SlideShare una empresa de Scribd logo
1 de 30
Descargar para leer sin conexión
INDUSTRIALIZING MACHINE LEARNING IN PHARMA:
Challenges, Use Cases, & ROI
Presenters: Daniel Kinney (Janssen), Ashish Sharma (Axtria), Kaiwen Zhong (Axtria)
1
AGENDA
1 Journey from Traditional Analytics to Industrialized Machine Learning
2 What Does It Take to Industrialize AI / ML?
3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”?
4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning
2
1 Journey from Traditional Analytics to Industrialized Machine Learning
AI / ML Beyond Pharma
3
Maturity of ML / AI in Other Industries
BANKING INSURANCE HCPGOVTRETAIL
MANUFACT-
URING
NATURAL
RESOURCES
UTILITIES
37% 48% 38% 26% 44% 40% 35% 33%
% of industry
ready to
deploy or
already
deployed
large scale
AI / ML
In 12 months
Top
Use Cases:
Fraud analysis on
transactional data
Chatbots
Market / Customer
Segmentation
Process
automation
Computer-assisted
diagnostics
Text / Opinion
Mining
Source: Gartner (March 2019)
4
Applications of ML / AI in Other Industries
BANKING INSURANCE HCPGOVTRETAIL
MANUFACT-
URING
NATURAL
RESOURCES
UTILITIES
Automation
of
Intelligence
Process
Planning &
Optimization
Real time
monitoring
Public health
surveillance
Disaster
planning
Fraud
detection
Demand
forecasting
Cyber attack
detection
Personalized
medicine
Predictive
optimization
Fault
analysis
Forecasting
Population
health mgmt
Demand
forecasting
Lifestyle
segmentation
Credit ratings
Common
Theme:
1. Enable additional and more accurate recommendations predictions
2. Provide timely insights
3. Automate repeatable steps to reduce SME costs and human errors
Source: Gartner (March 2019)
5
High-speed
trading
Appointment
scheduling
Claims
data entry
Machine
scheduling
Autonomous
drill bits
Autonomous
monitoring
E-service
Engineering
assistants
Scanning /
mapping
Product
design
AGENDA
1 Journey from Traditional Analytics to Industrialized Machine Learning
2 What Does It Take to Industrialize AI / ML?
3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”?
4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning
6
PMSA AI / ML Topic Clustering
Marketing (5)
HCP Targeting (13)
Patient Support (1)
NLP (7)
Provider (2)
Adverse Events (1)
New Patient ID (4)
Brand Performance Analyzer: Identifying Opportunities and
Commercial Performance Optimizer
Using AI / ML for More Insightful Pipeline Marketing Decision-Making
Automated Detection of Adverse Drug Reactions Using Social Media Text Data
A New Generation of Patient Support Programs
Application of Predictive Modeling to Define
HCP Potential for Prophylactic Treatment
Delighting Customers w/ Automated Content Personalization
Dynamically Improving NLP & Theme Discovery
Through Contextual Linguistics for Pharma
A ML Approach to Customer Acquisition
Discover latent Patient Insights Using Text Mining & ML
NLP to Understand HCP
Application of NLP & PCA in Life Sciences & Healthcare
ML Techniques Deliver Granular Insights to enable Improved HCP Marketing
Leveraging Event Pathways to Predict Disease
Progression in Contrast to ML Classification Algorithms
Applications of ML in Oncology Analytics: Combining Secondary & APLD
Advanced and Primary Analytics
A Novel Patient-Physician ML Approach on Diagnosing Rare Diseases and
Informing Commercial Strategies
Leverage Informed HCP Migration to Optimize Targeting & Multi-Channel Spend
ML Against the Provider Graph Patient Co-Pay Card Optimization
Use of ML and Patient Level Clinical Encounter Cues to Maximize Values of Sales
Force Activities
An All-Encompassing ML Attribution Model to Plan, Predict and Measure Impact
at the Channel and HCP Level
Unstructured to Structured Data: Can ML Help Make Sense of EMR/EHR Data?Strengthening Patient Journey with Emotional Insights via
Social Analytics
A Novel High-Dimensional Hybrid ML Approach Identifying
High-Value Rare Disease Specialists
Optimizing Physician Targeting to Find Undiagnosed Patients:
An Application of Advanced ML Methods to Hep C
ML to Better ID Physician Targets for New Product Launches
Enhance Accuracy and Effectiveness Of Physician Alert
7
Ecosystem Readiness for AI / ML in Pharma
DIGITAL LEADERS
AWS Machine Learning, SageMaker,
Comprehend, Rekognition
Verify, DeepMind Health,
Tesseract, Google Brain
ML Studio
AWS Machine Learning, SageMaker,
Comprehend, Rekognition
Verify, DeepMind Health,
Tesseract, Google Brain
ML Studio
REGULATIONS: FDA
12 ML / AI-based
algorithms approved, e.g.,
DreaMed supports diabetes
treatment decision
Viz.ai analyzes CT results and highlights
patients most likely to experience a stroke
12 ML / AI-based
algorithms approved, e.g.,
DreaMed supports diabetes
treatment decision
Viz.ai analyzes CT results and highlights
patients most likely to experience a stroke
GAME CHANGING TECHNOLOGIES
voted by CIOs
0% 10% 20% 30%
Data Analytics
AI / ML
Digital Transformation
Information Tech
IoT
CRM
Source: Gartner (March 2019)
8
PLATFORM
CONFIGURATION
DATA
COLLECTION
FEATURE
EXTRACTION
DATA
VALIDATION MACHINE RESOURCE
MANAGEMENT
ANALYSIS /
VISUALIZATIONPROCESS
MANAGEMENT
TOOLS
CONSUMPTION
INFRASTRUCTURE
MONITORING
ML
ALGORITHMS
PLATFORM
CONFIGURATION
DATA
COLLECTION
FEATURE
EXTRACTION
DATA
VALIDATION MACHINE RESOURCE
MANAGEMENT
ANALYSIS /
VISUALIZATIONPROCESS
MANAGEMENT
TOOLS
CONSUMPTION
INFRASTRUCTURE
MONITORING
ML
ALGORITHMS
De-Coding Industrialization of AI / ML
Most focusLeast focus
9
AGENDA
1 Journey from Traditional Analytics to Industrialized Machine Learning
2 What Does It Take to Industrialize AI / ML?
3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”?
4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning
10
Challenges Industrializing Machine Learning
 Downplaying the Role of Iterative Automation
 Data Readiness
 Adoption Barrier
11
3 Myths about Machine Learning
 Inability to show ROI if ML underperforms with poor data and short timeframe
 Inefficient one-off analysis with unnecessarily complicated ML algorithms
1. “BLACK BOX”
Most use cases do not require or are not fit for “black box”
algorithms like neural network.
2.IMMEDIATE
IMPROVEMENT
3. DISRUPTING
TODAY’S PROCESSES
Improvement of processes depends on data availability,
quality, and takes iterations.
Machine learning models do not need to be developed from
scratch. It can incorporate business rules / sit on top of rule-
based, knowledge-based systems.
HOWEVER
HOWEVER
HOWEVER
Consequences
12
 Increase use of publicly available data, which can
be labeled quickly and cheaply by online resources.
 ML (reinforcement learning) can also be used to
enrich training data.
It is daunting and time-
consuming to create large
amount of training data.
Data scientists & analysts
spend 80% time cleaning and
prepping data.
Data Readiness – “Garbage In, Garbage Out”
 Organizations loses competitive advantage as they are afraid to adopt machine learning and advanced analytics,.
 Underbudget for data and automation aspect of machine learning, leading to low ROI.
“DATA JANITORIAL WORK”
CREATE TRAINING DATA
SOLUTION
SOLUTION
Consequences
 Cloud & big data tech are
increasingly available for the
entire ML ecosystem.
 Data strategy, governance,
and management
13
Ability to Impact Decisions with ML Insights
 Inability to impact decision making eventually impact ROI of machine learning, making sustainable development
of machine learning in organization difficult.
FIELD ADOPTION
 Show relationship between increased sales and adoption.
 Tie adoption of insights into incentive compensation.
NON-PERSONAL
MARKETING
OTHER STRATEGIC
DECISIONS
 Use tools to automatically track target responses to create training data.
 Improve marketing vendor management with clear tracking system.
 Identify pain points: financial, productivity, processes, support.
 Avoid “technology for its own sake”.
SUGGESTION
SUGGESTION
SUGGESTION
Consequences
14
SUCCESSFUL INDUSTRIALIZATION OF ML:
Scraping Web Data for
Multiple Vaccines University
Requirements on Multiple Vaccines
Automated Predictive Targeting
with Real World Data, Payer
Insights, & HCP Data
15
Case 1: Predictive Targeting (1/4)
We want to identify opportunities where machine learning can be used to leverage real world
data for dynamic targeting across promotional channels
PROBLEM STATEMENT
OPPORTUNITY
IDENTIFICATION
• Call Targeting
• Call Frequency
• Product Priority
• Messaging
Optimization
• Segmentation
PERSONAL
PROMOTIONS
NON PERSONAL
PROMOTIONS
• Email
Targeting
• Media Buys
• Direct Mail
PEER TO PEER
• Speaker
Program
Attendee
Prospects
16
