Forecasting a Flap

There's a chaos-theory concept that refers to how a small, seemingly unimportant action in one part of the universe creates massive change in another part of the universe. The idea is this: Does the flap of a butterfly's wings in Brazil set off a tornado in Texas? The thinking is that the infinitesimal movement of air from the flap starts a reaction that ends with something in another part of the world being affected by the wing's movement. Coined to help illustrate the relationship between two seemingly unrelated variables, the idea has lately become a way to describe the often complex relationships that exists between consumers and the marketplace. For years, companies have sought to understand the relationship between marketplace factors and their customer bases' propensity to do business with them. Fortunately for businesses, the development of metrics applications over the past three decades has provided forecasters with the tools to take much of the mess out of figuring out the marketplace. These tools are forecasting and econometrics apps. The fiercely competitive nature of today's business landscape is forcing organizations to operate more efficiently. Years ago the data to segment customers was often inaccurate and the ability to pull results from econometric systems on a daily basis was virtually nonexistent. Companies are now looking to the past to predict the future by using econometric and forecasting solutions to measure the impact of economic and marketplace factors. The emphasis is on automation and simplification--it's becoming more about the pull and less about the push. Results are being drawn from forecasting and econometric applications on a monthly, weekly, or even daily basis, as opposed to being pushed to a department by management once a quarter. Mark Lush, a principal within Deloitte Consulting and coleader of the national marketing solutions practice, states his own, marketing-influenced version of the aforementioned scientific rubrik: "The butterfly flaps his wings in Beijing and suddenly lipstick sales go up in San Francisco. Econometrics is a classic example of the chaos theory. A company's customer base can be divided into literally thousands of segments. Forecasting and econometric tools identify the relationships and linkages between those segments at the highest level--they connect the dots." Econometrics for the Nincompoop? Just like their CRM analytical brethren (and sistren), the big push with econometrics today is on making these solutions more innate by automating the selection and customization processes that come with operating these apps. To do their job econometrics solutions use regression and time series analysis to model the relationships between groups of variables, determine the magnitudes of these relationships, and make predictions based on those models. The models by which these solutions slice and dice the data are by no means elementary. They involve heavy doses of calculus and a firm understanding on the part of the operator of both mathematical and economic theory. While too complex to go into at length, the models themselves are data cubes, virtual-3D megacubes of customer data. "We're talking about incredibly complex analytics," Lush says. "They're pushing and pulling, slicing and dicing massive quantities of data based on preset conditions or rules that are determined by the operator." While the variables change, the models remain a constant. Therefore, the key to these solutions lies in the ability of the operator to select and/or customize the model that will most accurately predict the type of information he is looking for. On the flipside these requirements have meant that these solutions were, for the longest time, relegated to those who had a firm understanding of the theories and mathematics involved in selecting the right model. Additionally, certain models were reserved for certain statisticians, all of which had their specialty within econometrics. "If you wanted to accomplish task A you would need to use product A, and there would be one specialist who would know that individual product and understand the results," says Marcus Hearne, senior product manager, SPSS. As a result, vendors like SAS and SPSS are now focusing on automating these processes so that enterprises don't need a rocket scientist to operate a forecasting solution. Vendors are focusing on bringing these tools into the front office via guided analysis and interpreted results for end users. "It's calculus on the fly," says Brenda Wolfe, product manager for forecasting and econometrics at SAS. "Many of these tools can now look at the data you're attempting to use, run a diagnostic test on it, and construct the model for you. They're capable of tailoring the models like a glove to the item you're forecasting." But if the user interfaces are becoming simpler, the analytical operations these tools are capable of performing are becoming increasingly complex. Once capable of dividing customer data into only summary segments, these applications are now capable of drilling down to minute segments with laserlike precision. "In the past the customer segments these solutions produced were chunky," says Andy Bober, director of customer intelligence product management at SAS. "There were fewer of them, they were big and unwieldy, and they were not as precise. Today, companies can look at smaller and smaller segments." Mike Gilliland, product marketing manager for forecasting at SAS, offers an example. "We have a retailer that is currently working on forecasting 32 million store item combinations, trying to figure out which items will sell best with the others," he says. "A number like that was unfathomable 10 years ago." Econometrics for the Masses? Despite vendors' best efforts to simplify and tailor the interfaces for the masses, a fundamental understanding of the data and procedures is still required to operate these systems. As a result, companies are leveraging these new and improved user interfaces and tailoring them for different sets, or levels, of users. These phased deployments are represented in one of the common trends seen in the marketplace today. To get the best of both worlds many enterprises have begun hiring small numbers of forecast statisticians as back-office power users to do the heavy-hitting analysis. This group, usually numbering less than 10, then distributes this data to analysts and forecasters throughout the company. Equipped with a simpler user interface, these analysts leverage the data from the Ph.D.s in the back office and customize the information for use in various departments throughout the enterprise. "Many forecast statisticians were worried they would be out of a job, because we were going to make these solutions too easy," Wolfe says. "Instead, it's had the opposite effect. Companies are hiring both, there's just different levels of users now." Delivering this data to the masses will be the