PREDICTIVE ANALYTICS
© 2015 Software AG. All rights reserved.
OVERVIEW
© 2015 Software AG. All rights reserved.
3 |
THE ANALYTICS SPECTRUM
ALL ANALYTICS ADD VALUE BUT ANSWER DIFFERENT QUESTIONS
Difficulty
Value
Descriptive
What
happened?
Predictive
What will
happen?
Diagnostic
Why did it
happen?
Streaming
What is
happening?
4 |
STREAMING ANALYTICS AND PREDICTIVE ANALYTICS
…while they can still change the outcome
BOTH TECHNIQUES COMPLEMENT THE OTHER
Predictive Analytics allows organizations to
build models that represent patterns of behavior
Streaming Analytics uses these models to
enable organizations to respond intelligently
5 |
PREDICTIVE ANALYTICS FOUNDATIONS
All predictive analytic methods and models are based upon one common premise:
What has happened in the past will likely happen again
So once we learn from the past, we know what to look for in the future
And all of this is driven by the data itself, so:
The more data you have, the more you need Predictive Analytics
A FEW SIMPLE PRINCIPLES MAKE THIS POSSIBLE
6 |
BUILDING A PREDICTIVE MODEL
Primarily, we are looking for whether a relationship exists between two variables
THE DATA DRIVES THE INSIGHTS
77
170
20
40
60
80
100
Dark Clouds Clear sky
UNEVEN
DISTRIBUTION
% did it rain
There is a likely relationship between dark clouds in the
sky and whether it rained
23 22 22 24 23 24 26
0
20
40
60
EVEN
DISTRIBUTION
% did it rain
No relationship between the day of the week and whether
it rained
7 |
BUILDING A PREDICTIVE MODEL
THE LOGIC CAN BE SIMPLE OR COMPLEX
An example of a decision tree showing whether it is likely to rain
8 |
Time
Business
Value
Business event
High value
response
Time to act
Low value
response
Forrester Research calls this
“perishable insight”
TIME VALUE OF DATA
THE LONGER THE REACTION, THE LOWER THE VALUE
9 |
PREDICTIVE ANALYTICS IN USE TODAY
Entertainment: recommendations suggest new movies based upon your viewing
history
– Leverages your tastes and the tastes of others like you
ALGORITHMS ARE ALL AROUND US
10 |
PREDICTIVE ANALYTICS IN USE TODAY
Finance: consumer credit scores model the likelihood of your paying the loan back
– Models your probable behavior based upon the defaults of many, many others
ALGORITHMS ARE ALL AROUND US
11 |
PREDICTIVE ANALYTICS IN USE TODAY
Health: heart monitors warn of undiagnosed heart problems
– Watches for known pulse irregularity patterns
ALGORITHMS ARE ALL AROUND US
12 |
THE BUSINESS VALUE OF
PREDICTIVE ANALYTICS
Fraud detection
Bank detects unusual spending
pattern on your card Retail
Making relevant offers at the right
time
Predictive Maintenance
Diagnosing a failing pump
13 |
SMART LOGISTICS
SUPPLY CHAINS CAN RUN MORE EFFICIENTLY
• Enable shippers to
plan activities
efficiently in the harbor
Objective
• Change Transport
Mode or lane
Automated action
• Arrival time
• Route failure
prediction
Predictive Analytics
• Position, Status, ETA
Always On Analytics
14 |
SMART STORE MONITORING
MERCHANTS CAN MAXIMIZE REVENUE
• Traffic density
• POS data
• Shelf sensors
Always On Analytics
• Increased revenue
• More effective
merchandising &
service
Objective
• Proactive staff re-
deployment
• Offer updates
Automated action
• When offers adjusted
• When queues appear
• When shelves need
replenishing
Predictive Analytics
15 |
PREDICTIVE MAINTENANCE
FIELD SERVICES CAN PREVENT OUTAGES
• Real-time conditions
Always On Analytics
• 99.999% uptime
• Increased 1st Call
Repair Rates
Objective
• Technician Dispatch
• Field Service
Automation
Automated action
• Failure prediction
• Remaining useful life
of components
Predictive Analytics
16 |
PREDICTIVE MAINTENANCE EXAMPLE
PREDICTIVE ALERTS ALLOW MORE TIME TO REACT
• Clear signal leading up to
failure…
– But reliant on human
intuition to interpret in real
time?
