Forecasting Demand

 

 

What is a forecast?

·         An estimate of future demand

Forecast Error

·         Difference between actual demand and forecast demand

·         Note:  it’s very important to monitor the accuracy of the forecast consistently and establish targets to improve accuracy

 

Why a Forecast?

·         Plan for future/reduce uncertainty

·         Manage/anticipate change

·         Increase communication between planning teams

·          Manage lead times/anticipate capacity and inventory demands

·         Project operations costs into budget

·         Improve productivity/competitiveness through decreased costs, improved delivery times, and responsiveness to customers.

 

Areas Impacted by the Forecast

·         Investment/capital equipment decisions

·         Inventory levels – desired level of customer service and safety stock buffers

·         Capacity

·         Budgets

·         Lead-time management

 

Forecast System Design

·         What information needs to be forecasted?  Define source of data and periods of data collection

·         Who is the person accountable for forecast accuracy?  There performance = accuracy of forecast

·         Forecast parameters

o   Forecast horizon

o   Forecast level of detail (product family? Subfamily? Model? SKU?)

o   Forecast periods

o   Forecast frequency (how often is it formally reviewed and revised?)

o   Forecast revision (how will revisions be recorded and tracked?)

·         Forecasting models/techniques based on volatility of demand

·         Collection of data for input into the forecast models

·         Test scenarios/models for accuracy

·         Record actual demand compared to forecasted demand

·         Report forecast accuracy

·         Root cause analysis on the forecast errors

·         Review forecasting system consistently – how is the forecast system performing?  Does it need to be changed?

 

Forecast Techniques

·         Qualitative:  estimated based on expert or informed opinions of future product demand.  Newly introduced products, derived from focus groups, research, surveys, etc…

Types:

o   Expert Opinion

o   Market Research:  usually conducted through surveys, need a sufficient sample size

o   Focus Groups:  group of customers who are asked to provide their opinion about a product/service.

o   Historical Analogy:  comparing the potential sales of a new product with the historical sales of a similar product/service.

o   Delphi Method:  specific question is asked to a group of experts about the future such as “what types of fuel will be used in 10 years to power vehicles?”

o   Panel Consensus:  group of people share opinions about the future and a facilitator brings the group to consensus.  The thought is that the group makes better decisions than individuals.

·         Quantitative:  mathematical formulas extrapolate historical demand into potential future demand, based on the idea that historical demand is a good indicator of future demand. 

Types:

o   Moving Average: 

§  Average of a certain number (n) of the most recent observations.  As each new observation is added, the oldest observation is dropped. 

§  Pros:

·         Fast and Easy

·         Filters out random variation

·         Longer periods = more smoothing out of demand

§  Cons:

·         Hard to see trends

·         Lags trends

o   Exponential Smoothing

§  Pros:

·         Good for fairly constant demand items

·         Satisfactory for short-term forecasts

§  Cons:

·         Lags trends

§  Smoothing Factor – a.k.a. Alpha (α)

·         Determines weight of historical data on projection

·         Sets responsiveness to changes in demand

·         clip_image002

 

·          

·          

·          

·          

 

·         Closer α is to 1 = more responsive the forecast will be to the latest demand value

·         Closer α is to 0 = the less responsive it is.

·         clip_image004

 

o   Regression Analysis

o   Adaptive Smoothing

o   Graphical Methods

o   Econometric Modeling

o   Life-Cycle Modeling

Forecast Data Methods

·         Intrinsic:  based on past demand, from data within the company

·         Extrinsic:  from data outside the company – e.g. research of economic and market trends

 

Demand:  A want or need for a product

·         Sources of Demand

o   Consumers – ultimate users

o   Customers – pay for the product

o   Referrers – recommend products to others

o   Dealers/Distributors – resells product to others

o   Intercompany – purchase product from a business unit within the company (e.g. between plants)

o   Service Parts

 

Independent demand:

·         Demand for an item that is not related to another item

·         Finished Goods for the customer typically

Dependant demand:

·         Demand for items used to make independent demand items

·         On the bill of materials, Semi-Finished

·         Always calculated from finished good requirement, NEVER forecasted

*I’ve actually been fed forecast for dependant demand items…large corporations aren’t perfect.  Learn this stuff and make some changes in your organization.

 

Influencing factors of Demand:

·         Internal:  Within the company itself

o   Promotions for products:  special packaging, special pricing, special location on a shelf/in a store, etc…

o   Substitution for products:  A customer can get by with a similar product if the original one is in short supply.

