transforming big data into supply chain analytics
TRANSCRIPT
Transferring Big Data into Supply Chain Analytics
Alan Milliken CFPIM CSCP CPFSr. Manager – Supply Chain Capability DevelopmentBASF
Transforming Big Data into Supply Chain AnalyticsSAPICS Conference 2015
Alan L. Milliken CFPIM CSCP CPF CSOPJune, 2015
� Term used to refer to the mass of information being generated today.
� In 2012, it was estimated that 2.5 exabytesof data were created each day. (1 exabyte = 1B gigabytes)
� The amount of data available is expected to double every 3 years.
� Technology increases data availability, enables communication of data and provides the ability to analyze the information.
What is “Big Data?”
3
What is/are Analytics?
The discipline of using math, statistics and models to extract knowledge from data. The knowledge gained is used to improve decision making and performance.
Businesses use analytics to describe, predict and improve performance.
4
What are the three major supply objectives which analytics help to improve?
5
In a 2012 *SAS-MIT survey with 2,500 respondents from over 20 industries, 67% indicated they are using analytics to improve overall performance.
?
*Source: “Reimagining the Possible with Data Analytics.” MIT Sloan Management Review, Spring, 2013 in collaboration with SAS.
Gathering and Structuring Data for Analysis
6
Take care to get what you like, or you will be forced to
like what you get! «
»
George Bernard Shaw
Sample of characteristics used to report sales and forecast performance:
� Material dimensions – SKU, Product, Product Group, etc.
� Customer dimensions – Sold-To, Payer, Customer Group, etc.
� Accounting dimensions – BU, SBU, Profit Center, etc.
� Geographical dimensions – country, region, sub-region, state, customer, etc.
Involve key users and process experts in definition phase.
Determine What Characteristics are Needed?
7
Determine What Key Figures are Required?
Note: All characteristics and key figures are available for analysis.
Sample of key figures used to generate a forecast
Sample of key figures used to report performance.
8
Note: If the sum of F-A = 0 there is no BIAS.
Forecast BIAS%:Measurement of a continuous under or over estimation of actual sales
Develop & Program Key Figure Definitions
Formula
n = # of months = 6 BIAS = -18%
Sales
FCST
Jan 04 Feb 04 Mrz 04 Apr 04 Mai 04 Jun 04
1.000.000
900.000
800.000
700.000
600.000
500.000
400.000
300.000
200.000
100.000
0
9
Sample Content:
� absolute values & quantities
� Sales by BU-Product-Region
� Forecast by SKU-Customer
region, sub region, country, company, plant
Organization
OD
BU
SBU
Main Group
α-code Geography
Material
Company hierachy
SupplyChain
hierachy
SKU, Product, etc.
Material hierachy
Determine Data Structure
Note: Multi-dimensional data cube provides key figures to support data mining.
10
Central Activities Copy
� North America
� South America
� Europe
� Asia Pacific
KPI calculation Data storage
Globally available
Extractions(raw)
Calculation of KPIs1
Store the data inaggregation levels to providefast access
2
Daily extractionin the regions
Gather and Process Data
11
Data Mining and Sample Reports
Data mining: the process of extracting information from a data set and transforming it into a usable structure, supports analytics.
Data
Business under-standing
Data under-standing
Data preparation
Modeling
Deve-lopment
Evaluation
12
What we want!
Data Mining – the process of extracting usable information from a dataset
“Big Data” Mine
13
Overview reports: Stay on a high level and provide a quick overview
� Overview
►Required for most of the report executions
►Only a limited number of characteristics � Short response time
►Might be relevant for every user
� Specialized
►Required for more special reporting needs
►Cover additional characteristics to allow specialized reporting
►Might only be relevant for certain users
Analysis reports
►Required for special purpose analysis
►High number of characteristics � Longer response times
►Detailed information, going down to the lowest level of thedocuments
Sample of Reporting Structure
14
(1) Select Standard Report (A) or Custom Query (B):
(2) Enter Query Characteristics Desired
A
B
Sample Standard Report –User Specific Characteristics
15
Sample Standard Report –User Specific Characteristics
Forecast Accuracy on Customer Level:
Note: Results are in excel format for easy analysis and charts.