Actionable,
frequent
insights as
prediction is
precise and
specific
Case 1: Predictive Targeting (2/4)
TRADITIONAL
APPROACH
MACHINE
LEARNING
AUGMENTATION
Aggregated Physician-
level Rx Data
Longitudinal Patient-
level Data (claims)
Payer & HCP
Features
Rx / Sales
Volume
Physician Prescription
Propensity
Predicted Biologic
Initiation
ML / AI FOR AUTOMATIONDATA READINESS
BUSINESS IMPACT
17
Case 1: Predictive Targeting (3/4)
Scale Up & Productionize
Pilot Changes to Test & Learn
Feasibility & PoC1
2
3
 Assess data availability
 Data acquisition & feature engineering
 Minimal viable product & integration
 Conduct analysis to demonstrate POC
 Implement & continuously improve
using test/control approach
 Continuous monitoring of
performance
 Productionize data
management/analytics
processes
 Optimize data infrastructure
Problem & Opportunity
Identification/Ideation
18
Case 1: Predictive Targeting (4/4)
FEASIBILITY & PROOF OF CONCEPT PILOT CHANGES TO TEST & LEARN SCALE & SYSTEMIZE CHANGES
PROCESS
Assess data availability and conduct
analysis to demonstrate proof of
concept
Implement & continuously improve
using test/control approach
Productionize data management/
analytics processes & optimize data
infrastructure
MODEL
Baseline model to predict patient
demand to proof out concept
Improved model with machine
learning for improved performance
Continuous tuning & evaluation of
model performance
DATA
Minimal viable product starting with
claims data
Incorporated additional HCP & Payer
features into model
Formalize data management support
model & governance
INFRASTRUCTURE Built locally leveraging Python
Started buildout in
analytics workbench
Operationalize & automate with
integration between analytics
workbench & source systems (CRM,
reporting, etc)
CHANGE
MANAGEMENT
Sales & Marketing leadership alignment
on potential value
Piloted with test/control
districts. Trained pilot users
Scale up to entire salesforce and
declare this is ‘business as usual’ way
of working
1 2 3
19
Case 2: Web Scraping for Targeted Marketing (1/5)
• Gather information from 10,000+ global university (English) website domains to see if
they mandate specific vaccinations for students / faculties.
• How to identify large amount of new marketing opportunities for a launch product?
• How to make data available for this project?
• How to extract useful information from the unstructured web data?
PAIN POINTS
PROBLEM
STATEMENT
20
Case 2: Web Scraping for Targeted Marketing (2/5)
PLATFORM
CONFIGURATION
DATA
COLLECTION
FEATURE
EXTRACTION
DATA
VALIDATION MACHINE RESOURCE
MANAGEMENT
ANALYSIS /
VISUALIZATIONPROCESS
MANAGEMENT
TOOLS
CONSUMPTION
INFRASTRUCTURE
MONITORING
ML
ALGORITHMS
SCALABILITY ITERATION ML ECOSYSTEM
ML is scaled to gather information from
1 vaccine from 200 domains,
to
1 vaccine from 10,000+ domains,
to
Multiple products in the department
from 10,000+ domains
 ML algorithms are built on top of
rules-based system (iteration 0)
 Time, resources, and data source
domains are managed to run
multiple iterations, with results
improving over time.
All components of industrialized ML
ecosystem were considered at the beginning
of the project to enable
 Automation,
 Efficient data gathering & analysis
 Consumption of information &
decision making
21
Case 2: Web Scraping for Targeted Marketing (3/5)
University Domain:
http://www.Princeton.edu
http://www.Harvard.edu
http://www.Columbia.edu
http://www.Yale.edu
http://www.Upenn.edu
http://www.Dartmouth.edu
http://www.Brown.edu
http://www.Cornell.edu
…
https://www.hbs.edu/Pages/default.aspx
https://alumni.harvard.edu/community/volunteer
https://huhs.harvard.edu/sites/default/files/Immunization_pac
ket_C2023_2019.docx.pdf
https://alumni.harvard.edu/programs-events/wtyc
https://alumni.harvard.edu/travel/mode-of-travel/50-years-of-
victory
https://alumni.harvard.edu/programs-events/harvard-alumni-
online-learning
…
Domain Sub URLs: Domain Content:
Illustration
22
Case 2: Web Scraping for Targeted Marketing (4/5)
• Data
Gathering /
validation of
10,000+
university
domains
SOURCES
• Set up data
structures in
AWS
• Connect with
scraping-related
APIs and
libraries
CONFIGURE SERVER
• Search all sub
URLs of main
university
domains
• Use existing
and build
custom web-
crawlers using
BS4, Requests,
etc.
SEARCH & CLEAN
DATA READINESSDATA READINESS AI / ML FOR AUTOMATIONAI / ML FOR AUTOMATION
1. FILTER URLS 2. ID REQUIREMENTS
Categorize URLs related
to vs. not related to
“student health” based
on URL and main
contents.
Categorize university
websites that require vs.
not require client’s
vaccines based on
website text contents.
… with the following NLP process:
A. Keyword based categorization
B. Create thesaurus for topics
C. Iterative ML model building & evaluation
BUSINESS IMPACTBUSINESS IMPACT
ANALYZE & VISUALIZE RESULTS
DETERMINE
MARKETING STRATEGY
Different commercial strategies for
universities that
1) Requires the vaccines
2) Recommend but not require
the vaccines
3) Mention but not require the
related disease
4) Neither require nor mention
the vaccines or the related
disease
23
Case 2: Web Scraping for Targeted Marketing (5/5)
STEPS
1.
FILTER
URLS
2.
IDENTIFY
REQUIREMENT
Keyword-based 3 5
Traditional ML Algorithms 2 4
BoW + Deep Learning 5 1
Hybrid with Traditional ML 1 (81%) 3
Hybrid with Deep Learning 4 1 (86%)
Ranking Classification Accuracy
on Test Data
Methods
Keyword-based:
Rules-based algorithm using keywords related to “student
health”, and particular vaccine
Traditional ML Algorithms:
Feature engineering for text processing,
Algorithms include Random forest, SVM, and Naïve Bayes
Deep Learning: Bag-of-words for text processing
Hybrid with Traditional ML:
Combining the results from keyword-based model with
results from the best performer of traditional ML algorithms
Hybrid with Deep Learning:
Combining the results from keyword-based model with
results from deep learning.
AGENDA
1 Journey from Traditional Analytics to Industrialized Machine Learning
2 What Does It Take to Industrialize AI / ML?
3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”?
4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning
25
Calculating AI / ML ROI: Why is it difficult?
 Both short term and long term values can be
spread across multiple parts of the organizations.
 It is hard to isolate the contribution of AI / ML in
improvements, especially in large business
outcomes.
26
Iterations to Reach Agreed Upon Target
Time
Error rate
Average error rate
by human specialists
Current error rate
Iteration 0
Target
error rate
Illustrative example:
Image classification error over time
Iteration 1
Iteration 2
Iteration N
…
Improved
accuracy
Reduced human
specialists costs
and errors
27
Considerations for Calculating AI / ML ROI
REVENUECOSTDELIVERY
SPEED
TEAM
EFFICIENCY
COMPETITIVE
ADVANTAGE
BRAND
EQUITY
DIGITAL
MATURITY
More
Tangible
Less
Tangible
Example:
Predict patient
diagnosis accurately
and target
marketing to HCPs
Example:
More productively
utilize highly paid
professionals & sales
reps
Example:
Provide rep NBAs
real time / daily instead
of monthly / quarterly
Example:
Extract more adverse
events per day than when
done by all manually
Example:
Identify un-
discovered side
effects faster
COMPLIANCE
RISK
Example:
Provide cost
efficient and
effective patient
support programs
Example:
Quickly react to positive &
negative media & online
coverage on the brand
Example:
Conduct targeted
and timely digital
campaigns and non-
personal detailing
28
Key Takeaway
You / your team does not need
to be machine learning unicorn,
But your organization should
combine three key aspects of
machine learning: data science,
business analytics, and data
engineering
For high ROI and sustainable
growth of machine learning.
MACHINE
LEARNING
Business Analytics
 Ensure ML model is set up to answer
important business question
 Communicate results to realize
impact of ML
 AND CHANGE MGMT
Data Science
 Develop and adjust ML models
 Validate and continuously monitor
model performance
Data Engineering
 Set up infrastructure to enable data
combination, and model run
 Scale models developed by data
scientists in big data environment
PLATFORM
CONFIGURATION
DATA
COLLECTION
FEATURE
EXTRACTION
DATA
VALIDATION
MACHINE
RESOURCE MGMT
CONSUMPTION
INFRASTRUCTURE
MONITORING
ML
ALGORITHMS
ANALYSIS /
VISUALIZATION
PROCESS
MGMT TOOLS
29
INDUSTRIALIZING MACHINE LEARNING IN PHARMA:
Challenges, Use Cases, & ROI
Presenters: Danny Kinney (Janssen), Ashish Sharma (Axtria), Kaiwen Zhong (Axtria)
30

