Slide Deck from the webinar
The odds of achieving successful outcomes when using enterprise demand management technology for supply chain planning are the same as a flip of the coin. Why? The programs are just too labor intensive and the engines are not fit for purpose. The odds improve when optimizers are added to increase the output from ERP systems like SAP and Oracle. In this webinar, we share insights on the benefits of using an optimizer for SAP. Gain insights from Lora Cecere, Founder of Supply Chain Insights, and Gerrott Faulkingham from ToolsGroup on the steps to take to reinvigorate your demand planning system.
24. Traditional Forecast Methods
1. Are adequate at handling fast moving items
2. Do not leverage existing data
3. Cannot take advantage of additional data streams/external inputs
25. Traditional Forecast Methods
1. Are adequate at handling fast moving items
2. Do not leverage existing data
3. Cannot take advantage of additional data streams/external inputs
27. Volatility (COV)
WMAPE
Forecast Error %
0 .5 1.0 1.5 2.0 2.5
20
30
40
50
60
70
80+
Opportunity
Demand Modeling
Traditional Forecasting
Tail Items
Risk
Forecast Error is a Difficult Problem to Manage
28. What is Demand Modeling
0
1
2
3
4
5
Demand
Demand Modeling is the Science of Calculating Probabilities or
Ranges of How Demand Could Occur
29. What is Demand Modeling
0
1
2
3
4
5
Demand Forecast
Demand Modeling is the Science of Calculating Probabilities or
Ranges of How Demand Could Occur
30. What is Demand Modeling
0
1
2
3
4
5
Demand Forecast
Demand Modeling is the Science of Calculating Probabilities or
Ranges of How Demand Could Occur
31. What is Demand Modeling
0
1
2
3
4
5
Demand Forecast Demand modeling
understands there is
inherent uncertainty
associated with future
demand whether that SKU
is a fast mover or a slow
mover
Demand Modeling is the Science of Calculating Probabilities or
Ranges of How Demand Could Occur
32. Traditional Forecast Methods
1. Are adequate at handling fast moving items
2. Do not leverage existing data
3. Cannot take advantage of additional data streams/external inputs
33. Point of Sale Data
Daily Ship-To
Daily Ship-From
Weekly Shipments
Monthly Shipments by
Customer
Leveraged using the “traditional” approach
Detail lost in “traditional” approach
Data Leveraged: Traditional vs. Probabilistic
34. Why Demand Details Matter
Same aggregate
historical sales
SKU: A
SKU: B
Traditional
35. Why Demand Details Matter
Same aggregate
historical sales
Same forecast
result
SKU: A
SKU: B
Traditional
36. Why Demand Details Matter
SKU: A
SKU: B
Same aggregate
historical sales
Traditional
Different detailed
ordering pattern
Probabilistic
Same forecast
result
37. Why Demand Details Matter
SKU: A
SKU: B
Same aggregate
historical sales
Vastly different
forecast certainty
Traditional
Different detailed
ordering pattern
Probabilistic
Same forecast
result
38. Traditional Forecast Methods
1. Are adequate at handling fast moving items
2. Do not leverage existing data
3. Cannot take advantage of additional data streams/external inputs
39. Probabilistic Forecast
Trend, Seasonality, Calendars and
Daily Sale Patterns
Market Intelligence
3
6
7
Trade Promotion
Media Event Effect4
Special Actions and Events
5 New Product Introduction
THE DAILY BASELINE
DemandInsightIncreasing
1
2
STOCHASTIC MODELING
MACHINE
LEARNING
DEMAND SHAPING
PLANNER
“Layers” of Demand Modeling
41. Traditional Forecast Methods
1. Are adequate at handling fast moving items
2. Do not leverage existing data
3. Cannot take advantage of additional data streams/external inputs
Why are companies still using the same traditional forecasting methods which have been around for
decades to solve the business problems of today?
Hinweis der Redaktion
Bryan Semple – Vice President Healthcare at ToolsGroup UK.
Previously was a customer of Systagenix when I worked in procurement and supply chain for the UK National Health Service
Systagenix then became a customer of mine when I joined ToolsGroup in 2012.