Product area Country Grouped Article Customer Group Calendar Year/Month 04.2009 05.2009 06.2009 07.2009 Over all Result
EMN DE Germany Stat FCA (Art-Cust) 100 100 100 100 100
Sales FCA (Art-Cust) 100 100 100 100 100
Stat FCA (Art-Cust) 100 100 100 100 100
Sales FCA (Art-Cust) 100 100 100 100 100
Stat FCA (Art-Cust) 0 100 0 0 67
Sales FCA (Art-Cust) 0 100 0 0 67
Stat FCA (Art-Cust) 0 100 0 0 67
Sales FCA (Art-Cust) 0 100 0 0 67
Stat FCA (Art-Cust) 0 100 0 0 67
Sales FCA (Art-Cust) 0 100 0 0 67
ES Spain Stat FCA (Art-Cust) 100 100 0 0 0
Sales FCA (Art-Cust) 100 100 0 0 0
Stat FCA (Art-Cust) 100 100 0 0 0
Sales FCA (Art-Cust) 100 100 0 0 0
Stat FCA (Art-Cust) 100 100 0 0 0
Sales FCA (Art-Cust) 100 100 0 0 0
US USA Stat FCA (Art-Cust) 100 100 0 0 0
Sales FCA (Art-Cust) 100 100 0 0 0
Calendar Year/Month 04.2009_07.2009
APO Planning Version 000 – Active Version
Grouped Article
Customer Group
Produkt area
16
Analytics – Education & Training
17
Descriptive Analytics
Used to measure performance, report what happened, why it happened and plan for improvement.
18
Standard Analytical Reports: Decision Cockpit:BASF Decision Cockpit – S&OP Demand Review
All key elements of the Analytical Reports can be displayed in the Decision Cockpit to enable management review..
S&OP Level Demand Review Dashboard
Analytics provide management with information to make better decisions.
Past PerformanceForecast Accuracy for S&OP Family XYZ: 69% (Target range: 50% – 70%)
Issues� Customer Demand in
France decreases about 7%� Price acceptance in Spain will
decrease
Planning Information Decisions� Adjust Demand Plan for S&OP
Family XYZ� Decrease sales price target for
S&OP Family XYZ and adjust budget quantities for Q4 2012
Time Series
Demand Review (ETA based)� 1a, b, c. KPI: FCA and BIAS (3 reports for 3 level)� 2. Forecast Qty (incl. Budget, Target)
19
Business Level –Forecast Accuracy KPI Report
„Dashboards“ to simplify the presentation of KPI‘s are best.
All KPI‘s should include „Targets“ and a way to identify those areas needing immediate attention.
Red-Yellow-Green light functionality is quite popular universally
SBU 2011 YE Nov-12 2012 YTD Target57% 65% 65% 70%
64% 62% 66% 70%
88% n/a 88% 70%
Total 76% 63% 73% 70%
Forecast Accuracy: Measures ability to forecast at CM + 2 lag
SBU 2011 YE Nov-12 2012 YTD Target6% 3% 9% 10%
1% 8% 6% 10%
4% n/a 1% 10%
Total 3% 6% 5% 10%
Bias:Measures direction of forecast error at CM + 2 lag
SBU 2011 YE Nov-12 2012 YTD Target36% 38% 42% 50%
45% 40% 41% 50%
55% n/a 59% 50%
Total 39% 39% 42% 50%
Hit Miss:Measures % of portfolio accurately forecasted
Forecast Accuracy & BIAS are global KPI‘s at BASF. Hit/Miss based on tolerances are also popular.
