Predictive analytics can help reduce volatility and improve decision making in procurement processes. It allows understanding of future costs, demand, and supply to overcome challenges. Predictive models analyze past data and behaviors to forecast trends and outcomes. As data sources like IoT sensors expand, predictive analytics is increasingly used for applications like manufacturing process improvement, predictive maintenance of equipment, and optimizing building energy usage.

1. Predictive analytics of
demand
for optimization of
Procurement Process
By :Prashant Bhatmule
VP-SCM
Bajaj Electricals Ltd

2. Points to cover
• Understanding the models, simulations and variables
• selection for better use of predictive analytics.
• How to reduce the volatility in decision making process using
• predictive analysis in procurement process
• How to overcome challenges in the procurement process ,
• inventory management understanding the future cost,
• demand, supply of resources using predictive analytics?
• Understanding multiple uses of predictive analytics
• in terms of suppliers and vendors
• Understand ways to address the challenges with managing
• Big Data and approaches for improving your supplier risk.

3. 3

4. Predictive Analytics - Definition
Predictive analytics is an area of data mining that deals
with extracting information from data and using it to
predict trends and behavior patterns. Often the
unknown event of interest is in the future, but predictive
analytics can be applied to any type of unknown whether
it be in the past, present or future.
--Wikipedia

5. Evolution of Analytics
5/25/2017
5
Seat of the pants:
Intuition, domain
knowledge
BI (data driven):
Visualizations, KPI’s,
OLAP, Dashboards
Predictive Analytics:
Modeling,
Experimental
verification
Real-time streaming
analytics:
Deep learning, A.I.
Complexity
Insights
Future looking
Past looking

6. Predictive Analytics Process
Model
Verification
Theoretical
Model
Treatments +
Controls + Error =
Response
Domain
knowledge,
1st
principles,
experience
Pre-existing
data

7. Traditional Predictive Analytics
• Applications
• Research and development in agriculture and industry
• Designed experiments for product and process improvement
• Verifying the effectiveness of healthcare treatments
• Clinical trials (randomized double blind are best)
• Predicting the weather
• Polls: opinions, election politics, Nielson ratings.
• Standardized tests: e.g. SAT’s to predict college success
• Actuarial Science: life expectancy, etc.
• Limitations
• Each data point must be planned and collected intentionally
• Expensive to design, collect the data

8. 21st CenturyTransformation: Big Data Content
1. Internet behavior and logs
• Every click and keystroke stored, server logs
• Automatic byproduct of the Internet
2. Internet content
• Text, emails, statuses, images, audio, video, etc.
• Voluntary product of Internet use, automatically harvested
3. Internet of things
• Sensors of all types: temperature, pressure, kinematics, images, audio, etc.
• Must be consciously set up and designed
4. Access to thousands of sources of traditionally collected data
• Open data movement (gov’t, academia)

9. Who is using big data?
“Big data is like teenage sex: everyone talks about it, nobody really knows how
to do it, everyone thinks everyone else is doing it, so everyone claims they are
doing it…”—Dan Ariely (2012)
Even if this was true in 2012, it is no longer true; many organizations are
successfully implementing modern predictive analytics with big data.

10. Analytic process changes
• Huge quantity and variety of data available with little effort/cost
• Substantial extra effort needed to extract and process data that have been collected without a
specific end use in mind (ETL)
• Unstructured data images, videos, audio, etc.
• Combination of a variety of data sources
• Some creativity required to compose relevant metrics
• Data availability leads to increased data mining, exploratory data analysis and data visualization
requirements
• Hundreds or even thousands of variables are “thrown in” to analyses
• More “statistically significant” relationships found
• Some new insights that are real
• Some mistakes due to over-fitting, random clusters (some algorithms exist to help filter out false “hits”)
• Domain knowledge / insights still crucial to distinguish real effects from random chance or allied effects

11. TraditionalTools andTechniques
• TraditionalTools
• Mostly commercial – SAS, SPSS, STATA, Minitab, Excel
• Open Source – R, Python
• TraditionalTechniques
• Various flavors of regression analyses
• Linear, multinomial, logistic, general linear models, regression discontinuity, instrumental variables,
panel models, time series, etc.
• Designed experiments,ANOVA
• Controls built in for known covariates
• Randomization to accommodate unknown covariates
• Sometimes designed to extract maximum possible info from fewest observations
11