• Condition monitoring can alert
on thresholds:
– Tells you something might
be wrong, but not what or
how urgent
– Ignores machine-specific
differences
Condition
Monitoring Alert
Predictive
Alert
Visual
Identification
F
A
I
L
U
R
E
TIME $
17 |
OPERATIONALIZING PREDICTIVE ANALYTICS
REAL-TIME PREDICTIONS NEED TO BE EASY TO DEPLOY
Data-
bases
Event Feed
Event Feed
Event Feed
Actions
Alerts
Notifications
Apama
Applications
DevelopmentRuntime
PMML
Predictive
Models
Data
Management
Applications
USE CASES AND
CUSTOMER EXAMPLES
19 |
SOFTWARE AG USE CASES FOR PREDICTIVE ANALYTICS
Solution Industries Predictive Use Cases Enhancement Core
Enabler
Predictive
Maintenance
• Manufacturing
• Failure Prediction
• Remaining Useful Life
x
Connected Customer
• Retail
• Hospitality
• Financial Services
• Telecom
• Next Best Offer
• Churn Detection
• Queue Prediction
• Path Analytics
• Facial Recognition
x
Smart Metering &
Manufacturing
• Utilities
• Manufacturing
• Electricity Theft Detection
• Quality Anomaly Detection
x
Smart Logistics
• Logistics
• Manufacturing
• Route Failure Prediction
• Arrival Forecasting
x
Fraud Detection
• Financial Services
• Retail
• Probabilistic Models x x
PARTNERING
21 |
THE SOFTWARE AG STRATEGY
• Partnering with organizations with data science skills
– This stuff isn’t impossibly complex
– But it does need specialist modeling skills
• We are building on the Software AG platform
– Gives us a fantastic integration message
– Includes new OEM components such as Predictive Analytics for Apama
– Along with components we resell such as KNIME and Predixion
• Open approach means we can work with the customer’s preferred modeling tools
© 2015 Software AG. All rights reserved. For internal use only
22 |
GO-TO-MARKET OPTIONS BY CUSTOMER TYPE
Standardized on
SAS/SPSS/etc
Export models in
PMML and execute
via Apama
No preferred
tools yet, R
Propose KNIME
or Predixion as a
platform
Willing to build
DS skills
Offer KNIME’s or
Predixion’s
simple, graphical
approach along
with trainings
Unwilling to
build DS skills
Introduce
services partners
like Mosaic
Customer has
in-house data science skills
Customer does not have
in- house data science skills
OPPORTUNITY BRAINSTORMING
24 |
MARKET OPPORTUNITY
• Streaming Analytics worth $2 billion* by 2020 (Markets and Markets)
• Predictive Analytics worth $7 billion by 2019 (Transparency Market Research)
© 2015 Software AG. All rights reserved. For internal use only
PREDICTIVE ANALYTICS IS EVEN BIGGER THAN STREAMING
*Arguably, the value of streaming analytics has been underestimated as
predictive analytics drives growth of streaming analytics Transparency Market Research © 2012
Market size by type of
application:
25 |
HOW TO SPOT AN OPPORTUNITY
Anywhere where an organization could benefit from knowing an event is
likely to happen before it happens!
– Are they using SAS, SPSS, or R?
– Are they using Apama?
– Do they have a Big Data initiative?
© 2015 Software AG. All rights reserved. For internal use only
26 |
ACCOUNT DISCUSSION
© 2015 Software AG. All rights reserved. For internal use only
27 |

Predictive analytics roadshow

  • 1.