Systagenix is now part of Acelity, the world’s largest wound care company. Acelity globally supplies more than 20 million advanced wound dressings per month. Revenue (2015) was approximately $1.9B
Offer a full range of dressings and therapeutic devices for healing severe wounds and preventing them from leading to more serious consequences.
We initially worked with the Systagenix company from 2012. Following their acquisition in 2014, we worked with them through their integration process and have expanded the solution to plan all KCI products as well.
Some of Acelity’s products are high volume, mature products. However, like many medical devices companies they are continually innovating and introducing new solutions which improve healthcare outcomes. There is therefore a continual change in dynamics of their demand flows and challenges in forecasting demand for new products and end of life products. This is further complicated by variable regulatory approval and clinical trial timescales.
Faced a challenging supply chain. Products were distributed to (100 countries?) via six regional 3PLs.
Some markets, including US, Canada and Western Europe are mature and serviced locally by own sales companies. Feature is regular ordering, often directly by hospitals and healthcare provider groups, relatively stable, although even this can be disturbed by regulatory and other factors..
Other regions including Middle East, Eastern Europe and Asia Pacific are serviced by distributors. Order profile is more variable with potentially large MOQs and irregular order intervals.
Emerging markets are often sold on a tender basis – some speculative, very large volumes, potentially disruptive and difficult to predict certainty and timing.
Following a successful seven month pilot, Systagenix went live with the hosted SaaS version of the ToolsGroup software in July 2013.
SO99+ extracts historical demand data from Systagenix’s SAP ERP system to automatically calculate a demand forecast. SO99+ has a unique ability to factor in the demand variability at the order-line level in order to optimize our safety stocks.
Next, the forecasts are refined further with input from the commercial, finance team before finally being used to calculate optimized safety stocks based on target service levels.
The system forecasts SKU level demand by individual market and then calculates safety stock targets at six 3PL stocking locations across the global supply chain for all 800+ SKUs.
Finally, the forecast and these dynamic safety stocks are loaded back into the ERP system, which then executes the planned replenishment actions.
Reduced time to produce a forecast by SKU from one week to one day. The forecaster’s remaining time is also used much more productively and satisfyingly, to refine the forecasts with input from the commercial team. The second forecaster now supports another part of the business.
Visibility of demand and ability to model potential changes to demand (scenario planning) is now possible.
Despite targeting higher 99 percent service levels at the 3PL distribution sites, inventory levels have been reduced by up to 15 percent.
On this measure alone, the investment paid off in the same year as implementation, allowing us to invest in other areas to maximize global service levels.
The monthly global forecast that used to take the two planners an entire week now can be accomplished by one full-time person in a single day, a 10 X improvement in planning productivity
A highly successful implementation. This has now also been adopted across the Acelity Woundcare business, which also included integration with KCI’s Oracle ERP and APCS solution
Picture and role and email
Cannot handle medium/slow movers at all
Requires a deep understanding of statistics and/or a PHD level education to correctly identify a method or algorithm which works for the business
“high touches” are required to keep the model running.
When planners do have to switch between algorithms because an item changes sales behavior – It adds unnecessary volatility into the model, making inventory and supply planning even more difficult because the forecast is a “moving target”
Does not scale well with a company as their complexity increases (forecasting 100 items can be trivial, forecasting 10,000 items can require a team of 20)
Weekly/ daily level granularity
Includes downstream data (POS, Neilson, IRI, etc)
Assumes that all items have a normal distribution of demand and orderlines
Prone to overfitting, does not properly predict future behavior
Businesses have many experts with a vast array of niche information
Stored both systemically and in a business process
Loses the ability to merge promotional data, external variables, product introduction, end of life planning, and other signals in one model
Is a type of problem that machine learning can leverage - vast amounts of data and complexity which has high impact to forecast error
The future of demand planning is rapidly approaching and business needs are ever changing. Companies need to be more agile than ever.
Cannot handle medium/slow movers at all
Requires a deep understanding of statistics and/or a PHD level education to correctly identify a method or algorithm which works for the business
“high touches” are required to keep the model running.