12. Big DataTools
• Big Data Predictive AnalyticsTools
• Open Source
• Apache Spark / MLlib
• Python scripts interfaced to Hadoop or other distributed data source
• Sometimes R or other traditional tools if data can be extracted and reduced to a size to fit on a single processor.
• Mondrian (OLAP – more BI than predictive analytics)
• Commercial / Proprietary
• RapidMiner (Radoop for Hadoop analytics)
• Pentaho – compare effectiveness of various models: operates against many data sources (being acquired by Hitachi)
• HPVertica Distributed R – ML algorithms pre-loaded
• BeyondCore - 1st through 4th order interactions automatically calculated
• SAS Enterprise Miner
• SPSS adding big data analytics modules
• IBM Predictive Maintenance and Quality
• Homegrown solutions internal to companies (hand-coded Python or Java algorithms)

13. Big DataTechniques
• Big Data Predictive AnalyticsTechniques
• Numerical responses / Categorical responses
• TraditionalAnalyticalTechniques – Linear, logistic regressions,GLM regressions, simulations
• Distributed Analytics – i.e. MapReduce
• Machine LearningAlgorithms - Bayes network, decision trees, rule engines
• A/BTesting – Designed randomized experiments to confirm model’s predictive power
• More use of natural experiments (due to increased sharing of data sets)

14. Predictive Analytics:
Current Use Cases
Directly related to the sponsoring organization’s goals
14

15. Use Cases
• What determines actual usage of analytics?
• As always: organizational goals
• Business / organizational goals drive initiatives
1. Improve outcomes (not obviously dollar related)
• For non-profits: improve client outcomes
• Fulfill the non-market based goal
• Improve products, customer satisfaction
• Provide the best product/experience and the customer will come
2. Increase profits by increasing revenue
3. Increase profits by reducing costs
15

16. Optimize Energy Usage - BuildingIQ

17. Legendary Pictures: Applied Analytics Division
• Legendary Pictures
• Movies such as Interstellar, Godzilla, Dracula Untold,
Man of Steel, Inception, Unbroken
• Global scale and reach
• Innovation factory, not a warehouse
• Use of predictive analytics now a part of key strategies to
inform creative decisions
• Concept and cast evaluation
• Green lighting concepts
• Cast strengths and weaknesses
• Build trailers on analytics
• Proprietary, commercial-grade software constantly analyzes
unique, extensive data across people, social media, and
content.
• 500 million emails, 250 million households
• Entertainment data
• Metadata on films since 1989
• Transform marketing
• Old fashioned: 4 quadrants: male/female/over 25/under 25; 4
groups of 80 million each
• Targeting: 80 million groups of 4
• Aiming at “swing groups” not fans nor never interested “Mom”
17

18. Manufacturing Process Improvement
• Intel - Using IoT and predictive analytics in chip manufacturing
• Process and product measurements taken upstream during the manufacturing process are able
to predict whether the chip will pass final test leading to a reduction in final testing
requirements.
• "Instead of running every single chip through 19,000 tests, we can focus tests on specific chips to
cut down test time.“
• This predictive analytics process, implemented on a single line of Intel Core processors in 2012,
allowed Intel to save $3 million in manufacturing costs. In 2017-18, Intel expects to extend the
process to more chip lines and save an additional $30 million, the company said.

19. Predictive Maintenance
• How Predictive Analytics ImprovesWindTurbine Maintenance

20. Model Building
logistic regression model to predict the probability of default
• Probability of bad = w1(Var1)+w2(Var2) +w3(Var1)……+w20(var20)
• Logistic regression gives us those weights
• Predicting the probability
• Odds of bad=0.13(number of cards)+ 0.21 (utilization)+…….+0.06(number of loan
applications)
• That’s it ….we are done

21. Actual model Building Steps
Objective &
Portfolio
identification
Customer score
vs. account
score
Exclusions
Observation
point and/or
window
Performance
window
Bad definition Segmentation
Validation plan
and samples
Variable
Selection
Model Building
Scoring and
Validation
Score
Implementation

22. Predictive AnalyticsTrends
• Cost of analytics will be driven down by:
• Standardization of IoT
• Spread of expertise
• Commercialization and widespread use of successful
techniques
• Development of “best practices” and user-friendly
software to support them in specific key applications
• Web and web-content data
• Profusion of analytical tools will consolidate to a
couple of winners
• Most used and useful tools will be identified and
commercialized
• New applications will continue to target identifiable
individuals in the “persuadable middle”
• Internet ofThings
• Development of interface standards will lead to
more commercially available tools as well as open
source solutions
• Healthcare analytics
• Will increasingly use Big Data as privacy and
regulatory issues resolved
• Leading edge
• Move from manual to dynamic model generation as
understanding of relevant factors improves (especially
in IoT, since things tend to have more consistent
behaviors than people)
• Analysis of streaming data becomes more
mainstream
22

23. 23

24. DataVisualizations
Data visualizations can help you make the right decisions quickly…
that will ultimately lead to success.

