Models

The model can help you to identify top-performing programs that should be expanded, and under-performing programs and need a 'get-well plan'. You should combine insights from the model with insights of field experts to make better decisions about programs. The model has done the deep thinking for you. As a result, the input data and interpretation of results is relatively straightforward.

The key output is an estimate of the 'contribution margin' of each marketing program. (A contribution margin is a form of profit margin where you don't have to account for all the costs in the organization, so it is not an operating margin or net income.) The main steps in the process are:

1. Allocate gross margin (revenue - cost of goods sold) to marketing programs. That is, reward programs for generating gross margin dollars, not revenue, so programs are rewarded more for selling high-margin products. Allocate gross margin from each sales order to specific Programs, Program Events, Lead Events that touch customers who place orders. This is the key step. Definitions:
• A 'Program' is a spending program, such as webinars that promote a particular product.
• A 'Program Event' is part of a Program that occurs on a particular date in a particular location for a particular product line.
• A 'Lead Event' consists of one prospective customer person who attends a Program Event.
2. Allocate expenses to Programs, Program Events, Lead Events using fixed expense and variable expense per Lead Event.
• Allocating program expense to individual programs is usually relatively easy, and data of reasonable quality and completeness is available in most organizations.
• The model allows you to distinguish fixed expense and variable expense.
  • Fixed expense does not change with the number of Lead Events (customer persons who are touched by the program). Examples: development cost of a seminar or webinar.
  • Variable expense changes with the number of Lead Events. Examples: rented space for an event, expense for staff that attend.
• Compute contribution margin (of a Program, Program Event or Lead Event) = allocated gross margin - allocated costs.

The model makes some key assumptions that improve the results.

• How a customer is defined important to the analysis.
• A customer is a decision network that collectively decides to buy a product or service.
• A customer is not generally a person who attends a program event, or who places an order, or who uses the product. (Such a definition would prevent the model from connecting a user of the product who attends a marketing event and a purchasing agent who places the order.)
• In practical terms, we suggest defining a customer to be a unique combination of company name, division, and address. This rule isn't perfect, but it is simple and usually yields good results.
• The time lag between a Lead Event and a resulting sales order at the same customer is in important factor in estimating the impact of the Lead Event on the sale.
• Each type of Program has an effective lifetime. A program gets less credit for an order as the time lag between the Lead Event and the order increases. You can specify the lifetime and how fast the program effectiveness drops off with time.
• The model allocates a fraction of total expense for a Program Event to each time period in wihwch there are some Program Events. The computation resembles depreciation of manufacturing equipment, in which depreciation expense is allocated to each time period during the life of the equipment.
• If a marketing program is not fully 'depreciated' at the end of model time, then some 'book value' remains in the program, and that value is not included in the analysis of what happened during model time.

The key results are collected on worksheet 'Contrib Margin'. Advanced versions include a pivot table of customer information.

We got great results using this method in a company with tens of thousands of marketing events annually, and dozens of products. In companies with marketing programs of this complexity, no experts or small team can carry in their heads all the knowledge that the model can capture. On the other hand, the analysis does not know all the facts that field personnel know, so you should take into consideration both the results of the analysis and expert opinions.

The model helps address two key pricing questions.

• What prices for a product line with several products will maximize its revenue?
• What prices for a product line of several products will maximize its profits? (Advanced version only)

The products can include your own products, competing products, and substitutes that are not direct competitors.

The key to estimating sales from pricing tests is to estimate a constant price elasticity for each product, and constant price cross-elasticities for each pair of products from test results.

• Price elasticity is defined by the following relationship: If I raise (lower) the price by x% (where x is small), then unit sales rise (fall) by elasticity * x%.
• Price cross-elasticity between Product A and Product B is defined by the following relationship: If I raise (lower) the price of Product B by x% (where x is small), then the unit sales of Product A rise (fall) by corss-elasticity * x%.

This model assumes that price elasticity and cross-elasticity is constant over the range of prices being considered.

The key step in computing profit margins is to estimate costs as a function of unit sales levels for the products. Costs can be cost of goods or total costs or whatever you choose.

The model offers two types of price elasticity analysis.

• Simple (or constant) price elasticity. (This is the only price elasticity method in the Standard version.) This is the standard 'textbook' model of price elasticity, characterized by the relation sales units = constant x price^elasticity, where elasticity is a constant derived from pricing test data.
• Generalize (or non-constant) price elasticity. (Advanced versions offer both elasticity methods.) This model allows the elasticity to change as a function of price. For a given set of pricing test data, this method is better at identifying prices that maximize revenue or profits.

Pricing tests proceed through the following steps.

1. Select a large market where you sell the product at a standard price, and you know how many units you sell at that price (per time period).
2. Select some smaller test markets within the large market that have similar response to price changes. The test markets should be chosen to minimize "cross-talk" between test markets about prices, and commuting between test markets to buy at a lower price. For example, test markets consisting of large cities separated by a few hundred miles generally meet these criteria.
3. Offer the product at different prices in the different test markets. You probably want to include some prices above the current standard price, and some below. Run the test for sufficient time that customers react to the test prices as if they were permanent prices. Measure the unit sales in each test market.
4. Enter the following data into the price testing template.
• The price and sales units for the entire market that prevailed before the pricing test
• The number of units sold in each test market before the pricing test
• The price and number of units sold in each test market during the pricing test

From this data the model estimates the price sensitivity (price elasticity) of the total market. It then estimates the revenue and sales units (and profit margin if that is included in your model) for the entire market at different prices that you specify.

The model has several other key features.

• The model provides some standard statistical measures of goodness of fit of the model and the market test data.
• The model reports r^2 ('r squared') for each product and pair of products. These measure the fraction of the variance in sales units for a product that is accounted for the change in price of that product or another product, and by the model. r^2 takes on values between zero and one. A value over 0.8 means that the constant price elasticity method accounts for nearly all of the variation in unit sales with price. A value below 0.2 means that other factors besides a constant price elasticity are needed to explain well the variations in sales units across test markets.
• The model reports a 'standard error of elasticity' for each product and each pair of products. These measure the likely range of price elasticity (and cross-elasticity of two products) that are reasonably consistent with the data.
• The template contains comments on factors that help to ensure that market tests yield valid conclusions about pricing behavior.

The current model includes these features. (For a simpler sales plan, see "Sales Plan for a Small Startup".)

• Uses linear regression on sales unit history to extract trends.
• Limits positive and negative growth rates, so that small and unstable segments do not blow up the overall plan.
• New products start their trend lines at product introduction dates.
• Forecasts revenue for product x location x channel x industry segments, using unit sales plan, list prices, and price discount percentages.
• Managers' revenue targets for various segments override historical trends so human judgment controls the plan.
• The model computes suggested adjustment factors to raw regression forecast to make the final forecast closely match managers' targets.
• If managers' targets are inconsistent (e.g. product managers targets canot be reconciled with channel managers' targets), the planner can choose which targets to emphasize.
• Includes adjustment factors for product x location x sales channel x industry segments.
• Forecasts actual average selling prices by segment from list prices and actual price discount history.
• When time grain is Quarters, the model extracts seasonality from historical data and applies it to the plan.
• Estimates growth trends free of seasonality.
• Estimates seasonal effects using regression methods.
• Adds historic seasonality back into forecast.
• Provides trailing-4-quarter history and plan, to eliminate seasonality and reduce effects of random large orders.
• Optionally uses local currency along with global currency - for now, Euros in the Euro Zone.
• Optionally accepts historical input data for revenue, sales units, prices (and variable costs if you select the margin forecasting feature) from databases.

Not all features are included in the Standard version.