals
4. Verify trainings
5. Verify substitutions
Our Vision for
simplicity in smaller
tasks:
10. Chatbot
Voice
Which employees will
be working next
Monday?
Bots Interfaces AI Engine
Intention: Attendance Bot
Entities:
• Agent: Employee 2345
• Range: 8/13 to 8/13
Natural Language Processing
15 to 1 minute average
saving time
↑ User experience
↑ Efficiency
1 wave – Informative
2 wave - Transactional
Who became
Personal AssistantsBots
Bots’ Farm
John Doo and Daniel Lee will be
working on 3/13/2018.
Carlos Torres will be on PTO.
11.
12. to the Digital
WorkforceJourney
RPA Factory
RPA + NLP
Phase 1
Large automations for batch
processing
Phase 2
Bots who can understand
commands expressed as natural
requests
Initial Stage
Simple Macros and
complicated
programming
14. Who make
DecisionsBots
Predictions in Collections
Who will pay within 30 and 60
days?
Classification
Boosted Decision Trees
ATM Replenishment
Determining optimal replenishment
Clustering & Regression
K-means & Logistic Regression
Return on Offers
How much a given offer will return?
Regression
Logistic Regression
Offers Acceptance
What customers will accept an offer?
Classification
Neural Network
ML models integrated with RPA
15. to the Digital
WorkforceJourney
RPA Factory
RPA + NLP
RPA + ML
Phase 1
Large automations for batch
processing
Phase 2
Bots who can understand
commands expressed as
natural requests
Phase 3
Bots who can learn from data
and make decisions
Initial Stage
Simple Macros and
complicated
programming
Rule Based Self Learning & Autonomous
16. Takeaways
• Start with quick wins to show
value
• Secure buy-in from your
executives
• Build an RPA CoE
• Create a strong Governance
• Work with your InfoSec and
Auditors from Day 1
• Develop a policy for the Bot’s
• Watch roles segregation
• Correct the process before
automation
• Cloud deployment
• Use Citizen Developer to
empower units and increase your
bot factory
• Integrate with AI to make smarter
bots
17. The Truly Smart Digital Workforce
Natural Language
Decision Making
Image Recognition
Sentiment Analysis
RPA Digital
Workforc
eSenses AI powered
F
Here is a sample of the bots we have automated so you can get an idea of what they look like
The first one is robotito, we name all of our bots, usually with funny names, at least this one is funny in Spanish
You can see at the top the description of the bot, in the initial circle the process took before the automation, then the time after the automation and then the savings in hours annually
Velocity // Challenging the false positive