g a r t n e r g r o u p
Integrated Logistics
Strategies ILS Strategic Planning,
SPA-500-075
B. Enslow Research Note
July 23, 1996
REVISION Forecasting Systems: Emerging
Functionality, Part 1
We outline the methodological advances that will occur
in demand forecasting systems during the next five years.
Core Topic
Applications Cross-Industry: Demand and Deployment
Planning Strategies and Systems
Key Issue
How will demand and deployment functionality change
during the next five years?
Strategic Planning Assumptions
Causal forecasting for demand planning will be adopted
by 75 percent of Type A (leading-edge) enterprises by 2001
(0.7 probability). Enterprises that use forecasting systems
that automatically determine the correct model form will
sustain a competitive advantage in demand planning through
2001 (0.7 probability).
Systems robust in forecasting all major types of demand
patterns will emerge by 2000 (0.8 probability).
Demand forecasting systems will undergo radical change
during the next five years. Enterprises that fail to keep
pace will find their business models inadequate, and they
will fall behind enterprises whose forecasting systems permit
them to react quickly to changes in demand (see Note 1).
Leading vendors will improve their demand forecasting systems
methodology and functionality. Here, we summarize the
critical methodological requirements for forecasting systems
during the next five years. In ILS Research Note
SPA-500-076, July 23, 1996, we summarize the critical
functional requirements of the products. Requirement:
Generate forecasts using historical data and causal data.
Today, forecasting systems range widely in their ability to
automatically spot seasonality, cyclical and general trends
in historical data and to deal with unusual values. This
ability is becoming more important as enterprises move to
forecasting at the SKU level and begin using point-of-sale
(POS) data. (During the next five years, enterprises will
increasingly use POS data because it closes the supply-chain
information loop and, by increasing data timeliness, improves
forecast accuracy.) Neural networks and expert systems are
being embedded in forecasting systems to address this need.
Increasingly, forecasting systems must also be able to
incorporate causal factors into the forecast, such as sales
of related products (e.g., shampoo and conditioner),
cannibalization effects and even the price of competitors
products. These systems must also account for events like
advertising and promotions, including lag effects (e.g.,
reduced sales the week after a promotion ends). Automatic
Forecasting System, Comshare, Manugistics and Neil Thall
Associates are among the vendors developing causal
forecasting techniques. Causal forecasting for demand
planning will be adopted by 75 percent of Type A enterprises
(aggressive adopters of technology) by 2001 (O.7
probability).
Requirement: Automatically determine the optimal model
form. The preset or "pick best" models found in many
forecasting systems today (e.g., Think Systems and LPA
Software) produce suboptimal forecasts. To produce
more-accurate forecasts, the system needs to automatically
tailor the model to the problem at hand, including selecting
the best lead and lag structures for each input series and
the best weightings. It needs to correct for omitted
variables (e.g., holidays or price changes that have affected
the historical data, but that the system has no knowledge of)
by identifying pulses, seasonal pulses, level shifts and
local time trends, and then adding the needed structure
through surrogate variables. Conversely, it needs to
eliminate unneeded structure (e.g., a statistically
unimportant causal variable) to keep the model manageable.
It should perform all these functions as part of its normal
routine without human intervention. It should also report
the statistical tests used to determine the model parameters,
and let users manipulate the coefficients and model structure
if they want. Manugistics and Automatic Forecasting Systems
are working on such methods. Enterprises that use
forecasting systems that automatically determine the correct
model form will sustain a competitive advantage in demand
planning through 2001 (0.7 probability).
Note 1
The Goals of Forecasting
Forecasting s role is to predict how much product will
be needed, when and where. Forecasts are typically divided
into three time frames long range, intermediate and short
range.
Long Range: Long-range forecasts (e.g., more than two
years) are used to decide whether to enter new markets,
develop new products or services, expand or create new
facilities (including plants and warehouses), or arrange
long-term procurement contracts.
Intermediate: Intermediate forecasts (e.g., three
months to two years) are used by: 1) finance, for budgetary
planning and cost control; 2) marketing, for new product
planning and sales force compensation plans; 3) operations,
for facility planning, capacity planning and process
selection; and 4) logistics, for warehouse and distribution
planning (e.g., transportation contracts).
Short Range: Short-range or operational forecasts
(e.g., fewer than three months) are used to make continual
decisions about planning, scheduling, inventory and staffing
in an enterprise s production, procurement and logistics
activities. These decisions usually involve scheduling
shipments and material flow through an enterprise s
facilities and on to the customer.
Requirement: Address special forecasting needs.
Depending on the enterprise, forecasting applications may
need to predict product and market life cycles, spare-part
demand, product returns, intermittent demand, seasonal
products and new products. For example, a leading system
will let users forecast demand for a new product by selecting
sales histories for several similar products and combining
them to create the forecast for the new product (e.g., the
new product may be given the geographical demand pattern of
one historical product and the seasonality of another).
Systems robust in forecasting all major types of demand
patterns will emerge by 2000 (0.8 probability).