When planners do have to switch between algorithms because an item changes sales behavior – It adds unnecessary volatility into the model, making inventory and supply planning even more difficult because the forecast is a “moving target”
Does not scale well with a company as their complexity increases (forecasting 100 items can be trivial, forecasting 10,000 items can require a team of 20)
Weekly/ daily level granularity
Includes downstream data (POS, Neilson, IRI, etc)
Assumes that all items have a normal distribution of demand and orderlines
Prone to overfitting, does not properly predict future behavior
Businesses have many experts with a vast array of niche information
Stored both systemically and in a business process
Loses the ability to merge promotional data, external variables, product introduction, end of life planning, and other signals in one model
Is a type of problem that machine learning can leverage - vast amounts of data and complexity which has high impact to forecast error
The future of demand planning is rapidly approaching and business needs are ever changing. Companies need to be more agile than ever.
Why the long tail is getting bigger
… this is why we need to move towards a demand model, and away from traditional historical level forecasting
Clear statement that there is a better way of doing this – “new way of modeling”
Cannot handle medium/slow movers at all
Requires a deep understanding of statistics and/or a PHD level education to correctly identify a method or algorithm which works for the business
“high touches” are required to keep the model running.
When planners do have to switch between algorithms because an item changes sales behavior – It adds unnecessary volatility into the model, making inventory and supply planning even more difficult because the forecast is a “moving target”
Does not scale well with a company as their complexity increases (forecasting 100 items can be trivial, forecasting 10,000 items can require a team of 20)
Weekly/ daily level granularity
Includes downstream data (POS, Neilson, IRI, etc)
Assumes that all items have a normal distribution of demand and orderlines
Prone to overfitting, does not properly predict future behavior
Businesses have many experts with a vast array of niche information
Stored both systemically and in a business process
Loses the ability to merge promotional data, external variables, product introduction, end of life planning, and other signals in one model
Is a type of problem that machine learning can leverage - vast amounts of data and complexity which has high impact to forecast error
The future of demand planning is rapidly approaching and business needs are ever changing. Companies need to be more agile than ever.
Two different items, A and B…
Results in the same forecast in traditional systems
However, in probabilistic demand models the devil is in the details
Forecast actually changes as well….
Many small orders result in more certainty than few large orders
Cannot handle medium/slow movers at all
Requires a deep understanding of statistics and/or a PHD level education to correctly identify a method or algorithm which works for the business
“high touches” are required to keep the model running.
When planners do have to switch between algorithms because an item changes sales behavior – It adds unnecessary volatility into the model, making inventory and supply planning even more difficult because the forecast is a “moving target”
Does not scale well with a company as their complexity increases (forecasting 100 items can be trivial, forecasting 10,000 items can require a team of 20)
Weekly/ daily level granularity
Includes downstream data (POS, Neilson, IRI, etc)
Assumes that all items have a normal distribution of demand and orderlines
Prone to overfitting, does not properly predict future behavior
Businesses have many experts with a vast array of niche information
Stored both systemically and in a business process
Loses the ability to merge promotional data, external variables, product introduction, end of life planning, and other signals in one model
Is a type of problem that machine learning can leverage - vast amounts of data and complexity which has high impact to forecast error
The future of demand planning is rapidly approaching and business needs are ever changing. Companies need to be more agile than ever.
Update logo “Cortana – look for star graph
Self learning, highly autonomous
Cannot handle slow movers at all
Requires a deep understanding of statistics and/or a PHD level education to correctly identify a method which works for the business
“high touches” are required to keep the model running.
Switching between algorithms because an item changes sales behavior adds unnecessary volatility into the model, making inventory and supply planning even more difficult
Does not scale well with a company as their complexity increases (forecasting 100 items can be trivial, forecasting 10,000 items can require a team of 20)
Includes downstream data (POS, Neilson, IRI, etc)
Assumes that all items have a normal distribution of demand and orderliness
Prone to overfitting, does not properly predict future behavior
Businesses have many experts with a vast array of niche information
Stored both systemically and in a business process
Loses the ability to merge promotional data, external variables, product introduction, end of life planning, and other signals in one model
Is a type of problem that machine learning can leverage - vast amounts of data and complexity which has high impact to forecast error
The future of demand planning is rapidly approaching and business needs are ever changing. Companies need to be more agile than ever.