·         External:  Outside the company

o   Seasonality:  winter clothes, back to school supplies, sunblock

§  clip_image006

o   Trends

o   Economic Cycles:  recession, booms, depressions, inflation

§  Good economic cycle = customers spend more money on purchases

§  Bad economic cycle = customers spend less money on purchases

o   Preference changes of the customer:  monthly instead of weekly shipments, higher quantity orders

o   Random Fluctuation:  factor is unknown that causes the randomness

 

 

 

Pyramid Forecasting

·         Product/Service Hierarchy from session 1

o   The most effective level to forecast:  Product family  or product subfamily level

·         Roll-Up

o   Two Products X and Y

o   Product level forecast

§  X Unit – 5,000

§  Price – $10

§  Y Unit – 3,000

§  Price – $20

§  Total Units in family forecasted = 8,000 (the manager makes qualitative decision to increase, or “Roll-Up” this forecast to 9,000 units)

§  Average Price = $15

§  Family-Adjusted Forecast = 9,000 units

 

·         Force-Down

clip_image008

Product X

clip_image010

 

Internal (Intrinsic) Data Sources/Factors

·         Product life-cycle management

o   4 Stages

1.       Introductory phase:  Usually no historical data; Qualitative methods used

2.       Growth phase:  Mix of Qualitative and Quantitative used

3.       Maturity phase:  Established product; Quantitative methods based on history used (moving average and exponential smoothing with adjustments for seasonality).

4.       Decline phase:  Decreasing demand, End-of-Life decisions are made.

·         Planned Price Changes:  Customers will buy more (higher demand) right before price increase.

·         Changes in Sales Force:  Hire more sales people = more sales or higher demand (if market isn’t saturated)

·         Resource Constraints:  Periods of constrained output in the past must be taken into account when creating a forecast.

·         Marketing and Sales Promotions:  Promotion decrease prices = higher demand; purpose – gain market share; artificial/transitory demand.

·         Advertising:  More advertising = higher demand

 

External (Extrinsic) Data Sources/Factors

·         Competition:  more/better competitors will decrease sales/demand for your product.

·         New Customers

·         Plans of major customers:  e.g. new store openings, new offerings.

·         Government policies

·         Regulatory concerns

·         Economic conditions

·         Environmental issues

·         Weather conditions

·         Global trends

 

Leading Indicators

·         Indicators or Causal Factors that influence the demand or volume purchase of a particular product or product family

o   Schools built influences the desk volume and dry erase board installations

o   Death rate influences funeral bouquets

 

New Product Introduction

·         Calculated risk

·         Has potential to be the next

o   Blockbuster

o   Lifesaver

o   Money loser

o   Disaster

o   Liability nightmare

 

Product Life Cycle

 

clip_image012

·         Length of each stage depends on:

o   Public acceptance

o   Social and economic conditions

o   Rate of competing products development

o   Rate of innovation

 

Focus Forecasting

·         Assumptions

o   One forecasting model is better than others

o   The most recent past is best indicator for future sales

·         Methods

o   All forecasting models for all items compared to recent sales history

o   Model selected with closest fit to actual sales

o   This model will be used to forecast this item this time

o   Next time, a different model may be selected

 

Data Issues for Forecasting

·         Availability

·         Consistency

·         Amount of Data History

·         Forecast frequency

·         Model reevaluation frequency

·         Cost & Time issues

·         Recording of true demand

·         Order date vs. ship date

·         Product units vs. financial units

·         Level of aggregation

·         Customer Partnering

 

Planning Horizon and Time Periods

·         Calendar:  decide how reporting will be done – every month, periods of four weeks each no matter the month, number of manufacturing days in a month/quarter.

·         Time periods and planning intervals: 

o   Short term forecast by Day? Week? Month?

o   Long term forecast by Years?  Quarters?

·         Planning Horizons

o   Planning Horizon = length of the period X number of periods

 

Data Preparation and Collection

·         Same Terms as forecast – Collected in same terms as the forecast (weekly, daily, monthly, etc…)

·         Actual sales NOT just shipments – we need to know how much our customer requested (a.k.a. DEMAND) not how much we were able to produce based on constraints.

·         Lost sales – record sales that would have been made had we had the product been available at the needed time.

·         Customer request date NOT the promised date – customer could have decide to cancel the order if we can’t meet their date

·         Separate demand recorded for unique customer segments

·         Exceptional Demand circumstances

 

 

Dealing with Outliers

·         Point of data differing extremely out of the norm that will distort the good data.

 

Decomposition of Data

·         Method of forecasting

o   Data separated into 4 components:

1.       Trend

2.       Seasonal

3.       Cyclical

4.       Random

o   Project the patterns individually then combining them to create a forecast.  Steps

1.       Purify Data

2.       Adjust Data

3.       Take out baseline and components

4.       Identify Demand Components (Trend, Seasonal, Cyclical, and Random)

5.       Measure random error

6.       Project Series

7.       Recompose

One Comment (+add yours?)

  1. Chon Swenson
    Sep 04, 2014 @ 01:46:50

    How can we collect or have a better estimation of ‘historical demand’?

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *