Category Archives: Case Studies

Should My Business Forecasting Process and System Be Weekly, Monthly (Or Both)?

One of the biggest questions in establishing  a new business forecasting process is the level of detail of the forecast from a product dimension as well as a geography/channel dimension.  The third key dimension is time.  The process stakeholders in operations usually want to drive all of the forecast dimensions down to the finest level of time detail, which is typically weekly, while finance is focused on the quarterly reporting time period and the monthly forecasts which build up into that time period.

The process owner can’t make everyone happy, or can they?

What are the factors which should drive that decision?  The first issue is always whether there is the organizational bandwidth to go to weekly forecasting.  If true weekly forecasts are desired then the organization needs to have the people and processes in place to manage and be accountable for the weekly numbers.  A simpler way to put it is never plant a lawn bigger than the one you want to mow.

A second factor is whether or not the new business process system eliminates at least half of the Excel spreadsheets that have the word ‘Forecast” in their title.  If it doesn’t, then the new system is probably forecasting at a level which is going to be too high for the needs of your organization.   That’s a sure sign that there is a need for weekly forecasting as well as other drill downs on product and geographic/channel attributes.

Do weekly and monthly numbers ever have to true up? If they do, you probably need to support both weekly and monthly in a formal system and have business rules to convert the forecasts back and forth between weekly and monthly.

Regardless of your decision, software like RoadMap GPS has the flexibility to operate in a weekly or monthly mode, as well as the ability to switch back and forth between modes at the touch of a button.

RoadMap GPS can switch from weekly to monthly mode at the touch of a button.
RoadMap GPS can switch from weekly to monthly mode at the touch of a button.

Case Studies – Energy and Utilities

RoadMap has built forecasting models for many different types of companies in energy and utilities:

  • Weekly Demand Forecasting by PADD Region for Gasoline, Aviation Fuel and Diesel
  • Hourly Electric Power Demand Across an ISO Power Grid
  • Daily Futures Contract Prices for Natural Gas, Gasoline Crude Oil and Heating Oil

Weather is the major demand determinant in Electric Utilities.  In the utility industry the interaction between reduced supply caused by network congestion, maintenance or reduced hydroelectric power availability combined with increased demand from extreme temperatures can cause major supply/demand imbalances and price shocks.

RoadMap’s ability to forecast electricity demand at a disaggregate level of detail and at intervals as short as 30 minutes based on weather patterns gives the utility or ISO a much clearer view of demand and potential supply imbalances.

Case Studies – Media and Entertainment

The Media and Entertainment industries are characterized by very simple supply chains and high profit margins.

RoadMap has built forecasting models for many different types of companies in these sectors:

  • Weekly Attendance at Theme Parks
  • Daily Newspaper Sales by Newsstand
  • Weekly DVD Sales by Title by Store
  • Monthly TV Station Revenue by Advertiser Class

Case Studies – Consumer Electronics

No industry in the world is as exciting as, or more difficult to forecast than consumer electronics, where waves of better, faster and cheaper products flood the marketplace every six months. RoadMap’s technology planning software was developed in concert with industry leaders such as Dell Computers, General Electric and Samsung.

As with the Life Sciences industry, product life cycle management is critical to an accurate forecast. RoadMap’s ability to phase-in and phase-out forecasts for product lines, as well as the ability to forecast price compression curves are essential to producing accurate forecasts.

In many consumer electronics industries, such as printers and copiers, forecasting sales of spare parts and consumables is as important as forecasting original equipment. Higher margins on sales of these service items allow the manufacturers to lower introductory prices of new models of equipment. However, without accurate forecasts of consumables, it is impossible to forecast margins and overall profits.

RoadMap’s ability to forecast across multiple hierarchies allows high technology companies to forecast their sales by brand, by end-user segment (Home vs. Small Business) as well as by classes of product (e.g. equipment vs. consumables).

Case Studies – Consumer Packaged Goods

RoadMap has worked with more than 10 of the largest consumer packaged goods companies in sales forecasting and marketing mix analysis.

