Classification and Regression in the Real World: Intelligent Automation Using RPA (Robotic Process Automation) and Machine Learning to Recognize Handwriting and Improve Email Classification, by Jidoka.
MLSEV 2019: 1st edition of the Machine Learning School in Seville, Spain.
5. BigML, Inc 5#MLSEV: Machine Learning & RPA
Jidoka
Welcome to the age of software robots
§ Developed by Novayre, software
development company founded in Seville in
2008.
§ Management team with over 20 years of
experience in IT and Business Consulting.
§ Headquarters in Spain (Sevilla and
Madrid). Offices in Colombia and the UK.
Background
is a software solution that facilitates the
automation of business processes on an
enterprise scale, providing a complete
platform for developing, implementing
and orchestrating robots software.
Internationally acknowledged presence
as the leader in Spanish-speaking
markets in its ‘Intelligent Automation
Continuum’
as Representative Vendor in
its ‘Market Guide for RPA
Software 2017’
as a Major Contender in its
‘Peak Matrix- RPA Technology
Vendor Assessment Study’
6. BigML, Inc 6#MLSEV: Machine Learning & RPA
Jidoka’s growing
partner network
reaches +25 countries
in 4 continents
Partners
Regional alliance with
technology companies
from multinationals
(Big Four) to highly
technology and
consultancy firms.
Clients
We work for global
organizations in a wide
range of industries and
business areas.
Jidoka
A global reference in Enterprise RPA
9. BigML, Inc 9#MLSEV: Machine Learning & RPA
Transition to digital
products and
services
Mobile and cross-
platform business
models, with new
challenges in user
experience and
security.
Rapidly changing
regulations
Strict regulations
constantly changing
and updating.
Digital transformation
New systems and
technologies
Coexistence of legacy
systems with new
solutions and
technologies such as
Blockchain or Fintech.
Operating
efficiency
Need to optimize
front and back office
tasks, reducing
costs and ensuring
quality compliance.
Big Data
New challenges in
managing and taking
full advantage of
generated data.
New challenges
For today businesses
10. BigML, Inc 10#MLSEV: Machine Learning & RPA
New workforce
Digital vs Human Workforce
Any technology that reduces fixed costs, especially staff-related costs, and improves
processes ensuring the quality of the service, becomes a key factor to gain a
competitive edge.
RPA (Robotic Process Automation) technology responds to the need for reducing
human intervention in processes, transferring functions and repetitive tasks to a "digital
workforce" of software robots.
Robot vs Human
Repeatable, High-
Volume and Well-
defined tasks
Skilled, Judgement and
Empathy based
activities
Digital Workforce Human Workforce
11. BigML, Inc 11#MLSEV: Machine Learning & RPA
Automates tasks: It works at the
individual tasks level on the screen, not
being necessary to intervene in the
entire end- to –end process.
Available in weeks or months: No
need to modify applications,
development time is shorter than
SOA/BPM projects.
It adapts to changing scenarios: It
recognizes text and images on the
screen, even if they change appearance
or position, with slight changes if
needed.
Robotic Process Automation
Software Robots
A “software robot” is a program that
uses applications the same way that
people do: using the user interface
(UI), reading and entering data on the
screen whether mouse inputs or
keystrokes.
Jidoka works without interfering with
the existing systems and
applications. RPA uses the existing
windows as if they were “macros”
(non-intrusive).
Software robots reduce manual
intervention so that human agents
can focus on activities that require
cognitive interpretation.
13. BigML, Inc 13#MLSEV: Machine Learning & RPA
Intelligent Automation
Get the best from both worlds
What RPA wins from ML What ML wins from RPA
Analysis of unstructured
documents and
connection between
doing and thinking in an
automated environment.
Connection to any system to
extract information needed
by ML algorithms or to
perform actions after
execution of an algorithm.
Typical tasks
Identify the category of data (classification)
Find natural groupings of data (clustering)
Predict the price of an asset (regression)
Extract information from a document
Opening emails and their attachments
Copy-paste of data
Entering information into systems
Login and logout
14. BigML, Inc 14#MLSEV: Machine Learning & RPA
Intelligent Automation
Application examples
Ticket routing for a technical helpdesk
RPA takes the ticket description and makes a call to the ML prediction service to fetch
the routing information to continue the automation process.
Customer support
ML identifies a customer request, understands the emotions, offers solutions and
triggers backend processes through RPA for quick implementation.
Loan process management
RPA guides the collection of the correct financial information and ML models optimize
loan approval processes and recommend interest rates for those that are approved.
Documentation in an insurance firm
The ML components help the system to analyze the different requirements of
customers and provide RPA with suitable data to generate documents.
16. BigML, Inc 16#MLSEV: Machine Learning & RPA
The Business Scenario
Citations Notices
The process that we need to automate consists of
digitalizing the citation notices issued by the mobility
secretary of a town hall.
