Demand in Business Forecasting

1
Demand in Business Forecasting

• Whats missing is often good measurement and a
  commitment to follow the data. We can do better.
  We have the tools at hand.
• Bill Gates
• Use of the law of demand is not simple or we
  would not be here. Successful application can
  have a significant impact on profits.

2
Traditional forecasting process

• The sales manager asks sales reps for forecast.
• The reps make guesses for next year, then
  subtract 10.
• The sales manager takes the forecasts and raises
  them because he knows the reps are fudging.
• The sales manager gives the forecast to top
  management which changes the numbers to match
  analyst expectations.
• Manufacturing ignores the numbers and orders raw
  materials based on last years actual sales.
• Actual sales may bear no relation to any of the
  above.
• Prior to earnings announcements, change the books
  so that they resemble the forecast.

3
Better forecasting process

• Build a model that predicts future buying
  behavior based upon previous years sales,
  seasonal changes in buying patterns, historical
  impact of marketing campaigns, overall state of
  the economy, fluctuations in currency exchange
  rates, and other relevant factors.
• Test the model against historical data to confirm
  that if it had been in place in the past if it
  would have predicted sales.
• Goal sales and marketing on providing information
  that hones the accuracy of the model.

4
Why Demand Is Not Easy to Measure

• Changes in the design of products and entry of
  new products mean limited lifecycles. Changes
  make forecasting more difficult because you use
  data from previous products and time periods.
• Study from Chicago and Columbia Business Schools
• 40 of household expenditures are on goods
  created in the last 4 years and 20 of
  expenditures are on goods that disappear in the
  next couple years. That is, in many markets there
  is rapid product entry and exit.

5
Accurate Measures Difficult

• Difficulty in interpretation of historical data
  Sales, orders, shipments and invoices are
  historical data that can cause confusion.
• This may be innocent butmay be evidence of
  theft, evidence of bad record keeping, or other
  internal problem.
• Besides trying to measure demand also an
  opportunity to understand company costs and
  operations better.

6
One Size May Not Fit All

• Expanding business to new markets means new
  demographics (customer base), facing new
  competitors, different seasonal factors,
  packaging requirements, and distributional
  channels. Seller of a product is likely really in
  multiple mini-markets that each require analysis.

7
Measurement Difficulties, But Big Benefits

• Demand usually underestimated. Were sales actual
  demand or did stock run out, thereby cutting
  sales?
• Most data is old look for more real time data.
  YRC Worldwide (transport) quickly reduces its
  fleet and employee base when it sees shipments
  shrinking. Checks weight and frequency of
  shipments to look for changes in industry
  conditions.
• HP works with Walmart to forecast PC demand. By
  getting earlier orders, HP saves on manufacturing
  costs, which are lower if ordered far in advance.
  Walmart gets better price.

8
Demand Forecasting

• Forecasts are statistical estimates for the
  future. They can be improved by determining
  probability distributions for demand points by
  location and for specific times.
• How much did actual demand deviates from prior
  demand forecasts? Improve the model. Revision a
  good idea as time passeswere the estimates made
  six months ago for next year still the best
  estimate?
• Based on experimenting with data, determine
  relevant time period. Example Anheuser-Busch
  uses five-year historical data to better
  understand product lifecycles and seasonal
  demand.

9
One Companys Application of Data

• Schwan Food6,000 sales reps deliver frozen foods
  to 3 m. customers at home. They looked at 6 weeks
  of orders to decide what to suggest to customers.
  Sales flat for years.
• More sophisticated match customers buying
  patterns offer new products and discounts via
  hand held devices used by reps. Revenues up 3-4
  because understand demand better.

10
Improving Demand Measures

• More measures of possibly relevant factors
  competitor prices, regional events, demographics,
  and weather. Some of this information is low
  cost.
• Use info from bar codes RFID chips.
• Think of how to use new information sources, such
  as social media. Insurance companies beginning to
  exploit life style information revealed in
  Facebook and such.

11
Data Shows Relationship between Weather and Sales

• Bottled water sales rise in Los Angeles in the
  winter when temperatures are below average and
  there is less wind than usual.
• Bottled water sales rise in Los Angeles in the
  fall when temperatures are above average and
  there is less wind than usual.
• In other areasdifferent factors are related to
  bottled water sales rates.

12

• Google Maps for Inside the Store
• Mobile Integration
• Shopping List
• Checkout using Smartphone's
• Directions using Smartphone's
• Using Security Cameras in the store

• Goal is to be Local in a Global Market place
• Demographic Data
• Purchasing Patterns
• Purchasing Reasons
• Impact on Demand.
• Increase Sales at Shelf.
• Weather Information

13

• Goal
• Identify Shrink
• Identify Phantom Inventory

• POG, shelf allocation display compliance
• Standards compliance
• Promo trade spend compliance
• Full category/competitor visibility
• Extensive Reporting
• SKU custom groups
• Product on display
• Quantity on display
• Store location
• Retail price
• POS accuracy
• Asset placement tracking
• Stores linked to reporting hierarchy

14
After Hurricane Katrina Drinking Water,
Batteries, Cleaning Supplies, Ready-to-Eat Food
15
Amazon VP of Digital Video and Music we let the
data drive what to put in front of customers we
dont have tastemakers deciding what our
customers should read, listen to and watch.

• Best Buy gets data on multiple offerings of
  products who is buying and using electronics?
• What they learned Many DVD players bought for
  young children. A store-brand with rubberized
  edges and spill resistant became good seller.
• Private label models do fine if have special
  features.
• Match.combetter algorithms for matching men and
  women.

16
Wide Range of Applications

• Pricing restaurant meals an drinks drinks have
  higher margins. Experiment with changing the mix
  of these services.
• Inventory controlWalgreen cut number of products
  carried about 20. Eliminate low value goods
  focus on profitable goods.
• Amazon runs many A-B experimentstwo versions of
  websites appear to matched sets of customers to
  see reactions.
• Google runs 100 experiments a day.

17
Successful Practice

• One form of this is in price-optimization
  software that looks to past sales to determine
  where to set initial prices today and when to
  begin to discount.
• This helps avoid panic discounting if initial
  sales are weaker than expected.
• Nordstroms attributed much of its increase in
  profit margin from 5.2 to 10.6 in two years to
  impact of the software.

18
Rapidly Changing World

• Harrahs casino was second rate. New CEO made it
  first tier as Caesars. Focus on data about
  customers from Total Rewards loyalty cards.
  Tested new promotions, price points, services,
  workflows, employee incentive plans and casino
  layouts. Let the customers tell you want they
  want.
• Ron Kohavi (Microsoft software architect)
  Objective data are replacing HiPPOs (Highest Paid
  Persons Opinions) as the basis for decision
  makingbetter cost control and better customer
  service.
• For many companies data doubles every year now.