RoadMap software can track and plan baseline and incremental volume due to promotion from both POS scanner data and factory shipments. Much of RoadMap’s proprietary forecasting technology was developed in the response to the CPG industry’s need for accurate baseline forecasting. Without an accurate baseline forecast, it is impossible to measure the sales or financial impact of trade and consumer promotions.  RoadMap’s baseline technology is so advanced, that it can generate weekly baseline forecasts at very low levels of detail – by UPC, by Account – and for new products.

RoadMap’s causal modeling capabilities allow marketing managers to estimate the impact of price reductions, feature ads, displays, consumer promotion and TV spending on brand sales. With accurate evaluation of historical marketing programs, managers can design and execute optimal marketing plans for their brands.

Case Studies – Life Sciences

RoadMap’s first clients were in the Life Sciences industry. More than 10 of the Top 25 Pharmaceutical companies have implemented RoadMap solutions for:

  • Sales and Operations Planning
  • Strategic Forecasting
  • Patient Persistency Models
  • Patient Adherence Analytics
  • Sampling Analysis
  • Promotion Analysis
  • Patient Website ROI

RoadMap software can analyze models of RX data by patient. These models can predict patient persistence and adherence to therapy based on patient demographics, sampling, payer behavior and ICD-9 Codes.  These predictive models have proven to be highly accurate in both short term and long term forecasting.

Predictive Analytics in Life Sciences: Co-Payment Mitigation

Specialty pharmacies have emerged as a strategic distribution channel for pharmaceutical and biotechnology companies, as the dollar share of distribution by this channel will soon reach 40% of the U.S. market. Specialty pharmacies provide a wide variety of services beyond filling an RX.  Patient support coordinators assist patients in administration of drugs, coordination of insurance benefits, adherence to therapy and scheduling of refills

Privacy concerns over patient data can be addressed by the process of de-identifying patient data which removes all personal information from prescription records while maintaining important data on demographic groups, insurance coverage and medical conditions which are necessary to model behavior at the patient level.  Patient level analytics allow drug companies to model persistency of therapy and adherence, which are often critical to patient outcomes.

Co-Payment Mitigation

With rising prescription costs and economic pressures, payers have shifted more specialty drugs into higher co-payment tiers.  Without co-pay mitigation, many patients face hundreds of dollars of copayments per month, which have an enormous impact on adherence to therapy.  With patient analytics, life science companies can develop and optimize programs to buy down co-payments for patients, therefore maintaining high quality access to needed medications.

Utilizing Big Data For Strategic Patient Insights:

Patient-Level Modeling Ensures That All Impacts on Patient Behavior Are Captured.

Not so long ago, pharmaceutical companies had to try to analyze the impact of marketing programs using monthly RX data by sales territory or physician decile.  Insights on patient behavior were gleaned from pharmaceutical call reporting systems and survey research with doctors.

RoadMap utilizes state of the art nonlinear statistical modeling techniques that are specifically designed to model the behavior of hundreds of thousands of individual patients on a daily basis.  Unlike linear models used for aggregate RX data, nonlinear models use statistical distributions designed for discrete patient choices such as Binomial and Multinomial Logits.

Event based models of individual patient persistency on therapy also require nonlinear methods like the Cox Proportional Hazards model.  Despite the sophistication of the models, RoadMap is committed to explaining and documenting all modeling results, so that management does not have to rely on a fragile “Black-Box” approach for critical strategic insights.Image

Global Issues in Predictive Marketing Analytics

Predictive Marketing Analysis has proven to be a key competitive advantage for American marketing companies.  Who hasn’t wondered how or Netflix can come up with plausible recommendations for that next book to purchase or that next film to stream down to your gaming device?  Outside the U.S., it’s a different story as some of the most sophisticated marketing analytical techniques must be adapted to each country’s unique combination of technology infrastructure, privacy laws, marketing regulations and competition.

Predictive Analytics in the European Union

The privacy laws in the EU are much stricter than in America.  In America, credit card companies, E-Commerce sites, credit reporting agencies can build a comprehensive portrait of an individual, family or neighborhood.  Each European country has a different Data Protection Authority which governs what information can be kept on each customer, so that an analysis that could make your company $1 million in profit in the U.S. could easily cost your company the same amount in fines to the government.  Fortunately, by aggregating transaction level data, you depersonalize it.  There’s no need to aggregate over time, so that you can still conduct promotion response analysis – you just can’t calculate promotion response curves by individual.