25. Data Analysis & Predictive Modeling
Increasingly, business rely on intelligent tools and techniques to analyze data systematically to
improve decision-making.
 Retail sales analytics
 Financial services analytics
 Telecommunications
 SupplyChain analytics
 Transportation analytics
 Risk & Credit analytics
 Talent analytics
 Marketing analytics
 Behavioral analytics
 Collections analytics
 Fraud analytics
 Pricing analytics

26. Dude we all know that
Data analytics rocks &
Predictive modeling is
going to rule!!!

32. How to
use
Data
>

34. • Automated Replenishments prevent stock outs and reduce cost
• Enable Flexible allocation scenario
• Can use integrated plans and automate allocation
• Helps in Sales and operations planning
• Allows to implement Consensus Demand planning
• Aligns the organisation

35. How does bank approve or decline CC and Loan
applications?
Any suggestions…
Chill
maddi
• Basis Predictive analysis out of Big Data !!!!!!!!

36. The Problem
Who will run away with my money?
• Question asked by all Credit card issuing banks ???

37. Citi has a similar problem?
Who is going to run away with my money?
• Given Historical of the customers we want to predict the probability of bad
• We have the data of each customer on
• Customer previous loans, customer previous payments, length of account credit
history, other credit cards and loans, job type, income band etc.,
• We want to predict the probability of default

38. How to predict?
We have historical data and we have statistics

39. Every Purchase Office builds a model that
gives a score to eachVendor
“Developing set of equations or mathematical formulation to forecast future
behaviors based on current or historical data.”
Applicants

40. Predictive Modeling
Lets try to understand predictive modeling
Predictive Modeling – Fitting a model to the data to predict the future.
• Predicting the future –and it is so easy some times
• Who is going to score more runs in IPL-2020?
• That’s it you predicted the future..
• Predicting the future based on historical data is
nothing but Predictive modeling

41. Predictive Modeling
Lets try to understand predictive modeling
Predictive Modeling – Fitting a model to the data to predict the future.
• Who is going to score more runs in IPL 2020?
• Predicting the future …well it is not that easy …

42. Predictive Modeling
Horse Race Analogy
How to bet on best horse in a horse race

43. The Historical Data
Win vs. Loss record in past 2 years
• Long legs: 75% (Horses with long legs won 75% of the times)
• Breed A: 55%, Breed B: 15 % Others : 30%
• T/L (Tummy to length) ratio <1/2 :75 %
• Gender: Male -68%
• Head size: Small 10%, Medium 15% Large 75%
• Country: Africa -65%

44. Given the historical data
What about best one in this lot?
Horse-1 Horse-2 Horse-3 Horse-4 Horse-5 Horse-6 Horse-7 Horse-8 Horse-9 Horse-10
Length of
legs
109 114 134 130 149 120 104 117 115 135
Breed C A B A F K L B C A
T/L ratio 0.1 0.8 0.5 1.0 0.3 0.3 0.3 0.6 0.7 0.9
Gender Male Female Male Female Male Female Male Female Male Female
Head size L S M M L L S M L M
Country Africa India Aus NZ Africa Africa India India Aus Africa

45. Even more variables

46. Identifying most impacting factors
Out of 500 what are those 20 important attributes

47. A technology platform that transforms your
business data, various demographics and
socio economic datasets into meaningful
location based analytics for making robust
and quick decisions.

49. Challenges/Key Requirements
1.Saturated markets? Any untapped markets still there.
2.What is the potential within the catchment of ‘X’ km around store.
3.How to organize/ re-define the territories for optimum distribution and sales?
4.Should I open new distribution point or cannibalize my existing non-performing
distribution points? But Where to open andWhat to cannibalize?
5.How to reach to the potential markets? Accessibility?
6.How a particular outlet/ distributor/ area is performing w.r.t. to the other
around it.

51. Model Building
Historical Data
The model
equation

52. Questions?
Prediction is very difficult, especially if it’s about the
future.
-- Niels Bohr

53. Thank You
Prashant Bhatmule