Más contenido relacionado

La actualidad más candente

Medical Assistant Design during this Pandemic Like Covid-19
Medical Assistant Design during this Pandemic Like Covid-19Medical Assistant Design during this Pandemic Like Covid-19
Medical Assistant Design during this Pandemic Like Covid-19
AI Publications
 
Forrester big data_predictive_analytics
Forrester big data_predictive_analyticsForrester big data_predictive_analytics
Forrester big data_predictive_analytics
Shyam Sarkar
 
Qais_Yahya_ Hatim_Resume_Q_2016
Qais_Yahya_ Hatim_Resume_Q_2016Qais_Yahya_ Hatim_Resume_Q_2016
Qais_Yahya_ Hatim_Resume_Q_2016
Qais Hatim
 
Amr assignment goutam shit - roll 11
Amr assignment   goutam shit - roll 11Amr assignment   goutam shit - roll 11
Amr assignment goutam shit - roll 11
Sourav Biswas
 

La actualidad más candente (20)

Credit card fraud detection through machine learning
Credit card fraud detection through machine learningCredit card fraud detection through machine learning
Credit card fraud detection through machine learning
 
An efficient feature selection algorithm for health care data analysis
An efficient feature selection algorithm for health care data analysisAn efficient feature selection algorithm for health care data analysis
An efficient feature selection algorithm for health care data analysis
 
IRJET- GPS based Medicine Informator
IRJET-  	  GPS based Medicine InformatorIRJET-  	  GPS based Medicine Informator
IRJET- GPS based Medicine Informator
 
ForresterPredictiveWave
ForresterPredictiveWaveForresterPredictiveWave
ForresterPredictiveWave
 