Predictive Analytics in Asia

In the emerging economies of Asia, marketing infrastructure is still evolving, as digital TV and Internet connectivity continue to rise.  Traditional retailers are still protected from competition from large stores in economies such as Japan.  Last of all, the greater concentration of the population in large cities means that traditional marketing strategies such as outdoor advertising retain their effectiveness to a much greater degree than in North America. So, there’s no reason to believe that the optimal marketing mix in an Asian country will look anything like the best strategy in the U.S.

Cultural Barriers to Adoption of Predictive Marketing Analytics

America is still a very young country relative to Europe and Asia.  American culture is a blend of many different cultures, which is generally bodes well for marketing American products internationally.  But, when it comes down to marketing analytics, brand and product managers in other countries often demand to “start from scratch” with their own view of the consumer and competition and then build a predictive model from that point.  Trying to force-fit a predictive model, regardless of how well it has worked in the U.S. is a recipe for a disastrous implementation of a marketing program.


Predictive analytics has transformed American marketing.  It has yet to have a global impact due to many regulatory and cultural factors.  However, companies who can successfully export their analytic strategies around the world can reap the benefits of profit-maximizing marketing strategies With all the differences between US and international markets, the base case U.S. marketing mix can be very different than the optimal nternational marketing mix in country after country.

Product Life Cycle Curves in the 21st Century

Product Life Cycle Curves in the 21st Century

One phenomenon that is common to many recent hi-tech product launches is the compression of the innovation part of the product life cycle into the initial product launch phase. This is due to the fact that many consumers in the innovator segment are very savvy purchasers. They often preplan their purchase, and thus preorder. This type of phenomenon takes out essentially the first leg out of the product life cycle curve and combines it with the launch month. initial buyers in the modern product life cycle are now compressed into the product launch month, which becomes the peak sales month. Therefore, in the classic product life cycle, sales may take as much as 8 months to hit the
peak. Launch weeks and launch months are now increasingly important in marketing products with very short life cycles. Given the rise and pervasiveness of social media technologies such as Facebook, Twitter, and YouTube, products with a very short life cycle can “go viral” across multiple social networks or fizzle out very quickly. Social networks are really a force multiplier for the “word of mouth” effect. They can make or break new
products very quickly. Consequently, now there are much higher marketing expenditures
in the pre-launch marketing phase in order to build the initial pulse of demand. Marketers cannot wait for the product to be released to build momentum. They want to build sales momentum prior to shipment. This makes the classic product life cycle model irrelevant in many industries. However, sales forecasting still needs to be done as accurately as possible, given the dynamics of the market. Therefore, the best way to forecast
products with very short product life cycles is to follow these steps:

Step 1: Gather the historical sales data by product

To forecast products with very short life cycles, you first have to gather data on all of current and predecessor products. Historical sales data are not enough. You also have to collect data of the past three years on the product attributes of products that are currently offered, as well their predecessors. The exact list of relevant attributes varies by market, but these are the ones that matter most:

  • Performance characteristics
  • Multifunctional capabilities of
  • Price
  • Order of market entry
  • Distribution partners
  • Media spending

Step 2: Develop Product Roadmaps.

A product roadmap is a bar chart that describes how new technologies will be introduced into various aspects of the product over time. For example, in a mobile phone product roadmap, one bar would show exactly when the company’s mobile phones would support the 4th generation communications network (4g) instead of the existing 3rd generation network (3g). If your marketing team has developed product roadmaps in the past, they are valuable tools to help you identify the truest predecessors to your upcoming line of new products. The product roadmap of Intel Corporation’s Notebook processor line is available at What matters most for forecasting is how consumers look at the market, not how the company looks at the market. For example, if you want to forecast sales for a mobile phone manufacturer, you have to know how large the market segment is for price driven phones (e.g., feature phones that consumers get for less than $99 with a two-year commitment) vs. web enabled Smartphones that currently cost between $99 and $199 with a two year commitment. Smartphones also require a separate subscription to a data plan to access the Internet, which increases the total cost of ownership (TCO) of the Smartphone. In addition, you have to realize that the words “small,” “large,” and “fast” are constantly being redefined in many consumer electronics industries.

