Facts and Fiction
SCL HUB 2016
Piet Buyck
VP of Sales, Forecasting Solutions
Demand Sensing
E2OPEN CONFIDENTIAL | © 2016 E2open 2
Welcome
• Since 1993 working with applications that
use applied mathematics to improve supply
chain solutions as designed by
organizations like Numetrix, Manugistics,
JD Edwards and Kronos
• VP of Sales for Terra since 2010, now
E2open
E2OPEN CONFIDENTIAL | © 2016 E2open 3
Agenda
• Demand Sensing
o Why is it necessary
o How does it work
o Myths
o Introducing more data
• Customer cases
• Questions
Terra Invented Demand Sensing in 2002
Accurate Most important goal is accuracy
Current
Forecasts reflect the latest information
available
Consistent
Forecasts are consistent with all observed
demand signals
Why is it necessary ?
Demand Sensing
E2OPEN CONFIDENTIAL | © 2016 E2open 6
Data Explosion Is Largely Untapped
o More data
o Richer data
o Unstructured data – mobile/social media
“Growth Through Innovation” is Driving Internal Complexity
Change in Network Complexity
E2OPEN CONFIDENTIAL | © 2016 E2open 8
Forecasting is Getting Harder
Direct Relationship between Forecast Error and Safety Stock
SDE 38% 81% 118% 149% 180%
Safety
Days
7 14 21 28 35
E2OPEN CONFIDENTIAL | © 2016 E2open 10
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10
%SalesorStock
Item Decile
% Sales % FG Stock
Higher Relative Capital Investment in 90% of Products
• Volume and Stock Distribution by Item Decile*
* Each group has 10% of the items
E2OPEN CONFIDENTIAL | © 2016 E2open 11
23X as Much Inventory Per $ of Sales in the Long Tail
80%
1%
45%
13%
0%
20%
40%
60%
80%
100%
Fast 10 Slow 50
%SalesorStock
% Sales % Capital Invested in FG Stock
E2OPEN CONFIDENTIAL | © 2016 E2open 12
Simplicity vs. Complexity - Navigation
Waterproof Water-resistant Destroyed by water
Can produce in a workshop Requires a printing press Requires factories
Maintenance-free Requires annual updates Has to be charged daily
Doesn’t work when cloudy Doesn’t work without light Works all of the time
Requires extensive training Requires some training Train yourself in a couple of
minutes
Cannot use when moving Cannot use when driving Works automatically when
driving
E2OPEN CONFIDENTIAL | © 2016 E2open 13
Technology – Accelerating the Pace of Change
E2OPEN CONFIDENTIAL | © 2016 E2open 14
Forecast Error Calculation Methodology
• All statistics are calculated at the item / ship-from location /
horizon-level
• MAPE – Mean Absolute Percent Error (normalized)
MAPE = sum(abs(Fcst – Ship)) / sum(Ship)
• Bias (normalized)
Bias = sum(Fcst – Ship) / sum(Ship)
E2OPEN CONFIDENTIAL | © 2016 E2open 15
Demand Sensing vs. Demand Planning
How does it work?
Demand Sensing
E2OPEN CONFIDENTIAL | © 2016 E2open 17
Why Is Forecasting Challenging?
Volume
E2OPEN CONFIDENTIAL | © 2016 E2open 18
0
200
400
600
800
1,000
1,200
1,400
10-Jul
24-Jul
7-Aug
21-Aug
4-Sep
18-Sep
2-Oct
16-Oct
30-Oct
13-Nov
27-Nov
A Big Order is Not Always an Increase in Volume
Oct28
E2OPEN CONFIDENTIAL | © 2016 E2open 19
To Create Accurate Forecasts with Better Math
• Uses advanced
pattern recognition
• Monitors your
demand signals
• Predicts what will be
ordered
Error dropped 60%.
