The Right Frame of Mind

Understanding analytic frameworks is necessary

By Erik Thomsen

Welcome to my new column, which will show you how to solve different types of decision-oriented problems by building the appropriate type of analytic model out of reusable, analytic building blocks. The building blocks comprise repeatable calculations and decision-oriented schemata or frameworks. Becoming more familiar with these analytic building blocks will make creating or defining the requirements for appropriate solutions for your organization much easier. Because no standard dimensional language exists, I will also introduce, where appropriate, pieces of an analytic language that I created to reason about calculations and schemata.

For this first installment, I will introduce four linked themes that will run throughout the column:

• Good decisions are made within appropriate analytic frameworks.
• Single analytic frameworks often require information from multiple sources.
• The required information is often derived.
• Myriad derivations can be created from the same set of repeatable calculations.

1. Decisions are made within particular analytic frameworks. Consider for a minute the abstractions you use on a regular basis in order to reason, make decisions, or otherwise operate in the world. Everybody thinks in terms of objects and processes. Furthermore, when engaged in problem solving, you are not just thinking about the existence of an object, but about the movement, composition, or attributes of that object or process, as well as relationships between collections of objects and processes.

These abstract, analytic frameworks form the basis for day-to-day problem solving. Depending on the problem, you may reason in terms of object attributes, process flow, or object motion. Whereas in day-to-day living you unconsciously select appropriate analytic frameworks, in business modeling you need to consciously select them. What follows is a description of several common analytic frameworks.

Paths. If you have ever followed a golf ball or baseball after it's hit, you are capable of making decisions in terms of paths. Paths trace the motion of objects. They have direction and speed, and their history traces a sequence of positions. You may need to predict where a path will lead, decide whether to alter a path, or simply study and compare different paths.

In the business world, if you're in charge of your company's e-sales, you must look at the paths made by visitors on your Web site. This scrutiny lets you predict which paths are likely to lead to abandoned carts; you can then either intervene with an offer before they leave or rearrange or redesign the relevant pages. If you're in charge of the internal network for a large corporation, you may study employee communication paths to determine if a different network topology might reduce bottlenecks.

Zones. If you have ever looked at cities, lakes, and countryside from an airplane, you are capable of making decisions in terms of zones. Zones are regions defined by boundaries. They can have different shapes, be embedded within other zones, or overlap. Zones can expand and contract, and they can do so in distinct areas or subzones. You may need to determine the boundary location between zones, predict whether a zone will expand over time, define the attributes of a zone, or simply measure its surface area.

For example, if you're the marketing manager for a catalog-based company, you may need to find the optimum demographic zones for targeted promotions.

States. When you're studying molecules as they transition from solid to liquid to gas or watching the NASDAQ as it moves into bear territory, you are thinking and capable of making decisions in terms of state changes. States are very general and may be attributed to any object or process. States may be defined in terms of the values of a single attribute or in terms of value ranges for a collection of attributes. Many behavioral properties follow from the state of an object - water expands when it turns to ice; couples buy a house after they get married. The intent of many state analyses is to either predict when an object is likely to change states or simply recognize that an object is in some state.

For example, if you're in charge of marketing, you may want to analyze why certain customers who were in a profitable state are now losing you money. If you are in charge of sales, you may try to find all customers who have recently changed state, such as from single to married, because such state changes are indicative of new purchasing patterns.

Transformations. If you've ever melted chocolate in a double boiler, warmed butter on the counter, stirred cream of tartar into beaten egg whites for a souffle, or wrestled with lumpy yarn as you knit a sweater, you're capable of making decisions in terms of transformation models. Transformations are operations performed on some material(s) yielding some other material(s) as a result. Basic operations include assembly (automotive or salads) and alteration (nuclear explosions or diamond formation). Transformations have inputs, transformation rates, possible intermediate results, and outputs (or simply throughputs).

For example, if you're in charge of production for a paper mill, you may need to analyze input quality, paper yields, quality of paper produced, and factors that most influence production costs.

Paths, zones, states, and transformations are four key analytic frameworks or schemata for thinking and making decisions. In subsequent installments, I will show you how and when to build each type (or mixed type) of analytic framework.

2. Single frameworks often require information from multiple sources. Analytic frameworks pull together information from a variety of sources. Consider the information required for a customer value state model that will assess the state of any particular customer, predict which customers are increasing or decreasing in value, and provide the basis for deciding how to treat any particular customer. You would need the following information:

• Sales: what was sold to the customer and at what price
• Marketing: what resources your company expended to attract the customer
• Distribution: the cost to get the product to this particular customer
• Inventory: whether any product is on hand when the customer asks for it
• Production: the cost to rush an order or meet a customer's special production requests
• Purchasing: the cost to meet a customer's requests for special inputs
• G&A: the cost to research whether customers' requests are permitted in their states.

In other words, a high-quality analytic framework for solving customer problems requires information from across the entire organization.

3. Required information is frequently derived. Not only does a good customer-oriented, analytic framework use information from around the organization, but most of that information was itself derived. For example, in addition to basic aggregations that apply to almost all information sources, sales information may need to undergo currency conversion, nominal dollar conversion, or unit conversion. You may need to allocate marketing costs by channel, region, time, and product. You may need to calculate distribution costs from container and segment costs, including amortization of relevant assets. Inventory costs may need to be calculated by taking into account losses, aging, and FIFO or LIFO accounting practices. Production information needs to reflect realtime activity-based costs. Purchasing costs need to reflect lifetime supplier value in addition to marginal materials cost. And G&A needs to reflect the opportunity costs of the persons handling the customer requests. In short, derivations may need to be created from all sources and combinations of sources to support a high-quality, analytic framework.

4. Myriad derivations are created from repetitions of the same types of calculations. Luckily, most derivations can be created from iterations of the same set of core calculation building blocks. I like to think in terms of calculation types and flavors. The main types of calculations are aggregation and allocation, ratio and product, ranking and randomizing, and links. The main flavors are simple, weighted, conditional, and overridden. For example, a cost comparison calculation may involve the ratio of the top 10 rankings of a weighted sum of a conditional product.

In subsequent installments, I will show how to define calculation building blocks and integrate information from multiple sources, within appropriately selected and defined analytic frameworks to create high-quality analytic solutions.