job of BI solutions. While most experts agree that the masses, that is, a company's sales force and marketing departments, will probably never use econometric and forecasting applications tools directly, BI solutions will be the vehicle by which the data is delivered to them. "The interfaces for econometric tools are never going to be as simple as a call center desktop, but they don't have to be," Lush says. "That's the job of the Business Objects and Cognoses of the world. The intersection of econometric data and the masses will be BI reporting tools." Groundwork The key to econometrics and forecasting is a robust foundation of customer data. "Five years ago companies weren't doing this sort of precision forecasting, because the data quality wasn't there," Lush says. "The tools are simply a means by which to manipulate the data so you can gain insight from it." New data quality solutions like CDI/MDM platforms and data warehouses are changing the game by providing the groundwork for an econometrics and/or forecasting tool to sit atop them. The importance of these data quality solutions is multiplied if a company is seeking to integrate the forecasting application into its everyday business processes, as the data must be kept cleansed and accurate, ready to go on a 24/7 basis. The data itself varies industry to industry, but in general is aggregated data that represents a customer's behavior in relation to a certain environmental factor. For example, a retailer would be looking for prior transactional or behavioral data, such as products purchased and money spent, combined with environmental variables, such as holidays or promotions, inventory levels, price changes, and/or competitive pricing, to determine what factors will influence a customer segment's propensity to buy. For a bank, information such as debits and credits, account balances, and frequency of activity might be measured against environmental pressures such as loans and interest rates. Even weather can now be taken into account (the butterfly effect, of course). "You're comparing the data to try to find a common dominator for that segment," Bober says. "You're using the transactional and behavioral information as a backdrop from which you can attribute an environmental factor." To get an idea of the big picture, many companies are increasingly turning to third-party, vertical data providers to complement their customer data with marketplace info and trends. IMS Health is one such example. The Connecticut-based consulting firm provides market intelligence to pharmaceutical businesses by providing market trend analysis that would otherwise be absent from a company's database. "They're giving companies the big picture data they don't already have by filling out the overall customer profile," Lush says. The leading vendors in the section, such as SPSS and SAS, design their solutions to mine from databases of varying quantity and quality, but, according to Lush, a good rule of thumb is to ensure the data is at least in a manageable, unified form before entering the vendor selection phase. Once the vendor has been selected, the two can work in concert to determine the next course of action. When the data is there, the results are compelling. "You can begin to forecast certain behaviors in relation to a specific customer segment, and what that segment might do in relation to all the other products and services your company offers," Lush says. A company can gain insight into marketing efforts by understanding which product features are important to particular audiences and by modeling customer choices based on segment attributes, thus allowing companies to predict customers' decisions. Econometrics solutions can also account for seasonal fluctuations and trends--using time series analysis, an end user can determine the effectiveness of previous promotions and then create a customer demand model for future ones, taking into account marketing activities and the impact of certain environmental factors, such as pricing, advertising, in-store merchandising, store distribution, inventory levels, and sales promotions. Using these figures companies can determine if they should invest in more marketing or augment their sales force to match their product portfolio. From the B2B side, companies can adjust production and supply chain schedules to satisfy retailers' demand, or to reduce maintenance fees, such as in the automotive and airline industries. Recalling his variation on what happens when a wing flaps in Beijing and lipstick sales go up in San Francisco, Lush says, "Today, a company just might be able to figure out why." Contact Assistant Editor Colin Beasty at cbeasty@destinationCRM.com. AutoZone Augments Its Retail Performance Rajeeve Kaul knows firsthand just how diverse Americans are when it comes to the cars they drive. Some are wealthier than others; some are farmers, some are outdoorsmen; some have growing families, others have none at all. As director of product and price optimization at AutoZone, the nation's leading auto parts retailer, Kaul realized that his company couldn't adopt a one-size-fits-all approach to stocking the shelves at its 3,300 stores. "It's one thing to earn the number one spot in automotive aftermarket retail--it's another to stay there," he says. To keep that top spot, the company implemented the SAS 9 analytics platform, an integrated data management and predictive analytics platform that comes with fit-to-task interfaces so companies can tailor the interfaces for varying skill levels and usage patterns. Before implementing SAS AutoZone used a variety of tools for querying relational databases. Over time, as the amount of information being gathered grew and the complexity of questions being asked increased, Kaul realized AutoZone needed more. Kaul says the "fit-to-task" interfaces within SAS 9 were important, allowing everybody within the company from the CEO down to analysts to leverage the data. With almost half a million products sold at its stores, information races into the data warehouse at speeds far outpacing even the fastest Corvette. "The information in our world is voluminous, and in many ways undefined," Kaul says. "SAS can handle that data and sort through all the meaningless parts to show which lines of information we really need in order to make a decision. It helps us optimally use our time by allowing us to focus on questions that affect our business." Using SAS, AutoZone can look at the performance of its stores as well as individual departments, products, and categories within each store. Looking at product performance, for example, AutoZone knows what actions to take on certain products, pricing, and promotional activities across all 3,300 stores based on the geographical customer segments individual stores are serving. Kaul says, "We can drill down to these areas so that if something is not working at one store, we can replicate what does work at another." --C.B.