    PREDICTIVE ANALYTICS © 2015Software AG. All rights reserved.
  • 2.
    OVERVIEW © 2015 SoftwareAG. All rights reserved.
  • 3.
    3 | THE ANALYTICSSPECTRUM ALL ANALYTICS ADD VALUE BUT ANSWER DIFFERENT QUESTIONS Difficulty Value Descriptive What happened? Predictive What will happen? Diagnostic Why did it happen? Streaming What is happening?
  • 4.
    4 | STREAMING ANALYTICSAND PREDICTIVE ANALYTICS …while they can still change the outcome BOTH TECHNIQUES COMPLEMENT THE OTHER Predictive Analytics allows organizations to build models that represent patterns of behavior Streaming Analytics uses these models to enable organizations to respond intelligently
  • 5.
    5 | PREDICTIVE ANALYTICSFOUNDATIONS All predictive analytic methods and models are based upon one common premise: What has happened in the past will likely happen again So once we learn from the past, we know what to look for in the future And all of this is driven by the data itself, so: The more data you have, the more you need Predictive Analytics A FEW SIMPLE PRINCIPLES MAKE THIS POSSIBLE
  • 6.
    6 | BUILDING APREDICTIVE MODEL Primarily, we are looking for whether a relationship exists between two variables THE DATA DRIVES THE INSIGHTS 77 170 20 40 60 80 100 Dark Clouds Clear sky UNEVEN DISTRIBUTION % did it rain There is a likely relationship between dark clouds in the sky and whether it rained 23 22 22 24 23 24 26 0 20 40 60 EVEN DISTRIBUTION % did it rain No relationship between the day of the week and whether it rained
  • 7.
    7 | BUILDING APREDICTIVE MODEL THE LOGIC CAN BE SIMPLE OR COMPLEX An example of a decision tree showing whether it is likely to rain
  • 8.
    8 | Time Business Value Business event Highvalue response Time to act Low value response Forrester Research calls this “perishable insight” TIME VALUE OF DATA THE LONGER THE REACTION, THE LOWER THE VALUE
  • 9.
    9 | PREDICTIVE ANALYTICSIN USE TODAY Entertainment: recommendations suggest new movies based upon your viewing history – Leverages your tastes and the tastes of others like you ALGORITHMS ARE ALL AROUND US
  • 10.
    10 | PREDICTIVE ANALYTICSIN USE TODAY Finance: consumer credit scores model the likelihood of your paying the loan back – Models your probable behavior based upon the defaults of many, many others ALGORITHMS ARE ALL AROUND US
  • 11.
    11 | PREDICTIVE ANALYTICSIN USE TODAY Health: heart monitors warn of undiagnosed heart problems – Watches for known pulse irregularity patterns ALGORITHMS ARE ALL AROUND US
  • 12.
    12 | THE BUSINESSVALUE OF PREDICTIVE ANALYTICS Fraud detection Bank detects unusual spending pattern on your card Retail Making relevant offers at the right time Predictive Maintenance Diagnosing a failing pump
  • 13.
    13 | SMART LOGISTICS SUPPLYCHAINS CAN RUN MORE EFFICIENTLY • Enable shippers to plan activities efficiently in the harbor Objective • Change Transport Mode or lane Automated action • Arrival time • Route failure prediction Predictive Analytics • Position, Status, ETA Always On Analytics
  • 14.
    14 | SMART STOREMONITORING MERCHANTS CAN MAXIMIZE REVENUE • Traffic density • POS data • Shelf sensors Always On Analytics • Increased revenue • More effective merchandising & service Objective • Proactive staff re- deployment • Offer updates Automated action • When offers adjusted • When queues appear • When shelves need replenishing Predictive Analytics
  • 15.