Step 3: Forecast Sales Over the Entire Product Life Cycle.

Once you have a good idea of the size of the market opportunity for the next generation of your product based on the previous generations of that product, you need to come
up with a statistical forecast of sales over the entire term of the product life cycle.
At this point, it’s useful to compare the statistical forecast to the sales force’s initial order forecast. The initial order forecast is the sales force forecast for the initial month of sales.
If the initial orders forecast for such a model of phone is about 25% of the total sales . One general rule of thumb in weighing the statistical forecast against the sales force estimate is that the forecast version that is lower would most likely be more accurate. Once that sanity check is completed on the total forecast for the product life cycle, the forecast is ready to be split into weeks for sales planning. Leading companies in consumer
electronics have built comprehensive E-commerce and demand planning systems that integrate product life cycle management data and collaborative planning and competitive intelligence with statistical forecasting capabilities to create the most accurate forecasts
and, consequently, the most responsive supply chains. Samsung Telecommunications
America, one of the leading mobile phone manufacturers, uses a system known as the Global Samsung Business Network (GSBN) which delivers comprehensive new product information, statistical sales forecasting, and inventory optimization capabilities to their retailers and channel partners.

Developing Predictive Analytical Applications with Big Data

How to Quench Your Thirst for Knowledge from Big Data

  • Daily Point of Sale Data from 3,000 stores for Ten of Thousands of SKUs
  • Daily RX Data for Hundreds of Thousands of Individual Patients
  • Real Time Demand for Electricity At Each Point Along a 250,000 Square Mile Electric Grid

We all thirst for information and learning from these huge data sets. But, in the time it take to read this post, terabyte upon terabyte of transactions data has been piling up in corporate databases around the world.  How can we harness the power of Big Data, while not drowning in it? Identify the Strategic Drivers of Your Business By identifying what is really important to know in your business and focusing on analyzing just those issues, you can cut down the real time and daily data requirements for your business. Key Performance Indicators or KPIs have been around for a generation.  It’s very important to have a subclass of these indicators which are real time KPIs. This enables you to whittle down you Big Data set into hopefully a manageable one. These KPIs can be as unique as the industries that they are in.

  • Mobile Phones: Pre-orders for Soon to be Launched Phones
  • Fashion Apparel: Color Mix Across all Styles and Lines
  • Pharmaceuticals: List of Patients More than 7 Days Overdue for a Refill
  • Utilities: Hourly Sunshine and Temperature Forecasts by City
  • On-line Retailer: Recommendation Engine

Using Automated Tools to Analyze the Data As Yogi Berra once said, “you can’t think and hit at the same time” – meaning that a ball player does not have a lot of time to think about whether or to swing or not and what type of swing to take, it must be a trained instinct.   In a similar manner, predictive analytics for Big Data must be analyzed automatically.  In other words, if it takes you a day to some up with a forecast of tomorrow’s sales, you are never going to be able to come up with a forecast in time to make a decision based on the forecast.  RoadMap’s forecasting algorithms make extensive use of re-entrant coding as well as in-memory data caching which allows for fast and efficient automated processing. Look for a Total Solution, Not an ERP Suite or a Single Package Depending on the complexity of the predictive model, your existing suite of software may just not have enough horsepower.  That’s why open source  projects such as the R Project for Statistical Computing can be such as valuable part of the total solution.  R, an Open Source project which is supported by numerous corporations, universities and hospitals has developed over 5000 statistical models called packages.  So, if your Big Data analysis engine needs to be turbocharged, it’s a good bet that there will be some predictive model in R which will help fill the bill. R Statistical Packages Your analytic needs may also exceed your IT infrastructure. That’s why cloud based solutions work so well with Big Data.  IT departments usually work off a 18 month long backlog of applications requests.  Strategic projects can’t wait 18 months; they have to be done immediately, with whatever resources are available. In summary, corporations that can cut their predictive analytics down to size by focusing on strategic imperatives, develop automated systems, and use multi-vendor approaches, which may even include open source or could solutions, stand the best chance of gaining business insight.