Fraud detection ML
Fraud detection MLFraud detection ML
Fraud detection ML
 
Anomaly detection- Credit Card Fraud Detection
Anomaly detection- Credit Card Fraud DetectionAnomaly detection- Credit Card Fraud Detection
Anomaly detection- Credit Card Fraud Detection
 
Machine Learning for Business - Eight Best Practices for Getting Started
Machine Learning for Business - Eight Best Practices for Getting StartedMachine Learning for Business - Eight Best Practices for Getting Started
Machine Learning for Business - Eight Best Practices for Getting Started
 
Fault detection of imbalanced data using incremental clustering
Fault detection of imbalanced data using incremental clusteringFault detection of imbalanced data using incremental clustering
Fault detection of imbalanced data using incremental clustering
 
Medical Assistant Design during this Pandemic Like Covid-19
Medical Assistant Design during this Pandemic Like Covid-19Medical Assistant Design during this Pandemic Like Covid-19
Medical Assistant Design during this Pandemic Like Covid-19
 
Statistics And Probability Tutorial | Statistics And Probability for Data Sci...
Statistics And Probability Tutorial | Statistics And Probability for Data Sci...Statistics And Probability Tutorial | Statistics And Probability for Data Sci...
Statistics And Probability Tutorial | Statistics And Probability for Data Sci...
 
Improving Credit Card Fraud Detection: Using Machine Learning to Profile and ...
Improving Credit Card Fraud Detection: Using Machine Learning to Profile and ...Improving Credit Card Fraud Detection: Using Machine Learning to Profile and ...
Improving Credit Card Fraud Detection: Using Machine Learning to Profile and ...
 
Forrester big data_predictive_analytics
Forrester big data_predictive_analyticsForrester big data_predictive_analytics
Forrester big data_predictive_analytics
 
Credit Card Fraud Detection Using Unsupervised Machine Learning Algorithms
Credit Card Fraud Detection Using Unsupervised Machine Learning AlgorithmsCredit Card Fraud Detection Using Unsupervised Machine Learning Algorithms
Credit Card Fraud Detection Using Unsupervised Machine Learning Algorithms
 
Credit card fraud detection using machine learning Algorithms
Credit card fraud detection using machine learning AlgorithmsCredit card fraud detection using machine learning Algorithms
Credit card fraud detection using machine learning Algorithms
 
Credit card fraud detection
Credit card fraud detectionCredit card fraud detection
Credit card fraud detection
 
Qais_Yahya_ Hatim_Resume_Q_2016
Qais_Yahya_ Hatim_Resume_Q_2016Qais_Yahya_ Hatim_Resume_Q_2016
Qais_Yahya_ Hatim_Resume_Q_2016
 
IRJET - Employee Performance Prediction System using Data Mining
IRJET - Employee Performance Prediction System using Data MiningIRJET - Employee Performance Prediction System using Data Mining
IRJET - Employee Performance Prediction System using Data Mining
 
PROVIDING A METHOD FOR DETERMINING THE INDEX OF CUSTOMER CHURN IN INDUSTRY
PROVIDING A METHOD FOR DETERMINING THE INDEX OF CUSTOMER CHURN IN INDUSTRYPROVIDING A METHOD FOR DETERMINING THE INDEX OF CUSTOMER CHURN IN INDUSTRY
PROVIDING A METHOD FOR DETERMINING THE INDEX OF CUSTOMER CHURN IN INDUSTRY
 
Amr assignment goutam shit - roll 11
Amr assignment   goutam shit - roll 11Amr assignment   goutam shit - roll 11
Amr assignment goutam shit - roll 11
 
Risk mgmt-analysis-wp-326822
Risk mgmt-analysis-wp-326822Risk mgmt-analysis-wp-326822
Risk mgmt-analysis-wp-326822
 

Similar a INDUSTRIALIZING INDUSTRIALIZING MACHINE LEARNING IN PHARMA: Challenges, Use Cases, & ROI

Driving Results through Strategic Data Sourcing and Optimization: Life Line G...
Driving Results through Strategic Data Sourcing and Optimization: Life Line G...Driving Results through Strategic Data Sourcing and Optimization: Life Line G...
Driving Results through Strategic Data Sourcing and Optimization: Life Line G...
Vivastream
 
How AI and ML Can Optimize the Supply Chain.pdf
How AI and ML Can Optimize the Supply Chain.pdfHow AI and ML Can Optimize the Supply Chain.pdf
How AI and ML Can Optimize the Supply Chain.pdf
Global Sources
 

Similar a INDUSTRIALIZING INDUSTRIALIZING MACHINE LEARNING IN PHARMA: Challenges, Use Cases, & ROI (20)

Big data symposiam.pptx
Big data symposiam.pptxBig data symposiam.pptx
Big data symposiam.pptx
 
AI IN PREDICTIVE ANALYTICS: TRANSFORMING DATA INTO FORESIGHT
AI IN PREDICTIVE ANALYTICS: TRANSFORMING DATA INTO FORESIGHTAI IN PREDICTIVE ANALYTICS: TRANSFORMING DATA INTO FORESIGHT
AI IN PREDICTIVE ANALYTICS: TRANSFORMING DATA INTO FORESIGHT
 
855 sponsor gazdak_using our laptop
855 sponsor gazdak_using our laptop855 sponsor gazdak_using our laptop
855 sponsor gazdak_using our laptop
 
940 sponsor gazdak_using our laptop
940 sponsor gazdak_using our laptop940 sponsor gazdak_using our laptop
940 sponsor gazdak_using our laptop
 
"Healthcare Data for Healthy CRM & PRM" TrendLab Webinar
"Healthcare Data for Healthy CRM & PRM" TrendLab Webinar"Healthcare Data for Healthy CRM & PRM" TrendLab Webinar
"Healthcare Data for Healthy CRM & PRM" TrendLab Webinar
 
artificial intelligence in Pharmacy field.pptx
artificial intelligence in Pharmacy field.pptxartificial intelligence in Pharmacy field.pptx
artificial intelligence in Pharmacy field.pptx
 
How Pharma Can Fully Digitize Interactions with Healthcare Professionals
How Pharma Can Fully Digitize Interactions with Healthcare ProfessionalsHow Pharma Can Fully Digitize Interactions with Healthcare Professionals
How Pharma Can Fully Digitize Interactions with Healthcare Professionals
 
10 Amazing Benefits of Machine Learning You Should Be Aware Of!
10 Amazing Benefits of Machine Learning You Should Be Aware Of!10 Amazing Benefits of Machine Learning You Should Be Aware Of!
10 Amazing Benefits of Machine Learning You Should Be Aware Of!
 