    15 | PREDICTIVE MAINTENANCE FIELDSERVICES CAN PREVENT OUTAGES • Real-time conditions Always On Analytics • 99.999% uptime • Increased 1st Call Repair Rates Objective • Technician Dispatch • Field Service Automation Automated action • Failure prediction • Remaining useful life of components Predictive Analytics
  • 16.
    16 | PREDICTIVE MAINTENANCEEXAMPLE PREDICTIVE ALERTS ALLOW MORE TIME TO REACT • Clear signal leading up to failure… – But reliant on human intuition to interpret in real time? • Condition monitoring can alert on thresholds: – Tells you something might be wrong, but not what or how urgent – Ignores machine-specific differences Condition Monitoring Alert Predictive Alert Visual Identification F A I L U R E TIME $
  • 17.
    17 | OPERATIONALIZING PREDICTIVEANALYTICS REAL-TIME PREDICTIONS NEED TO BE EASY TO DEPLOY Data- bases Event Feed Event Feed Event Feed Actions Alerts Notifications Apama Applications DevelopmentRuntime PMML Predictive Models Data Management Applications
  • 18.
  • 19.
    19 | SOFTWARE AGUSE CASES FOR PREDICTIVE ANALYTICS Solution Industries Predictive Use Cases Enhancement Core Enabler Predictive Maintenance • Manufacturing • Failure Prediction • Remaining Useful Life x Connected Customer • Retail • Hospitality • Financial Services • Telecom • Next Best Offer • Churn Detection • Queue Prediction • Path Analytics • Facial Recognition x Smart Metering & Manufacturing • Utilities • Manufacturing • Electricity Theft Detection • Quality Anomaly Detection x Smart Logistics • Logistics • Manufacturing • Route Failure Prediction • Arrival Forecasting x Fraud Detection • Financial Services • Retail • Probabilistic Models x x
  • 20.
  • 21.
    21 | THE SOFTWAREAG STRATEGY • Partnering with organizations with data science skills – This stuff isn’t impossibly complex – But it does need specialist modeling skills • We are building on the Software AG platform – Gives us a fantastic integration message – Includes new OEM components such as Predictive Analytics for Apama – Along with components we resell such as KNIME and Predixion • Open approach means we can work with the customer’s preferred modeling tools © 2015 Software AG. All rights reserved. For internal use only
  • 22.
    22 | GO-TO-MARKET OPTIONSBY CUSTOMER TYPE Standardized on SAS/SPSS/etc Export models in PMML and execute via Apama No preferred tools yet, R Propose KNIME or Predixion as a platform Willing to build DS skills Offer KNIME’s or Predixion’s simple, graphical approach along with trainings Unwilling to build DS skills Introduce services partners like Mosaic Customer has in-house data science skills Customer does not have in- house data science skills
  • 23.
  • 24.
    24 | MARKET OPPORTUNITY •Streaming Analytics worth $2 billion* by 2020 (Markets and Markets) • Predictive Analytics worth $7 billion by 2019 (Transparency Market Research) © 2015 Software AG. All rights reserved. For internal use only PREDICTIVE ANALYTICS IS EVEN BIGGER THAN STREAMING *Arguably, the value of streaming analytics has been underestimated as predictive analytics drives growth of streaming analytics Transparency Market Research © 2012 Market size by type of application:
  • 25.
    25 | HOW TOSPOT AN OPPORTUNITY Anywhere where an organization could benefit from knowing an event is likely to happen before it happens! – Are they using SAS, SPSS, or R? – Are they using Apama? – Do they have a Big Data initiative? © 2015 Software AG. All rights reserved. For internal use only
  • 26.
    26 | ACCOUNT DISCUSSION ©2015 Software AG. All rights reserved. For internal use only
  • 27.