§ The mobility secretary is going through the process of
digital transformation.
§ They need to introduce citations information into a
database.
§ The citations contain both printed and handwritten
data.
Transform scanned documents into structured data.
The client
The current
situation
Our goal
17. BigML, Inc 17#MLSEV: Machine Learning & RPA
Characteristics that rings the RPA bell
The Business Scenario
Well defined, repetitive, process
High volume of transactions
Time consuming for humans and error
prone
Suitable for distribution of workload
Read handwritten Numbers
18. BigML, Inc 18#MLSEV: Machine Learning & RPA
Citations Notices
Document to process
OCR extraction of printed data:
§ Name
§ Address
§ City
§ Date
§ ID number
Extraction of handwritten data:
§ Reference Number
19. BigML, Inc 19#MLSEV: Machine Learning & RPA
Powerful allies
§ Integration with Free
Software OCR Engine
Tesseract.
§ Integration with Open
Source Computer Vision
Library Open CV.
§ Integration with the
Machine Learning
platform BigML.
OCR
+
+
Machine
Learning
Citations Notices
20. BigML, Inc 20#MLSEV: Machine Learning & RPA
Citations Notices
Document to process
23. BigML, Inc 23#MLSEV: Machine Learning & RPA
The Business Scenario
Email classification
The customer service department of a large company
receives on a daily basis a very large number of
emails addressed to different departments.
§ Time wasted processing emails and redirecting them
to the right department.
§ Tasks executed manually, processing emails one by
one.
§ Many requests are not being dealt with as quickly as
would be desirable.
Make this whole process more agile and responsive.
The client
The current
situation
Our goal
24. BigML, Inc 24#MLSEV: Machine Learning & RPA
Characteristics that rings the RPA bell
The Business Scenario
Well defined, repetitive, process
High volume of transactions
Time consuming for humans
Suitable for distribution of workload
Unstructured texts
25. BigML, Inc 25#MLSEV: Machine Learning & RPA
Process Workflow
Email classification
Check for
new emails
Start
Open ticket
application
Process
current
email
Predict
department
Save ticket
Close ticket
application
End
New
emails to
process?
New
emails?
Yes
No
No
Yes
26. BigML, Inc 26#MLSEV: Machine Learning & RPA
Email Classification
Solution
Check for
new Emails
Jidoka Robots can communicate with lots of protocols,
including SMTP.
Open Ticket
Application
Jidoka Robots can integrate with the Operating System
(Windows or Linux) to open and close applications.
Process
Current
Email
Jidoka Robots have all the power of Java language at their
disposal.
Close
Ticket
Application
Create new
Ticket
Jidoka Robots can easily automate web applications
through DOM access.
Predict
Department
BigML offers both a Web Application and a powerful and
robust Java API to create, configure and efficiently
manage all their resources.
27. BigML, Inc 27#MLSEV: Machine Learning & RPA
BigML Source File
Source File Uploaded via Java API,
the CSV Source file
includes 5 fields:
§ From
§ To
§ Subject
§ Body
§ Department
q Info
q Administration
q Support
Email Classification
28. BigML, Inc 28#MLSEV: Machine Learning & RPA
BigML Dataset
The dataset suitable to
be processed by BigML
is easily generated.
Dataset
Email Classification
29. BigML, Inc 29#MLSEV: Machine Learning & RPA
Model Generation and Evaluation
BigML Model Generation
§ Predictive Model
Generation (Decision tree).
§ Evaluation and training of
the model to ensure
accuracy.
Email Classification
30. BigML, Inc 30#MLSEV: Machine Learning & RPA
Predicted Results
Email Classification
The BigML predictive model
returns:
§ The predicted department.
§ The probability of the right
result.
Department prediction
Information Support
Administration Unknown
Results with probability < 90%
are discarded and assigned to
an unknown department.
31. BigML, Inc 31#MLSEV: Machine Learning & RPA
Before and After
Email Classification
Before
Automation
The high volume of received emails made the
process of knowing who’s responsible for answering
them tedious and time-consuming, making the
response time too slow.
After
Automation
Results
Reduce manual intervention
We have achieved
Reduce response time
Optimize the process
Classification
Service 24h
Response Time
-35%
The automatic classification of the emails makes
easier and faster the response, avoiding it to be
bouncing from department to department wasting
time in the meanwhile.
34. BigML, Inc 34#MLSEV: Machine Learning & RPA
Robotic Process Automation
The foundation of Intelligent Automation
Robotics (RPA)
Repetitive and well-
defined tasks
Machine learning
Pattern and knowledge-based task
Chatbots
User interaction
Artificial Intelligence
Decision making
1
2
3
4
10-15%
<10%
15-20%
60-70%
You might say that RPA is the arms and legs, and the machine
learning component is the brain of an intelligent automation.