Driving Results through Strategic Data Sourcing and Optimization: Life Line G...
Driving Results through Strategic Data Sourcing and Optimization: Life Line G...Driving Results through Strategic Data Sourcing and Optimization: Life Line G...
Driving Results through Strategic Data Sourcing and Optimization: Life Line G...
 
Mit tech review_machinelearning
Mit tech review_machinelearningMit tech review_machinelearning
Mit tech review_machinelearning
 
Digital & Beyond
Digital & BeyondDigital & Beyond
Digital & Beyond
 
20-Minute Channel Bytes: 5 Tips for Gaining Traction in the Healthcare IT Market
20-Minute Channel Bytes: 5 Tips for Gaining Traction in the Healthcare IT Market20-Minute Channel Bytes: 5 Tips for Gaining Traction in the Healthcare IT Market
20-Minute Channel Bytes: 5 Tips for Gaining Traction in the Healthcare IT Market
 
How An AI-Powered Trade Promotion Optimization Software Can Improve Consumer ...
How An AI-Powered Trade Promotion Optimization Software Can Improve Consumer ...How An AI-Powered Trade Promotion Optimization Software Can Improve Consumer ...
How An AI-Powered Trade Promotion Optimization Software Can Improve Consumer ...
 
Empowering the financial institutions with machine learning
Empowering the financial institutions with machine learningEmpowering the financial institutions with machine learning
Empowering the financial institutions with machine learning
 
Ibm 1901 2019_marktr
Ibm 1901 2019_marktrIbm 1901 2019_marktr
Ibm 1901 2019_marktr
 
Data trends redefine leading brands
Data trends redefine leading brandsData trends redefine leading brands
Data trends redefine leading brands
 
ClickZ/Fospha: The State of Marketing Measurement, Attribution, and Data Mana...
ClickZ/Fospha: The State of Marketing Measurement, Attribution, and Data Mana...ClickZ/Fospha: The State of Marketing Measurement, Attribution, and Data Mana...
ClickZ/Fospha: The State of Marketing Measurement, Attribution, and Data Mana...
 
Advanced analytics playing a vital role for health insurers
Advanced analytics playing a vital role for health insurersAdvanced analytics playing a vital role for health insurers
Advanced analytics playing a vital role for health insurers
 
How Pharma Can Fully Digitize Interactions with Healthcare Professionals
How Pharma Can Fully Digitize Interactions with Healthcare ProfessionalsHow Pharma Can Fully Digitize Interactions with Healthcare Professionals
How Pharma Can Fully Digitize Interactions with Healthcare Professionals
 
How AI and ML Can Optimize the Supply Chain.pdf
How AI and ML Can Optimize the Supply Chain.pdfHow AI and ML Can Optimize the Supply Chain.pdf
How AI and ML Can Optimize the Supply Chain.pdf
 

Último

VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...
VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...
VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...
JRRolfNeuqelet
 
Forensic medicine MCQ for early learners
Forensic medicine MCQ for early learnersForensic medicine MCQ for early learners
Forensic medicine MCQ for early learners
AlaguPandi5
 
Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan 081901222272 Obat Penggugur Kandu...
Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan  081901222272 Obat Penggugur Kandu...Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan  081901222272 Obat Penggugur Kandu...
Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan 081901222272 Obat Penggugur Kandu...
Halo Docter
 
Unit 4 Pharmaceutical Organic Chemisty 3 Quinoline
Unit 4 Pharmaceutical Organic Chemisty 3 QuinolineUnit 4 Pharmaceutical Organic Chemisty 3 Quinoline
Unit 4 Pharmaceutical Organic Chemisty 3 Quinoline
AarishRathnam1
 
Connective Tissue II - Dr Muhammad Ali Rabbani - Medicose Academics
Connective Tissue II - Dr Muhammad Ali Rabbani - Medicose AcademicsConnective Tissue II - Dr Muhammad Ali Rabbani - Medicose Academics
Connective Tissue II - Dr Muhammad Ali Rabbani - Medicose Academics
MedicoseAcademics
 
Physiologic Anatomy of Heart_AntiCopy.pdf
Physiologic Anatomy of Heart_AntiCopy.pdfPhysiologic Anatomy of Heart_AntiCopy.pdf
Physiologic Anatomy of Heart_AntiCopy.pdf
MedicoseAcademics
 
Cytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose Academics
Cytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose AcademicsCytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose Academics
Cytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose Academics
MedicoseAcademics
 

Último (20)

supply cas 5449-12-7 BMK glycidic acid(powder) in stock EU pick-up
supply cas 5449-12-7 BMK glycidic acid(powder) in stock EU pick-upsupply cas 5449-12-7 BMK glycidic acid(powder) in stock EU pick-up
supply cas 5449-12-7 BMK glycidic acid(powder) in stock EU pick-up
 
Top 10 Most Beautiful Chinese Pornstars List 2024
Top 10 Most Beautiful Chinese Pornstars List 2024Top 10 Most Beautiful Chinese Pornstars List 2024
Top 10 Most Beautiful Chinese Pornstars List 2024
 
VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...
VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...
VIII.1 Nursing Interventions to Promote Healthy Psychological responses, SELF...
 
High Purity 99% PMK Ethyl Glycidate Powder CAS 28578-16-7
High Purity 99% PMK Ethyl Glycidate Powder CAS 28578-16-7High Purity 99% PMK Ethyl Glycidate Powder CAS 28578-16-7
High Purity 99% PMK Ethyl Glycidate Powder CAS 28578-16-7
 
Forensic medicine MCQ for early learners
Forensic medicine MCQ for early learnersForensic medicine MCQ for early learners
Forensic medicine MCQ for early learners
 
Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan 081901222272 Obat Penggugur Kandu...
Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan  081901222272 Obat Penggugur Kandu...Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan  081901222272 Obat Penggugur Kandu...
Obat Aborsi Ampuh Usia 1,2,3,4,5,6,7 Bulan 081901222272 Obat Penggugur Kandu...
 
ROSE CASE SPINAL SBRT BY DR KANHU CHARAN PATRO
ROSE  CASE SPINAL SBRT BY DR KANHU CHARAN PATROROSE  CASE SPINAL SBRT BY DR KANHU CHARAN PATRO
ROSE CASE SPINAL SBRT BY DR KANHU CHARAN PATRO
 
Unit 4 Pharmaceutical Organic Chemisty 3 Quinoline
Unit 4 Pharmaceutical Organic Chemisty 3 QuinolineUnit 4 Pharmaceutical Organic Chemisty 3 Quinoline
Unit 4 Pharmaceutical Organic Chemisty 3 Quinoline
 
Connective Tissue II - Dr Muhammad Ali Rabbani - Medicose Academics
Connective Tissue II - Dr Muhammad Ali Rabbani - Medicose AcademicsConnective Tissue II - Dr Muhammad Ali Rabbani - Medicose Academics
Connective Tissue II - Dr Muhammad Ali Rabbani - Medicose Academics
 
Face and Muscles of facial expression.pptx
Face and Muscles of facial expression.pptxFace and Muscles of facial expression.pptx
Face and Muscles of facial expression.pptx
 
spinal cord disorders and paraplegia .
spinal cord disorders  and  paraplegia .spinal cord disorders  and  paraplegia .
spinal cord disorders and paraplegia .
 