Editor's Notes

  • #7 Mike’s slide
  • #8 Mike’s slide
  • #13 Jeremy’s slide
  • #14 CASE STUDY: Royal Dirkzwager This picture you see, is a dashboard of the shipping around the harbor of Rotterdam. Royal Dirkzwager is one of the world’s leading maritime service providers. Core capability of Dirkzwager is to “gather full shipping movement details (such as positions) and relate them to clients in real-time’. The accurate and continuous flow of this information enables clients (shippers) to plan activities and processes as efficiently as possible. CCO Jeroen Kortsmit says; “Royal Dirkzwager’s core process is collecting, interpreting, validating and distributing information to logistics suppliers, warehousing companies, merchants and other interested parties in the shipping industry.” OBJECTIVE Enable shippers to plan activities efficiently in/nearby harbor Based on the real-time and accurate information they can; Steer third parties (providing food for crew, cleaning) to support the ships on time in full Optimize the harbor capacity. ALWAYS ON ANALYTICS CCO Jeroen Kortsmit says; “Royal Dirkzwager’s core process is collecting, interpreting, validating and distributing information to logistics suppliers, warehousing companies, merchants and other interested parties in the shipping industry.” The always on analytics contains European ship positions, status and ETA. PREDICTIVE ANALYTICS Dirkzwager uses Apama to enable shippers to predict the Estimated Arrival Time (ETA) of any ship anywhere and then revise the ETA continuously as circumstances change. The service tracks ships as they cross virtual lines and feeds clients continuous location-based event updates. AUTOMATED ACTIONS Automated messages are sent to clients what the ETA means anticipating on activities when they arrive in the harbor Automated messages are sent as soon as a monitored vessel enters (or exits) a zone, the subscriber receives details by email or text. Through this messaging service, Royal Dirkzwager directly supports its clients in optimizing their own operational processes.
  • #15 CASE STUDY: Retailers No reference. OBJECTIVE Increased revenue More effective merchandising & customer service ALWAYS ON ANALYTICS Traffic density POS data Shelf sensors PREDICTIVE ANALYTICS + AUTOMATED ACTIONS Predict queues: automatically re-deploying staff when there is going to be a problem/queue in the next 10 mins, Predict lack of specific inventory; adjusting a promotion (with altered signage) when it is performing different to plan (ie not well enough or too well that you will run out of inventory) Predict empty shelves; replenishing shelves before they are completely empty.  Check Smart Store Monitoring Deck and the Planet Retail White paper http://salesweb.eur.ad.sag/Solutions/solutions/industries/retail/#SMSTMO
  • #16 CASE STUDY: GE Jenbacher This picture is of a General Electric Jenbacher power generator used by the plastics factory. Worldwide, about 11.000 Jenbacher generators are operating. These massive engines, 8.4 meters long and 4.7 meters wide, can power everything from large factories to small cities. OBJECTIVE Reliability is key: factories must produce, cities must be livable. GE has aggressive SLAs with an objective of 99.999% uptime from each engine. Besides, GE Software has the mission to make GE’s customers’ systems 1% more efficient Sounds like not too difficult, but this equates to a saving of 66 billion dollars over 15 years for GE customers. ALWAYS ON ANALYTICS The trick is “fix it before it breaks down”. But this is difficult since the engines are used in variety of ways: some run at a constant rate – others experience peaks – and differences in light/heavy usage. GE needs to monitor constantly and real-time the conditions of the machines to know what could go wrong and when. PREDICTIVE ANALYTICS A breakdown must be predictable, otherwise GE can never meet the objective of 99.999% uptime. Solution with Software AG: Using smart system software that includes 300 sensors reviewed continuously The data is analysed & visualised and compared to predictive maintenance rules over time Smart analytics determine when a generator bearing will ‘wear out’ or ‘malfunction’ or predict how much life is left AUTOMATED ACTIONS Automated actions -in time- will increase efficiency and uptime that will generate the 1% improvement Actions such as initiating technician dispatch and automated task management for Field Service.
  • #18 Mike’s slide
  • #22 Mike’s slide
  • #25 Jeremy’s slide
  • #26 Jeremy’s slide