EM – 2012 Forecast Accuracy
EM – 2012 Bias
EM – 2012 Hit Miss
20
Sample Diagnostic AnalyticProvides exceptions by individual responsibility:
Exceptions: � Sales but no forecast
� Forecast but no sales
� Difference in sales and forecast > 50%
21
Customer Marketing 03-2010 03-2010 03-2010 03-2010
SBU Article # Article Description Group Manager Actual Sales Statistical FC Sales FC Fin. Reg. Market. FC
ZAC 79345968 Dispersion Bulk ABC 001 J. Smith 0.00 KG 6,803.89 KG 6,803.89 KG 6,803.89 KG
ZAC Result 0.00 KG 6,803.89 KG 6,803.89 KG 6,803.89 KG
ZAC 78459312 Dispersion 190 KG ABC 002 J. Smith 0.00 KG 1,416.00 KG 1,416.00 KG 1,416.00 KG
ZAC ABC 003 J. Smith 0.00 KG 123.43 KG 123.43 KG 123.43 KG
ZAC ABC 004 J. Smith 1,520.00 KG 740.57 KG 740.57 KG 740.57 KG
ZAC ABC 005 J. Smith 190.00 KG 0.00 KG 0.00 KG 0.00 KG
ZAC ABC 006 J. Smith 5,320.00 KG 0.00 KG 0.00 KG 0.00 KG
ZAC Result 7,030.00 KG 2,280.00 KG 2,280.00 KG 2,280.00 KG
ZAC 74698213 Dispersion 25 KG ABC 007 J. Smith 0.00 KG 15.54 KG 15.54 KG 0.62 KG
ZAC ABC 008 J. Smith 0.00 KG 15.54 KG 15.54 KG 1.54 KG
ZAC ABC 009 J. Smith 0.00 KG 21.75 KG 21.75 KG 0.00 KG
ZAC ABC 010 J. Smith 0.00 KG 31.07 KG 31.07 KG 2.78 KG
ZAC ABC 011 J. Smith 0.00 KG 6.22 KG 6.22 KG 0.62 KG
ZAC Result 0.00 KG 90.11 KG 90.11 KG 5.56 KG
Sample Exception Report
Analytics should include financial impact when feasible.
Top 20 Forecast Accuracy DriversRegion Article Number Article Name ABC Forecast Accuracy Error (kgs) Error ($)N-NAFTA 5089 A 0% 17775 $159,442
N-NAFTA 5074 A 50% 17584 $193,424
N-NAFTA 5005 A 56% -16000 $199,040
N-NAFTA 5254 A 0% -12350 $469,300
MX 5004 A 34% -11480 $98,269
N-NAFTA 5000 A 0% 11280 $259,891
N-NAFTA 5013 A 50% -8300 $0
N-NAFTA 5508 A 60% -8000 $136,000
MX 5508 A 60% -5600 $94,360
N-NAFTA 5508 A 54% 5050 $99,788
N-NAFTA 5239 B 0% 5000 $61,000
N-NAFTA 5253 A 0% 4500 $832,500
N-NAFTA 5676 A 47% -4460 $38,178
N-NAFTA 5001 C 0% 3750 $46,125
N-NAFTA 5007 C 0% -3500 $0
N-NAFTA 5000 B 0% 3200 $224,000
N-NAFTA 5002 A 0% 3000 $43,590
N-NAFTA 5006 B 39% -2600 $42,172
N-NAFTA 5001 A 0% 2400 $33,528
MX 5002 B 28% -1970 $38,474
Subtotal (721) $837,495
22
Forecast Accuracy Measures are Descriptive
Descriptive Analytics provide feedback to improve Predictive Analytics.
MPE = ∑ (Actual – FCST) x100 MAPE= ∑ │(Actual – FCST) x100
Actual Actual
No. Of Observations No. Of Observations
Period Sales (Units) FCST (Units) Error % Error Abs. % Error
1 999 1319 -320 -32.03 32.03%
2 1178 1141 37 3.14% 3.14%
3 1247 1168 79 6.36% 6.36%
4 1469 1141 328 22.31% 22.31%
5 1074 1298 -224 -20.86% 20.86%
6 1568 1263 305 19.43% 19.43%
7 1159 1370 -211 -18.23% 18.23%
8 1383 1267 116 8.39% 8.39%
9 1552 1370 182 11.73% 11.73%
10 1174 1365 -191 -16.24% 16.24%
Sum -16.01% 158.72%
MPE -1.60%
MAPE= 15.87%
│
23
COV = Std. Dev’n in Period Demand/Average Demand
Using Variance to Test for Forecastability
Descriptive Analytics support key decisions in supply chain management.
� 29 of 46 products are forecastable.
� Over 90% of volume is forecastable.
� 37% of products represent 6% of sales volume. Check profitability and place on make-to-order status.
Product Group A
37%
26%
37%
67%
27%
6%
0%
10%
20%
30%
40%
50%
60%
70%
80%
0-0.50 0.51-1.0 >1.0
% of Total Products
% of Total Sales
COV Range
24
Predictive Analytics
The analysis of current and/or historical data to make predictions about the future.
25
Sample Forecasting Techniques (Predictive)
Descriptive Analytics provide feedback to improve Predictive Analytics.