DP Forecast
Actual History
Demand Sensing Forecast
Actual DS Forecast
DP Forecast
E2OPEN CONFIDENTIAL | © 2016 E2open 20
Demand Sensing Reconciles All Signals
Touchless
Every Item
Every Day
E2OPEN CONFIDENTIAL | © 2016 E2open 21
Demand Sensing vs. Demand Planning Performance
E2OPEN CONFIDENTIAL | © 2016 E2open 22
Better Forecasts: Promotions and New Items
• Universal performance across all segments of the business
• Promotions, new products, seasonal, fast-moving, slow-
moving
Promotions Performance New Product Performance
E2OPEN CONFIDENTIAL | © 2016 E2open 23
Fewer Disruptions from Extreme Error
• Primary source of supply chain disruptions and stock-outs
• Direct impact to service levels and on-shelf availability
Velocity
Extreme Under-Forecast
DP DS DS vs. DP
1 8% 2% -71%
2 10% 3% -72%
3 11% 4% -68%
4 14% 5% -67%
5 25% 9% -65%
E2OPEN CONFIDENTIAL | © 2016 E2open 24
Volatility and Resilience
Extreme Undersell Error
Excess Stock
Demand Planning
Demand volatility
Demand Sensing
Extreme Oversell Error
Order Cuts, Logistics Costs
E2OPEN CONFIDENTIAL | © 2016 E2open 25
• Cut forecast error by more than 40%
• Decreased safety stock by more than 30%
• Reduced finished goods inventory by $1.3 billion
• Cut forecast error 40%
• Removed 5 days of inventory
• Achieved significant freight savings
• Cut short-term forecast error 40%
• Produced the right product mix, decreased costs
and better served customers
• Cut forecast error by 50%
• Decreased safety stock 20%
• Improved order fill 5%
• Saved $20 million in inventory
Some Published Customer Results
E2OPEN CONFIDENTIAL | © 2016 E2open 26
Debunking Demand Sensing Myths
• Black Box
• It is daily, so I cannot react to it
• It is daily so it will disturb my world
• Does it flow back into my demand plan ?
E2OPEN CONFIDENTIAL | © 2016 E2open 27
Using Data Across the Value Chain
Multi-Enterprise Demand Sensing
E2OPEN CONFIDENTIAL | © 2016 E2open 28
Myth: Will POS Data Drive My Forecast
E2OPEN CONFIDENTIAL | © 2016 E2open 29
Myth: Demand Sensing Provides Planners with a Workbench
E2OPEN CONFIDENTIAL | © 2016 E2open 30
Reconciles All Demand Signals
Touchless
Every Item
Every Day
Business 1
Business 2
Orders
DP Fcst
Store Inv
DC Inv
Scanner
DC Ship to Store
P&G Shipments
Source – P&G presentation at the 2009
Terra Users’ Conference
What is predictive in one
business is not for another!
Let the Engine Determine the Predictors
E2OPEN CONFIDENTIAL | © 2016 E2open 32
Predictors Also Vary By Time
Customer Examples
© IGD 2014
Benefits – Inventory reduction – 35%
The benefits
With improved forecast accuracy Unilever has been able to reduce its safety stock levels
over time. Overall, Unilever has been able cut inventory in Europe by 35%.
Foods -43%
Personal Care -18%
Home Care -29%
Refreshment -38%
Inventory changes following implementation:
A leading position in
Operating Working Capital while service improves
Operating Working Capital
(% OWC / Revenues)
-10%
-5%
0%
5%
10%
15%
20%
25%
30%
AkzoNobel
TgT
E2OPEN CONFIDENTIAL | © 2016 E2open 36
Conclusions
• Volume and complexity of problem requires automation and computer
power, but complexity is behind the scenes like a GPS
• Error is reduced with 38% average and is the single biggest
opportunity to reduce working capital, increase resilience or service
levels
• Applicable for forecasting items that are made to stock in all industries,
CPG is just the most mature in this area
• More data provides more accuracy including Social Media, but biggest
gain is possible with internal data. 80% of our customers do not use
POS data
Thank you

Demand Sensing

  • 1.
    Facts and Fiction SCLHUB 2016 Piet Buyck VP of Sales, Forecasting Solutions Demand Sensing
  • 2.
