This document discusses different types of forecasting methods and their accuracy levels. Naive forecasting is the simplest but least accurate, assuming past demand will repeat. Statistical forecasting fits curves to historical data and incorporates seasonality and trends, improving accuracy. Demand planning further increases accuracy by using more granular downstream data and reducing effects that cause demand fluctuations. The most accurate methods are demand modeling and machine learning, which consider demand drivers and patterns from extended or big data, decreasing errors and lost sales by 33%.

1. ACCURACY
ERROR
Purely reactive
Becomes a nightmare to
manage in Excel
Leverages more granular
and downstream data to
get a cleaner demand signal
and reduce volatility and
bullwhip effect
Fits a forecast curve
through historical
demand quantities
Incorporates
seasonality, trend data,
and moving averages
Includes techniques that
are usually associated
with short-term demand
sensing to dramatically
increase long-term accuracy
Assumes last year’s
or last month’s
demand value will occur
again this month
Is often done in Excel Hierarchy and causal
effects are
incorporated into the
forecast
Statistically predicts
monthly or weekly
demand patterns
Relies on powerful
models to consider
demand drivers such
as promotional details, new
product introductions, social
media, etc.
Takes advantage of
extended and even
big data to further increase
accuracy
ToolsGroup, Inc.
Us-info@toolsgroup.com
617-263-0080 EXT.1
75 Federal St., Boston, MA 02110 www.toolsgroup.com
40%
60%
50%
50%
70%
30%
85%
15% 90%
10%
NoForecasting NaiveForecasting StatisticalForecasting DemandPlanning DemandModeling MachineLearning
Copyright © 2015 ToolsGroup. All rights reserved
Improvements in forecast are most dramatic when there is a fundamental change in the approach to forecasting
(from No Forecasting to Naive, from Statistical to Demand Planning, and from Demand Planning to Demand Modeling)
The combination of Demand Modeling and Machine
Learning will decrease errors and lost sales by 33%
THE EVOLUTION OF FORECASTING