A physicist, a chemist and an economist are on an island with only a can of beans. The three are contemplating methods for opening the sealed can. The physicist says: "gravity will open the can if we climb that cliff and drop the can"; the chemist says: "heat will melt the can if we place it under a magnifying glass in the hot sun"; the economist says: "my solution is easy. Assume a can opener!" With predictive analytics software, the theory is one could predict the proposed solutions by each of the experts.
Many firms now employ advanced analytic models to predict behavior and to understand markets and other systems. As any good quantitative analyst will agree, however, models are only as good as the assumptions underpinning them. Increases in computing power and to data quantity have opened the door for major innovations in the field of advanced analytics, but a human touch is still required to make sure those models accurately reflect the world around them.
Enacted in 1981, the federal Research and Development (R&D) Tax Credit allows a credit of up to 13% of eligible spending for new and improved products and processes. Qualified research must meet the following four criteria:
Eligible costs include employee wages, cost of supplies, cost of testing, contract research expenses, and costs associated with developing a patent. On January 2, 2013, President Obama signed the bill extending the R&D Tax Credit for 2012 and 2013 tax years.
For years, firms have developed models to try to reflect human behavior. Choice modeling has attempted to understand, for example a customer's willingness to pay for quality improvements to a product. The key assumption underneath such models is that human behavior is inherently rational and therefore based on the maximization of one's utility given a finite amount of money to spend.
Added computing power and data processing capability in recent years has allowed ever more robust models to be constructed. Software programs like Sawtooth have developed to harness the expanding power of computer modeling, allowing analysts to consider more variables in their analysis and to construct models with subcomponents so as to provide greater precision in within one's results. One type of newer modeling methodology is called the menu based choice model. This type of modeling allows firms to analyze how to price an entire bundle of services, such as how a cable company would choose to price a bundle of cable, DVR capability, internet, and whatever other features might make up a given suite.
Advances to computing power have inspired development in many aspects of modeling. However, the best market analysts do not turn everything over to technology. Modeling analysts must continually ask themselves whether their results make sense. Often, erratic results occur when a critical factor has not been adequately contemplated within a model, or when an unnecessary factor must be removed. Highly trained and experienced technical professionals, typically software engineers, build out these models and carefully examine them to make sure that the assumptions underlying their results are valid.
Descriptive models play an exciting, complimentary role to prescriptive models. Descriptive models look at macro trends. Understanding the relationships between macro trends is the goal of descriptive modeling, whereas predicting the actions of any one agent within a system is the domain of prescriptive modeling.
As an example, imagine an ant brood. Each ant in the brood has a different quality of vision which dictates how far from the brood the ant is able to travel to procure food. Perhaps the ants also differ in how much food they can carry, or how long they can go without food. When these ants begin to move within a simulation, macro patterns can be observed, and relationships between those patterns can be better understood. How, for example, does density of a food source affect how much food the ants can collect? If vision improves twofold, do the ants procure twice as much food?
Descriptive models are currently being used to better understand the U.S. energy grid, traffic, and terrorism, among many other topics. In terrorism, for example, some models attempt to understand what conditions affect self-radicalization - individuals banding together into a terrorist cell. A better understand of those conditions and their interrelationships can help shape policy so as to forestall the self-radicalization process.
Modern psychology recognizes five dimensions of personality: extroversion, agreeableness, conscientious, neuroticism, and openness to experience. Experts believe that twitter tweets can be analyzed to predict the tweeter's preferences. This personality profiling is based on a study of bloggers performed by Dr. Tal Yarkoni of the University of Colorado. Dr. Yarkoni first determined the bloggers' personality traits by questionnaire. He then correlated the words and categories of words they used in their blogs with their personality traits. Using these concepts Dr. Eben Haber from IBM's Almaden Research Center in San Jose, California has developed software that takes streams of tweets and searches them for words that indicate a tweeters personality values and needs. In a Big Data application Dr. Haber analyzed three months of data from 90 million twitter users. The conclusion is that one's presumptive personality can be determined from just 50 tweets and can be well identified from 200 tweets. We call these predicted social media personality characteristics "Behavioral Twitterlytics."
Netflix is the largest provider of commercial streaming video programming in the United States. It uses predictive analytics to power its strategy and business model which, as a result, impacts its customer profiles.
Netflix did their own behavior analysis test on their (33 million worldwide) customers. That in depth analysis of their subscriber viewing preferences would pay off. Accordingly, they invested $100 million on two 13-episode seasons of the remake of the 1990 BBC series, "House of Cards". Their data indicated that the same subscribers who loved the original "House of Cards" also thoroughly enjoyed both Kevin Spacey movies and movies directed by David Fincher. So, they put all three together based on their data and predicted a big hit.
Netflix gathers an astronomical amount of data on its customers. It gathers information such as the device a customer watches on, what night of the week it's watched on, what genre movie was watched, the positive or negative rating given, bookmarks, social media references, location, and more. It even takes into consideration when a subscriber fast forwards, rewinds, or pauses a specific scene and at what point the viewer stops watching.
Mohammad Sabah, senior data scientist at Netflix, reported at the Hadoop Summit that Netflix was even "capturing specific screen shots to analyze in-the-moment viewing habits" to "consider things such as volume, colors, and scenery that might give valuable signals about what viewers like. " But one has to wonder what happens when more than one person uses the same account. How would the algorithm accurately portray a subscriber's movie or television preference when there can be many different ones in a household? All in all, Netflix maintains that due to their use of predictive analysis, they influenced 75 percent of its subscribers to choose an additional Netflix offering. Netflix's video streaming is itself disruptive technology for the movie, cable and traditional media companies. By adding more advanced predictive analytics, Netflix accelerates the technology based disruption.
Many feel that the key to many pressing social problems lay in the power of descriptive models. These models are exceptionally customizable to the problem they are trying to solve and to the assumptions which have been chosen to form the basis of the model. Once again, human intelligence is crucial to make sure the model reflects reality as accurately as it possibly can. These modeling endeavors, and the considerable man-power required to develop them and refine them, are eligible activities for the R&D tax credit.
Charles R Goulding Attorney/CPA, is the President of R&D Tax Savers.
Daniel Audette is a Senior Tax Analyst with R&D Tax Savers.
Charles G Goulding is a practicing attorney with experience in R&D tax credit projects for a host of industries.
