By Admin, Sep 13, 2019
According to KDNuggets, the top 5 data science and machine learning methods for 2018/2019 are all statistical approaches. In addition, the top 5 did not change at all since 2017.
George Box has said:
"Statisticians, like artists, have a bad habit of falling in love with their models."
It is no surprise to see the top analysis approaches are somewhat simplistic. Simple analyses should be used to answer simple questions. However, when complex business questions arise there is a need for a more complex analysis that is still easy to use. This is where market simulation is gaining momentum among analytics trailblazers.
Simulation is not a new concept and has been used for decades. NASA has simulated space systems since the 1960s. Meteorologists regularly use simulations to forecast the weather and pilots use flight simulations for training. The application of simulation for business purposes has started to emerge in the last decade as a solution to test ideas in a simulated environment and predict outcomes.
Historically, building a simulation required a team of PhDs to manually build a model using agent-based modeling. This process was cumbersome and did not yield answers at the required pace to become actionable. Recent technology advancements in machine learning, cloud-computing and software development now make simulation possible at the speed business demands for a wide variety of use cases. Financial organizations use simulation to spot indicators of the next financial crisis. IT leaders use simulation to predict data breaches. Manufacturing companies use the technology to support production planning. Meanwhile, cities and government organizations leverage simulation to create better transportation routes.
In business, leaders—from the CFO and CMO to head of sales—are beginning to use market simulation to better understand which insights and strategies will deliver growth and investment efficiencies. For example, broadcasting company E.W. Scripps leverages simulation to forecast media spend ROI for its advertisers.
Put simply, market simulation makes it possible to analyze ecosystems. By blending available data, domain expertise and machine learning, BI teams can recreate the dynamics and rules of how a population of people in a given market behave, influence each other, and make decisions. With so many variables and rules at play, there is a level of complexity involved that is best understood through the support of machine learning-powered simulation models.
The underlying science for market simulation is rooted in agent-based modeling and enhanced with behavioral economics, network science, attribution and machine learning. For those new to agent-based modeling or simulation concepts, there is often great skepticism. Here is an overview of how one builds a market simulation:
The first step in building a market simulation is to define the market competition. For every decision a person makes, they choose between a series of alternatives. These alternatives may include competing products, services, brands, or the choice to do nothing at all.
For example, a new TV streaming service would identify companies like Netflix, Amazon Prime Video and Hulu as direct competitors along with cable, satellite and premium channels. However, it’s important to also consider that a consumer may opt out of watching any programming.
Once the market is defined, then you need to identify the attributes that influence the decisions of individuals in the market and what their current preferences or perceptions are for the alternatives in the market. The simulated population may be divided into segments of consumers that have different behaviors, attitudes, and preferences. This recreates a population in a simulated environment. In order to do this, many companies use market research, segmentation studies, or other business intelligence to weight whether people care more about price or quality, convenience or speed.
Next, you incorporate all the other factors that influence individuals’ choices. Oftentimes, this includes the marketing activity of a particular company, competitive actions, and in some cases word-of-mouth.
Here is an example of a framework used to simulate the streaming video market:
Using this framework, data and domain expertise are input to apply data points to each variable.
The goal of creating a simulation is that it mirrors the real-world. In order to do this, the simulation must be calibrated by adjusting settings until there is enough confidence that the simulation can reliably predict future outcomes.
This is an on-going process since there will always be new data and learnings that need to be added to represent an evolving market. People aren’t static, so a simulation can’t be either. Over time, this allows the simulation to become smarter and even more reliable.
With a calibrated simulation model, a company is able to begin testing insights to see how various product updates, pricing strategies, marketing campaigns, and other factors can influence consumer decision making. Ultimately, the goal is to identify which “levers” to pull or strategies to implement in order to increase the likelihood that consumers select your product over other offerings. For example, a car manufacturer may seek to understand the impact of adding TV monitors to its line of minivans. Through market simulation, that manufacturer can clearly see the impact on sales and may decide to invest in TV monitors or pull a different lever altogether.
Companies are also able to leverage simulation to better understand the impact of scenarios they don’t have control over, such as a new competitor entering the market or the rise of gas prices.
Simulation is meant to build trust in data-driven insights, empowering businesses to test new strategies before making an investment. It’s the final step in realizing data as a powerful business asset that delivers ROI.