PeriodActualSales ($M)
3-Period Sum
3-PeriodAvg. (FCST)
1 356
2 372
3 374
4 380 1102 367
5 365 1126 375
6 373 1119 373
7 367 1118 373
8 373 1105 368
9 374 1113 371
10 380 1114 371
11 368 1127 376
12 371 1122 374
Avg. 371
σ 6.6
400
390
380
370
360
350
3401 2 4 6 83 5 7 9
Units
Period
Sales
FCST
26
Statistical FC
Supply Planning:� Feasibility check� Production quantities
–1 CD
Sales FC
Regional Marketing FC
Demand Validation Meeting
Qualitative input may be supported with quantitative analyses.
Predictive Analytics May Include Qualitative Inputs
1–2 CD 3–6 CD 7–9 CD 10–12 CD 13 CD 17 CD
Unit M 09/ 2009 M 10/ 2009 M 11/ 2009 M 12/ 2009 M 01/2010
Statistical FC KG 4,428,853 4,307,289 4,307,289 4,307,289 4,307,289
Final Sales FC KG 4,089,409 3,833,127 2,821,739 3,021,739 3,387,431
RegionalMarketing FC
KG
Final Regional Marketing FC
KG 4,093,316 3,836,253 2,832,553 3,032,553 3,396,082
Constraint FC KG 3,922,763 3,832,553 2,832,553 3,032,553 3,396,082
27
Advanced Predictive Analytic
Leading Indicators include:
� Average Weekly Hours Manufacturing
� Weekly Claims for Unemployment
� New Orders – Manufacturing
� Housing Starts
� Stock Prices (500 stocks)
� Money Supply & Interest Rate
Actual Sales:
Quantities & Dates at Customer Ship-to Level
3-months EWS indicator
Model includes exponential smoothing and multi-regression analysis.
120
110
100
90
80
70
60
3- months Index (100 = Avg Actuals 2012)
Jan10
Jul10
Jul11
Jul12
Jul13
Jan11
Jan12
Jan13
Jan14
28
Predicting Forecastability
Statistical ForecastSales: ReviewMarketing: Review
Statistical ForecastSales: Proactive InputMarketing: Review
Statistical ForecastSales: Proactive Input only for selective(large) customers (� 80/20 Rule)Marketing: Review selective Customers
Only statistical forecastSales: No routine action requiredMarketing: No routine action required
Not possible to forecastSwitch to MTO or ex-Stock SalesNo routine action by any party
Forecasting RolesDetailed Forecast (Article-Customer
A B C
X
Y
Z
Volume
Variation of demand
29
Using Analytics in Optimization of Processes
This application requires advanced math and statistics enabled by advanced software.
The goal: Maximize Profits
30
Inventory
Develop the Business Objective
Maximize profits while balancing demand & supply within capacity constraints and inventory limits:
Productioncapacity
Variable demand
Contract demand
Goal: Maximize EBIT
31
Sales targetsfor regions(for Reg. Bus. Mgmt.)
Production plan(for Production)
External Purchases
Raw material demand
Distribution plan(for SCM)
Inventory plan(for information)
Demand forecasts(from marketing)
Capacities (from production)
Inventories(from BW reports)
BOM‘s, Routings(from production)
Value parameters(from Controlling and BW)
Optimization
Optimization: Input-Output Information
Profit optimized
32
Sample of Data and Process Structure
Rules:
� Sales order requests must be within plan.
� Sales order must be provided 4 weeks before expected shipment.
� Unplanned sales orders are reviewed by marketing & operations management.
Data & Process Information:
� Customer segmentation analysis and results.
� Total delivered cost from each plant to customer.
� On-line, real-time check of requests versus optimized plan.
� Exception report monitoring of planned vs. actual sales.
33
� More is better at this point.
� What information is needed?
� Multi-dimensional: e.g. sale by product & customer.
� Characteristics (e.g. Business Unit) and Key Figures (e.g. Forecast in Units)
� Enable identification of trends, patterns, exceptions, etc.. . Generate KPI’s and Dashboards.
Summary
1
Gather & Store Data
2
Develop Target Data Implement Data Structure
3
Transform Data to Information
Transform Info to Knowledge
Take action to improve
4 5 6
34
IBF Events – Analytics
Please go to: IBF.org for more details. !
December 17 – 19, 2014IBF‘s Demand Planning and Forecasting Boot Camp w/ Predictive Analytics / Big Data Workshop New York City, NY
Atlanta, GAApril 25, 2015 Predictive Business Analytics, Forecasting and Planning Conference April 23 – 24, 2015
35
36
37