Demand in Business Forecasting What s missing is often

Demand in Business Forecasting “What’s 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.

Traditional forecasting process • The sales manager asks sales reps 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 year’s 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.

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.

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.

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; but—may 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.

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 minimarkets that each require analysis.

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.

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 passes—were 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.

One Company’s Application of Data Schwan Food— 6, 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.

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.

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 areas—different factors are related to bottled water sales rates.

ü Google Maps for Inside the Store ü Mobile Integration ü Shopping List ü Checkout using Smartphone's ü Directions using Smartphone's ü Using Security Camera’s in the store ü Goal is to be Local in a Global Market place ü Demographic Data ü Purchasing Pattern’s ü Purchasing Reason’s ü Impact on Demand. ü Increase Sales at Shelf. ü Weather Information CONTAINS CONFIDENTIAL & U. S. PATENT PROTECTED INFORMATION PAGE 12

Goal ü Identify Shrink ü Identify Phantom Inventory • • POG, shelf allocation & display compliance Standards compliance • • Promo & trade spend compliance Full category/competitor visibility • Extensive Reporting ü SKU & custom groups ü Product on display ü Quantity on display ü Store location ü Retail price ü POS accuracy ü Asset placement & tracking • Stores linked to reporting hierarchy CONTAINS CONFIDENTIAL & U. S. PATENT PROTECTED INFORMATION PAGE 13

After Hurricane Katrina: Drinking Water, Batteries, Cleaning Supplies, Ready-to-Eat Food

Amazon VP of Digital Video and Music: “we let the data drive what to put in front of customers; we don’t 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. com—better algorithms for matching men and women.

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

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. Nordstrom’s attributed much of its increase in profit margin from 5. 2% to 10. 6% in two years to impact of the software.

Rapidly Changing World Harrah’s casino was second rate. New CEO made it first tier as Caesar’s. 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 (Microsoftware architect): Objective data are replacing Hi. PPOs (Highest Paid Person’s Opinions) as the basis for decision making—better cost control and better customer service. For many companies data doubles every year now.