Overview on the Automatic pill identifier
Overview on the Automatic pill identifierOverview on the Automatic pill identifier
Overview on the Automatic pill identifier
 
Sell 5cladba adbb JWH-018 5FADB in stock
Sell 5cladba adbb JWH-018 5FADB in stockSell 5cladba adbb JWH-018 5FADB in stock
Sell 5cladba adbb JWH-018 5FADB in stock
 
Histopathological staining techniques used in liver diseases
Histopathological staining techniques used in liver diseasesHistopathological staining techniques used in liver diseases
Histopathological staining techniques used in liver diseases
 
Physiologic Anatomy of Heart_AntiCopy.pdf
Physiologic Anatomy of Heart_AntiCopy.pdfPhysiologic Anatomy of Heart_AntiCopy.pdf
Physiologic Anatomy of Heart_AntiCopy.pdf
 
Top 15 Sexiest Pakistani Pornstars with Images & Videos
Top 15 Sexiest Pakistani Pornstars with Images & VideosTop 15 Sexiest Pakistani Pornstars with Images & Videos
Top 15 Sexiest Pakistani Pornstars with Images & Videos
 
Anti viral drug pharmacology classification
Anti viral drug pharmacology classificationAnti viral drug pharmacology classification
Anti viral drug pharmacology classification
 
SEMESTER-V CHILD HEALTH NURSING-UNIT-1-INTRODUCTION.pdf
SEMESTER-V CHILD HEALTH NURSING-UNIT-1-INTRODUCTION.pdfSEMESTER-V CHILD HEALTH NURSING-UNIT-1-INTRODUCTION.pdf
SEMESTER-V CHILD HEALTH NURSING-UNIT-1-INTRODUCTION.pdf
 
Cytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose Academics
Cytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose AcademicsCytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose Academics
Cytoskeleton and Cell Inclusions - Dr Muhammad Ali Rabbani - Medicose Academics
 
Drug development life cycle indepth overview.pptx
Drug development life cycle indepth overview.pptxDrug development life cycle indepth overview.pptx
Drug development life cycle indepth overview.pptx
 

INDUSTRIALIZING INDUSTRIALIZING MACHINE LEARNING IN PHARMA: Challenges, Use Cases, & ROI