    E2OPEN CONFIDENTIAL |© 2016 E2open 2 Welcome • Since 1993 working with applications that use applied mathematics to improve supply chain solutions as designed by organizations like Numetrix, Manugistics, JD Edwards and Kronos • VP of Sales for Terra since 2010, now E2open
  • 3.
    E2OPEN CONFIDENTIAL |© 2016 E2open 3 Agenda • Demand Sensing o Why is it necessary o How does it work o Myths o Introducing more data • Customer cases • Questions
  • 4.
    Terra Invented DemandSensing in 2002 Accurate Most important goal is accuracy Current Forecasts reflect the latest information available Consistent Forecasts are consistent with all observed demand signals
  • 5.
    Why is itnecessary ? Demand Sensing
  • 6.
    E2OPEN CONFIDENTIAL |© 2016 E2open 6 Data Explosion Is Largely Untapped o More data o Richer data o Unstructured data – mobile/social media
  • 7.
    “Growth Through Innovation”is Driving Internal Complexity Change in Network Complexity
  • 8.
    E2OPEN CONFIDENTIAL |© 2016 E2open 8 Forecasting is Getting Harder
  • 9.
    Direct Relationship betweenForecast Error and Safety Stock SDE 38% 81% 118% 149% 180% Safety Days 7 14 21 28 35
  • 10.
    E2OPEN CONFIDENTIAL |© 2016 E2open 10 0% 20% 40% 60% 80% 100% 1 2 3 4 5 6 7 8 9 10 %SalesorStock Item Decile % Sales % FG Stock Higher Relative Capital Investment in 90% of Products • Volume and Stock Distribution by Item Decile* * Each group has 10% of the items
  • 11.
    E2OPEN CONFIDENTIAL |© 2016 E2open 11 23X as Much Inventory Per $ of Sales in the Long Tail 80% 1% 45% 13% 0% 20% 40% 60% 80% 100% Fast 10 Slow 50 %SalesorStock % Sales % Capital Invested in FG Stock
  • 12.
    E2OPEN CONFIDENTIAL |© 2016 E2open 12 Simplicity vs. Complexity - Navigation Waterproof Water-resistant Destroyed by water Can produce in a workshop Requires a printing press Requires factories Maintenance-free Requires annual updates Has to be charged daily Doesn’t work when cloudy Doesn’t work without light Works all of the time Requires extensive training Requires some training Train yourself in a couple of minutes Cannot use when moving Cannot use when driving Works automatically when driving
  • 13.
    E2OPEN CONFIDENTIAL |© 2016 E2open 13 Technology – Accelerating the Pace of Change
  • 14.
    E2OPEN CONFIDENTIAL |© 2016 E2open 14 Forecast Error Calculation Methodology • All statistics are calculated at the item / ship-from location / horizon-level • MAPE – Mean Absolute Percent Error (normalized) MAPE = sum(abs(Fcst – Ship)) / sum(Ship) • Bias (normalized) Bias = sum(Fcst – Ship) / sum(Ship)
  • 15.
    E2OPEN CONFIDENTIAL |© 2016 E2open 15 Demand Sensing vs. Demand Planning
  • 16.
    How does itwork? Demand Sensing
  • 17.
    E2OPEN CONFIDENTIAL |© 2016 E2open 17 Why Is Forecasting Challenging? Volume
  • 18.
    E2OPEN CONFIDENTIAL |© 2016 E2open 18 0 200 400 600 800 1,000 1,200 1,400 10-Jul 24-Jul 7-Aug 21-Aug 4-Sep 18-Sep 2-Oct 16-Oct 30-Oct 13-Nov 27-Nov A Big Order is Not Always an Increase in Volume Oct28
  • 19.
    E2OPEN CONFIDENTIAL |© 2016 E2open 19 To Create Accurate Forecasts with Better Math • Uses advanced pattern recognition • Monitors your demand signals • Predicts what will be ordered Error dropped 60%. DP Forecast Actual History Demand Sensing Forecast Actual DS Forecast DP Forecast
  • 20.