Bottom Line: During the next five years, business
demands for accuracy and customer responsiveness will mandate
new requirements for forecasting systems. Increasingly,
enterprises must: 1) more accurately forecast at the point
of demand, 2) incorporate causal factors like promotions and
price into the forecast methodology, and 3) handle
specialized demand patterns. Incorporate these requirements
when evaluating forecasting products, and analyze the
business value that improved forecasting capabilities will
have on sales, marketing, manufacturing and logistics
activities.
Copyright 1996 ILS: SPA-500-075
July 23, 1996
Integrated Logistics Strategies ILS
Strategic Planning, SPA-500-076
B. Enslow Research Note
July 23, 1996
REVISION Forecasting Systems: Emerging
Functionality, Part 2
We outline the functionality that will be emerging in
demand forecasting systems during the next five years.
Core Topic
Applications: Cross-Industry: Demand and Deployment
Planning Strategies and Systems
Key Issue
How will demand and deployment functionality change
during the next five years?
Strategic Planning Assumptions
All major demand planning systems will integrate with
multidimensional databases or have them embedded by the end
of 1998 (0.8 probability). True enterprisewide forecasting
systems will emerge in 1999 (0.7 probability). Data handling
and automation features rather than forecast accuracy will be
the key differentiators for replenishment planning
applications through 2000 (0.7 probability).
The simulation of demand will become a standard category
management activity for Type A enterprises (aggressive
adopters of technology) by 2000 (0.8 probability). Mass
customization, vendor-managed inventory (VMI) and other
business initiatives are pushing many enterprises demand
forecasting systems to the breaking point. Leading
enterprises are evaluating products that not only alleviate
this, but that also let the demand plan be leveraged
throughout the enterprise. Here, we summarize the critical
functionality requirements for forecasting systems during the
next five years.
In "Forecasting Systems" Emerging Functionality, Part 1,
we summarize the critical methodological requirements.
Requirement:
Multidimensional capabilities. As enterprises move to
forecasting at the SKU level, handling data becomes a
challenge. To optimize speed and performance, demand
forecasting at the hundreds-of-thousands-of-SKUs level should
be done using online analytical processing (OLAP) with data
stored in a multidimensional database, not a relational
database. This will enable the system to arrange information
in multiple levels or definitions (e.g., by product line,
SKU, customer, geography and time) so users throughout the
enterprise can slice and dice the forecast to make effective
plans. Many traditional forecasting systems (e.g.,
Manugistics and American Software) still rely on relational
databases. Products from Think Systems and Comshare work off
multidimensional databases. All major demand planning
systems will integrate with multidimensional databases or
have them embedded by the end of 1998 (0.8 probability).
Note 1
Change Propagation
Automatic top-down, bottom-up and middle-out change
propagation lets users make adjustments to a forecast at any
level of the database. For instance, a forecast can be
developed at the pack level (e.g., 16 ounces of diet soda)
and be aggregated to total soda sales and disaggregated to
all component SKUs (e.g., open stock and floor displays).
Similarly, top management can mandate, say, a 10 percent
sales increase for Sales Region C and the changes will be
propagated accordingly.
Requirement: Useful for multiple levels of the
enterprise. As enterprises move to cross-functional
forecasting teams and single-number forecasts, forecasting
packages must be flexible enough to meet the needs of
everyone from the regional sales manager to the VMI planner
to the procurement manager. Among the necessary features:
1) automatic top-down, bottom-up and middle-out change
propagation (see Note 1); 2) "bookmarks" for customizing
views; 3) intuitive user interfaces; and 4) easy-to-use tools
for forecast overrides and clear documentation of overrides.
Systems from American Software, Comshare, Manugistics and
Think Systems are useful for pieces of the enterprise, but no
vendor has a product that enables people throughout the
enterprise. True enterprisewide forecasting systems will
emerge in 1999 (0.7 probability).
Requirement: Is automated and includes an exceptions
feature. To reduce the head count needed to create the
forecast, the system must process the forecasts
automatically, including data input and forecast generation.
However, the system must produce alerts or exception reports
when odd-looking data (e.g., outliers) are encountered, and
it must allow users to manipulate the forecast models when
desired. Automation is becoming increasingly important as
enterprises adopt VMI and want to do weekly SKU forecasting
at the customer level. Data handling and automation features
rather than forecast accuracy will be the key
differentiators for replenishment planning applications
through 2000 (0.7 probability).
Requirement: Contain advanced simulation capabilities.
Simulation lets users store forecasts and do "what if?"
analysis to determine, say, how a price change will affect
the flow of goods through a distribution center. Advanced
simulation lets a vice-president of sales use drag-and-drop
technology to add a promotion and then graphically see the
effect on the sales forecast at the national, regional and
local levels. Comshare, Manugistics and Think Systems are
refining their products simulation capabilities, though the
methods of determining incremental lift remain relatively
crude. The simulation of demand will become a standard
category management activity for Type A enterprises
(aggressive adopters of technology) by 2000 (0.8
probability).
Bottom Line: During the next five years, advances in
forecasting products will enable multiple parties (e.g.,
sales, marketing, manufacturing and logistics) to enhance,
manipulate and use the forecast. Clients facing pressure
from their customers to implement VMI and do more-effective
category management should pay close attention to the
emerging functional enhancements in demand forecasting
systems.