  • 1. INDUSTRIALIZING MACHINE LEARNING IN PHARMA: Challenges, Use Cases, & ROI Presenters: Daniel Kinney (Janssen), Ashish Sharma (Axtria), Kaiwen Zhong (Axtria) 1
  • 2. AGENDA 1 Journey from Traditional Analytics to Industrialized Machine Learning 2 What Does It Take to Industrialize AI / ML? 3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”? 4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning 2 1 Journey from Traditional Analytics to Industrialized Machine Learning
  • 3. AI / ML Beyond Pharma 3
  • 4. Maturity of ML / AI in Other Industries BANKING INSURANCE HCPGOVTRETAIL MANUFACT- URING NATURAL RESOURCES UTILITIES 37% 48% 38% 26% 44% 40% 35% 33% % of industry ready to deploy or already deployed large scale AI / ML In 12 months Top Use Cases: Fraud analysis on transactional data Chatbots Market / Customer Segmentation Process automation Computer-assisted diagnostics Text / Opinion Mining Source: Gartner (March 2019) 4
  • 5. Applications of ML / AI in Other Industries BANKING INSURANCE HCPGOVTRETAIL MANUFACT- URING NATURAL RESOURCES UTILITIES Automation of Intelligence Process Planning & Optimization Real time monitoring Public health surveillance Disaster planning Fraud detection Demand forecasting Cyber attack detection Personalized medicine Predictive optimization Fault analysis Forecasting Population health mgmt Demand forecasting Lifestyle segmentation Credit ratings Common Theme: 1. Enable additional and more accurate recommendations predictions 2. Provide timely insights 3. Automate repeatable steps to reduce SME costs and human errors Source: Gartner (March 2019) 5 High-speed trading Appointment scheduling Claims data entry Machine scheduling Autonomous drill bits Autonomous monitoring E-service Engineering assistants Scanning / mapping Product design
  • 6. AGENDA 1 Journey from Traditional Analytics to Industrialized Machine Learning 2 What Does It Take to Industrialize AI / ML? 3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”? 4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning 6
  • 7. PMSA AI / ML Topic Clustering Marketing (5) HCP Targeting (13) Patient Support (1) NLP (7) Provider (2) Adverse Events (1) New Patient ID (4) Brand Performance Analyzer: Identifying Opportunities and Commercial Performance Optimizer Using AI / ML for More Insightful Pipeline Marketing Decision-Making Automated Detection of Adverse Drug Reactions Using Social Media Text Data A New Generation of Patient Support Programs Application of Predictive Modeling to Define HCP Potential for Prophylactic Treatment Delighting Customers w/ Automated Content Personalization Dynamically Improving NLP & Theme Discovery Through Contextual Linguistics for Pharma A ML Approach to Customer Acquisition Discover latent Patient Insights Using Text Mining & ML NLP to Understand HCP Application of NLP & PCA in Life Sciences & Healthcare ML Techniques Deliver Granular Insights to enable Improved HCP Marketing Leveraging Event Pathways to Predict Disease Progression in Contrast to ML Classification Algorithms Applications of ML in Oncology Analytics: Combining Secondary & APLD Advanced and Primary Analytics A Novel Patient-Physician ML Approach on Diagnosing Rare Diseases and Informing Commercial Strategies Leverage Informed HCP Migration to Optimize Targeting & Multi-Channel Spend ML Against the Provider Graph Patient Co-Pay Card Optimization Use of ML and Patient Level Clinical Encounter Cues to Maximize Values of Sales Force Activities An All-Encompassing ML Attribution Model to Plan, Predict and Measure Impact at the Channel and HCP Level Unstructured to Structured Data: Can ML Help Make Sense of EMR/EHR Data?Strengthening Patient Journey with Emotional Insights via Social Analytics A Novel High-Dimensional Hybrid ML Approach Identifying High-Value Rare Disease Specialists Optimizing Physician Targeting to Find Undiagnosed Patients: An Application of Advanced ML Methods to Hep C ML to Better ID Physician Targets for New Product Launches Enhance Accuracy and Effectiveness Of Physician Alert 7
  • 8. Ecosystem Readiness for AI / ML in Pharma DIGITAL LEADERS AWS Machine Learning, SageMaker, Comprehend, Rekognition Verify, DeepMind Health, Tesseract, Google Brain ML Studio AWS Machine Learning, SageMaker, Comprehend, Rekognition Verify, DeepMind Health, Tesseract, Google Brain ML Studio REGULATIONS: FDA 12 ML / AI-based algorithms approved, e.g., DreaMed supports diabetes treatment decision Viz.ai analyzes CT results and highlights patients most likely to experience a stroke 12 ML / AI-based algorithms approved, e.g., DreaMed supports diabetes treatment decision Viz.ai analyzes CT results and highlights patients most likely to experience a stroke GAME CHANGING TECHNOLOGIES voted by CIOs 0% 10% 20% 30% Data Analytics AI / ML Digital Transformation Information Tech IoT CRM Source: Gartner (March 2019) 8
  • 9. PLATFORM CONFIGURATION DATA COLLECTION FEATURE EXTRACTION DATA VALIDATION MACHINE RESOURCE MANAGEMENT ANALYSIS / VISUALIZATIONPROCESS MANAGEMENT TOOLS CONSUMPTION INFRASTRUCTURE MONITORING ML ALGORITHMS PLATFORM CONFIGURATION DATA COLLECTION FEATURE EXTRACTION DATA VALIDATION MACHINE RESOURCE MANAGEMENT ANALYSIS / VISUALIZATIONPROCESS MANAGEMENT TOOLS CONSUMPTION INFRASTRUCTURE MONITORING ML ALGORITHMS De-Coding Industrialization of AI / ML Most focusLeast focus 9
  • 10. AGENDA 1 Journey from Traditional Analytics to Industrialized Machine Learning 2 What Does It Take to Industrialize AI / ML? 3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”? 4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning 10
  • 11. Challenges Industrializing Machine Learning  Downplaying the Role of Iterative Automation  Data Readiness  Adoption Barrier 11
  • 12. 3 Myths about Machine Learning  Inability to show ROI if ML underperforms with poor data and short timeframe  Inefficient one-off analysis with unnecessarily complicated ML algorithms 1. “BLACK BOX” Most use cases do not require or are not fit for “black box” algorithms like neural network. 2.IMMEDIATE IMPROVEMENT 3. DISRUPTING TODAY’S PROCESSES Improvement of processes depends on data availability, quality, and takes iterations. Machine learning models do not need to be developed from scratch. It can incorporate business rules / sit on top of rule- based, knowledge-based systems. HOWEVER HOWEVER HOWEVER Consequences 12
  • 13.  Increase use of publicly available data, which can be labeled quickly and cheaply by online resources.  ML (reinforcement learning) can also be used to enrich training data. It is daunting and time- consuming to create large amount of training data. Data scientists & analysts spend 80% time cleaning and prepping data. Data Readiness – “Garbage In, Garbage Out”  Organizations loses competitive advantage as they are afraid to adopt machine learning and advanced analytics,.  Underbudget for data and automation aspect of machine learning, leading to low ROI. “DATA JANITORIAL WORK” CREATE TRAINING DATA SOLUTION SOLUTION Consequences  Cloud & big data tech are increasingly available for the entire ML ecosystem.  Data strategy, governance, and management 13
  • 14. Ability to Impact Decisions with ML Insights  Inability to impact decision making eventually impact ROI of machine learning, making sustainable development of machine learning in organization difficult. FIELD ADOPTION  Show relationship between increased sales and adoption.  Tie adoption of insights into incentive compensation. NON-PERSONAL MARKETING OTHER STRATEGIC DECISIONS  Use tools to automatically track target responses to create training data.  Improve marketing vendor management with clear tracking system.  Identify pain points: financial, productivity, processes, support.  Avoid “technology for its own sake”. SUGGESTION SUGGESTION SUGGESTION Consequences 14
  • 15. SUCCESSFUL INDUSTRIALIZATION OF ML: Scraping Web Data for Multiple Vaccines University Requirements on Multiple Vaccines Automated Predictive Targeting with Real World Data, Payer Insights, & HCP Data 15
  • 16. Case 1: Predictive Targeting (1/4) We want to identify opportunities where machine learning can be used to leverage real world data for dynamic targeting across promotional channels PROBLEM STATEMENT OPPORTUNITY IDENTIFICATION • Call Targeting • Call Frequency • Product Priority • Messaging Optimization • Segmentation PERSONAL PROMOTIONS NON PERSONAL PROMOTIONS • Email Targeting • Media Buys • Direct Mail PEER TO PEER • Speaker Program Attendee Prospects 16