    E2OPEN CONFIDENTIAL |© 2016 E2open 20 Demand Sensing Reconciles All Signals Touchless Every Item Every Day
  • 21.
    E2OPEN CONFIDENTIAL |© 2016 E2open 21 Demand Sensing vs. Demand Planning Performance
  • 22.
    E2OPEN CONFIDENTIAL |© 2016 E2open 22 Better Forecasts: Promotions and New Items • Universal performance across all segments of the business • Promotions, new products, seasonal, fast-moving, slow- moving Promotions Performance New Product Performance
  • 23.
    E2OPEN CONFIDENTIAL |© 2016 E2open 23 Fewer Disruptions from Extreme Error • Primary source of supply chain disruptions and stock-outs • Direct impact to service levels and on-shelf availability Velocity Extreme Under-Forecast DP DS DS vs. DP 1 8% 2% -71% 2 10% 3% -72% 3 11% 4% -68% 4 14% 5% -67% 5 25% 9% -65%
  • 24.
    E2OPEN CONFIDENTIAL |© 2016 E2open 24 Volatility and Resilience Extreme Undersell Error Excess Stock Demand Planning Demand volatility Demand Sensing Extreme Oversell Error Order Cuts, Logistics Costs
  • 25.
    E2OPEN CONFIDENTIAL |© 2016 E2open 25 • Cut forecast error by more than 40% • Decreased safety stock by more than 30% • Reduced finished goods inventory by $1.3 billion • Cut forecast error 40% • Removed 5 days of inventory • Achieved significant freight savings • Cut short-term forecast error 40% • Produced the right product mix, decreased costs and better served customers • Cut forecast error by 50% • Decreased safety stock 20% • Improved order fill 5% • Saved $20 million in inventory Some Published Customer Results
  • 26.
    E2OPEN CONFIDENTIAL |© 2016 E2open 26 Debunking Demand Sensing Myths • Black Box • It is daily, so I cannot react to it • It is daily so it will disturb my world • Does it flow back into my demand plan ?
  • 27.
    E2OPEN CONFIDENTIAL |© 2016 E2open 27 Using Data Across the Value Chain Multi-Enterprise Demand Sensing
  • 28.
    E2OPEN CONFIDENTIAL |© 2016 E2open 28 Myth: Will POS Data Drive My Forecast
  • 29.
    E2OPEN CONFIDENTIAL |© 2016 E2open 29 Myth: Demand Sensing Provides Planners with a Workbench
  • 30.
    E2OPEN CONFIDENTIAL |© 2016 E2open 30 Reconciles All Demand Signals Touchless Every Item Every Day
  • 31.
    Business 1 Business 2 Orders DPFcst Store Inv DC Inv Scanner DC Ship to Store P&G Shipments Source – P&G presentation at the 2009 Terra Users’ Conference What is predictive in one business is not for another! Let the Engine Determine the Predictors
  • 32.
    E2OPEN CONFIDENTIAL |© 2016 E2open 32 Predictors Also Vary By Time
  • 33.
  • 34.
    © IGD 2014 Benefits– Inventory reduction – 35% The benefits With improved forecast accuracy Unilever has been able to reduce its safety stock levels over time. Overall, Unilever has been able cut inventory in Europe by 35%. Foods -43% Personal Care -18% Home Care -29% Refreshment -38% Inventory changes following implementation:
  • 35.
    A leading positionin Operating Working Capital while service improves Operating Working Capital (% OWC / Revenues) -10% -5% 0% 5% 10% 15% 20% 25% 30% AkzoNobel TgT
  • 36.
    E2OPEN CONFIDENTIAL |© 2016 E2open 36 Conclusions • Volume and complexity of problem requires automation and computer power, but complexity is behind the scenes like a GPS • Error is reduced with 38% average and is the single biggest opportunity to reduce working capital, increase resilience or service levels • Applicable for forecasting items that are made to stock in all industries, CPG is just the most mature in this area • More data provides more accuracy including Social Media, but biggest gain is possible with internal data. 80% of our customers do not use POS data
  • 37.