  • 17. Actionable, frequent insights as prediction is precise and specific Case 1: Predictive Targeting (2/4) TRADITIONAL APPROACH MACHINE LEARNING AUGMENTATION Aggregated Physician- level Rx Data Longitudinal Patient- level Data (claims) Payer & HCP Features Rx / Sales Volume Physician Prescription Propensity Predicted Biologic Initiation ML / AI FOR AUTOMATIONDATA READINESS BUSINESS IMPACT 17
  • 18. Case 1: Predictive Targeting (3/4) Scale Up & Productionize Pilot Changes to Test & Learn Feasibility & PoC1 2 3  Assess data availability  Data acquisition & feature engineering  Minimal viable product & integration  Conduct analysis to demonstrate POC  Implement & continuously improve using test/control approach  Continuous monitoring of performance  Productionize data management/analytics processes  Optimize data infrastructure Problem & Opportunity Identification/Ideation 18
  • 19. Case 1: Predictive Targeting (4/4) FEASIBILITY & PROOF OF CONCEPT PILOT CHANGES TO TEST & LEARN SCALE & SYSTEMIZE CHANGES PROCESS Assess data availability and conduct analysis to demonstrate proof of concept Implement & continuously improve using test/control approach Productionize data management/ analytics processes & optimize data infrastructure MODEL Baseline model to predict patient demand to proof out concept Improved model with machine learning for improved performance Continuous tuning & evaluation of model performance DATA Minimal viable product starting with claims data Incorporated additional HCP & Payer features into model Formalize data management support model & governance INFRASTRUCTURE Built locally leveraging Python Started buildout in analytics workbench Operationalize & automate with integration between analytics workbench & source systems (CRM, reporting, etc) CHANGE MANAGEMENT Sales & Marketing leadership alignment on potential value Piloted with test/control districts. Trained pilot users Scale up to entire salesforce and declare this is ‘business as usual’ way of working 1 2 3 19
  • 20. Case 2: Web Scraping for Targeted Marketing (1/5) • Gather information from 10,000+ global university (English) website domains to see if they mandate specific vaccinations for students / faculties. • How to identify large amount of new marketing opportunities for a launch product? • How to make data available for this project? • How to extract useful information from the unstructured web data? PAIN POINTS PROBLEM STATEMENT 20
  • 21. Case 2: Web Scraping for Targeted Marketing (2/5) PLATFORM CONFIGURATION DATA COLLECTION FEATURE EXTRACTION DATA VALIDATION MACHINE RESOURCE MANAGEMENT ANALYSIS / VISUALIZATIONPROCESS MANAGEMENT TOOLS CONSUMPTION INFRASTRUCTURE MONITORING ML ALGORITHMS SCALABILITY ITERATION ML ECOSYSTEM ML is scaled to gather information from 1 vaccine from 200 domains, to 1 vaccine from 10,000+ domains, to Multiple products in the department from 10,000+ domains  ML algorithms are built on top of rules-based system (iteration 0)  Time, resources, and data source domains are managed to run multiple iterations, with results improving over time. All components of industrialized ML ecosystem were considered at the beginning of the project to enable  Automation,  Efficient data gathering & analysis  Consumption of information & decision making 21
  • 22. Case 2: Web Scraping for Targeted Marketing (3/5) University Domain: http://www.Princeton.edu http://www.Harvard.edu http://www.Columbia.edu http://www.Yale.edu http://www.Upenn.edu http://www.Dartmouth.edu http://www.Brown.edu http://www.Cornell.edu … https://www.hbs.edu/Pages/default.aspx https://alumni.harvard.edu/community/volunteer https://huhs.harvard.edu/sites/default/files/Immunization_pac ket_C2023_2019.docx.pdf https://alumni.harvard.edu/programs-events/wtyc https://alumni.harvard.edu/travel/mode-of-travel/50-years-of- victory https://alumni.harvard.edu/programs-events/harvard-alumni- online-learning … Domain Sub URLs: Domain Content: Illustration 22
  • 23. Case 2: Web Scraping for Targeted Marketing (4/5) • Data Gathering / validation of 10,000+ university domains SOURCES • Set up data structures in AWS • Connect with scraping-related APIs and libraries CONFIGURE SERVER • Search all sub URLs of main university domains • Use existing and build custom web- crawlers using BS4, Requests, etc. SEARCH & CLEAN DATA READINESSDATA READINESS AI / ML FOR AUTOMATIONAI / ML FOR AUTOMATION 1. FILTER URLS 2. ID REQUIREMENTS Categorize URLs related to vs. not related to “student health” based on URL and main contents. Categorize university websites that require vs. not require client’s vaccines based on website text contents. … with the following NLP process: A. Keyword based categorization B. Create thesaurus for topics C. Iterative ML model building & evaluation BUSINESS IMPACTBUSINESS IMPACT ANALYZE & VISUALIZE RESULTS DETERMINE MARKETING STRATEGY Different commercial strategies for universities that 1) Requires the vaccines 2) Recommend but not require the vaccines 3) Mention but not require the related disease 4) Neither require nor mention the vaccines or the related disease 23
  • 24. Case 2: Web Scraping for Targeted Marketing (5/5) STEPS 1. FILTER URLS 2. IDENTIFY REQUIREMENT Keyword-based 3 5 Traditional ML Algorithms 2 4 BoW + Deep Learning 5 1 Hybrid with Traditional ML 1 (81%) 3 Hybrid with Deep Learning 4 1 (86%) Ranking Classification Accuracy on Test Data Methods Keyword-based: Rules-based algorithm using keywords related to “student health”, and particular vaccine Traditional ML Algorithms: Feature engineering for text processing, Algorithms include Random forest, SVM, and Naïve Bayes Deep Learning: Bag-of-words for text processing Hybrid with Traditional ML: Combining the results from keyword-based model with results from the best performer of traditional ML algorithms Hybrid with Deep Learning: Combining the results from keyword-based model with results from deep learning.
  • 25. AGENDA 1 Journey from Traditional Analytics to Industrialized Machine Learning 2 What Does It Take to Industrialize AI / ML? 3 Challenges and Pharma Success Stories: Why some organizations still call AI / ML a “hype”? 4 Future of Machine Learning in Pharma: ROI & sustainable growth of machine learning 25
  • 26. Calculating AI / ML ROI: Why is it difficult?  Both short term and long term values can be spread across multiple parts of the organizations.  It is hard to isolate the contribution of AI / ML in improvements, especially in large business outcomes. 26
  • 27. Iterations to Reach Agreed Upon Target Time Error rate Average error rate by human specialists Current error rate Iteration 0 Target error rate Illustrative example: Image classification error over time Iteration 1 Iteration 2 Iteration N … Improved accuracy Reduced human specialists costs and errors 27
  • 28. Considerations for Calculating AI / ML ROI REVENUECOSTDELIVERY SPEED TEAM EFFICIENCY COMPETITIVE ADVANTAGE BRAND EQUITY DIGITAL MATURITY More Tangible Less Tangible Example: Predict patient diagnosis accurately and target marketing to HCPs Example: More productively utilize highly paid professionals & sales reps Example: Provide rep NBAs real time / daily instead of monthly / quarterly Example: Extract more adverse events per day than when done by all manually Example: Identify un- discovered side effects faster COMPLIANCE RISK Example: Provide cost efficient and effective patient support programs Example: Quickly react to positive & negative media & online coverage on the brand Example: Conduct targeted and timely digital campaigns and non- personal detailing 28
  • 29. Key Takeaway You / your team does not need to be machine learning unicorn, But your organization should combine three key aspects of machine learning: data science, business analytics, and data engineering For high ROI and sustainable growth of machine learning. MACHINE LEARNING Business Analytics  Ensure ML model is set up to answer important business question  Communicate results to realize impact of ML  AND CHANGE MGMT Data Science  Develop and adjust ML models  Validate and continuously monitor model performance Data Engineering  Set up infrastructure to enable data combination, and model run  Scale models developed by data scientists in big data environment PLATFORM CONFIGURATION DATA COLLECTION FEATURE EXTRACTION DATA VALIDATION MACHINE RESOURCE MGMT CONSUMPTION INFRASTRUCTURE MONITORING ML ALGORITHMS ANALYSIS / VISUALIZATION PROCESS MGMT TOOLS 29
  • 30. INDUSTRIALIZING MACHINE LEARNING IN PHARMA: Challenges, Use Cases, & ROI Presenters: Danny Kinney (Janssen), Ashish Sharma (Axtria), Kaiwen Zhong (Axtria) 30