Hi, I'm Jang Saet-byeol and in each entry in this 'Big insight Big !nspiration' column, I have tried to share many interesting cases to inspire not only my Amorepacific colleagues, but also our customers by introducing and using new insights that have been obtained from big data. Even though big data can be somewhat distant or difficult, or perhaps not even important for you at all, we have still been able to discuss some interesting cases that have a close relation to your job. For example, we explored product planning (Ch. 1), manufacturing (Ch. 2), shipping (Ch. 3), and customer service/marketing (Ch. 4 & 5).

 In the final chapter, I'd like to mention important factors and considerations we must pay attention to when handling big data. I'd also like to discuss some keywords as these are important in the same way that we read precautions and side-effects when taking medication. It is my hope that this information will once again be of use to you.

 In the previous chapters, I discussed several cases that demonstrated successful usage of big data. This time, however, I'd like to explore things using a different perspective. It is true that big data can be a very powerful tool in this rapidly changing business environment. But, it is not a magic wand that can be used to solve every problem or provide you with all the answers you need. In fact, some companies have gone downhill after making misjudgments based on their analyses of big data.

 A Danish company The Lego Group, which is best known for the manufacture of Lego-brand toys, has experienced a decline in performance as other digital entertainment mediums have emerged and grown in popularity. This lead to the company utilizing big data to overcome this crisis. The results of their analysis showed that millennials tend to not be patient and instead prefer instant satisfaction. What was further implied by these findings was that existing Lego products were less attractive compared to the video games, which offer more immediate gratification.

 As such, in the year 2003, Lego made some changes based on their analysis and increased the size of the iconic plastic bricks to allow for easier assembly. This approach, however, backfired and saw a decrease in sales by about 30% compared to the previous year. In fact, the company was at risk of default in 2004. In contrast, Lego was able to regain its popularity and increase performance using small data.

 Martin Lindstrom, who is thought to be the best branding expert on this planet by many, introduced this case in his recent book Small Data : The Tiny Clues that Uncover Huge Trends. Small data, in his opinion, begins with the deep, close observations of a company's competition. An example of this could involve engaging in a more familiar activity, such as interviewing a customer. In fact, he would often open the fridge and drawers or even look through the trash cans to gain further insight into the customer's life and living environment (in the house or office). In other words, small data is all about capturing the reality with no restrictions or boundaries.

 Lindstrom said that it was actually a German boy's old sneakers that could save Lego from their impending crisis. At that time, while trying to gauge the behavior of millennials in small- and medium-sized cities, Lindstrom asked the boy about which of his belongings was the proudest of. The eleven-year-old boy, also a big fan of Lego as well as a skateboarder, promptly showed his old pair of sneakers. The shoes were evidence of the boy's hard work at becoming the best skateboarder in the city. For him, they had a deep meaning which might perhaps be lost on others. The reality of this observation struck home with the Lego team who realized that millennials found value in their efforts to achieve difficult goals, which was actually in contrast to the popular belief at the time that they sought easiest path to satisfaction.

 Now, I'd like to discuss an old movie. It's the 2002 American film Minority Report featuring Tom Cruise as the leading actor. The film is set in Washington D.C. in 2054, which became a safe city thanks to "Pre-crime," a specialized police department that apprehends criminals based on foreknowledge, or pre-cognition, provided by three psychics also called "precogs."

 The film opens with a prediction of the chief of Pre-crime, John Anderton (portrayed by Tom Cruise), becoming a criminal. In the film, the predictions are made by psychics and are definitely not based on big data. However, because of the prediction, which is a central theme of the film, John felt victimized as it seemed impossible that the chief of Pre-crime who never thought of killing someone is suspected of murder. The current reality in the U.S. sees the use of CCTV, GIS and criminal records being used to predict crime, which is of course different to the film, but it does indicate that prediction based on data can become a reality in the future.

 If you still think of this more as science fiction for the distant future rather than science fact, let's take a look at a more practical example. How would you feel if you received an email saying, "Based on our prediction using big data, your probability of moving to another company is within the top 10% of the company"? Moreover, what if you had never thought of moving to another company or leaving the company? In fact, the world-renowned PC manufacturer, The Hewlett-Packard Company, which had been struggling with a high turnover rate of 20%, did in fact develop a prediction model to calculate the probability of someone leaving the company based on data analysis.

 It is quite easy to discover unknown facts, be they past or present, using big data rather than to predict something that hasn't happened yet. Target Corporation, a large retailer in the U.S., found itself at the center of a controversy when it sent a coupon to a high school girl, who was pregnant but hadn't told her parents yet, after predicting that she was pregnant based on her purchase data.

 As you may have felt from the abovementioned examples, prediction or response based on big data casts a shadow that cannot be easily overlooked, especially when discussing privacy. Privacy has always been an important issue and has seen many discussions and debates ever since the early days of big data. The realization has been that if big data is used with an analysis model, you must take into account the inevitable damage that can be caused by a prediction error.

 A particular example of this error in prediction can be seen in the revelation that one in four Facebook users provide false data as a means of escaping the stress of prediction or marketing they do not wish to be exposed and as a response to their privacy concerns. It is perhaps time that more people and companies start to consider more the cons of big data as evidence is starting to reveal that in some ways, the cons do overwhelm the pros.

 In the beginning, as big data began to emerge, it was pointed out that while big data is a very powerful tool for understanding correlation, it is difficult to analyze causality using big data. This is a serious problem experienced in fields that make use of big data. This kind of problem usually occurs in so called "data-driven" companies that advocate big data just to follow the trends while maintaining a traditional culture or decision-making process. In particular, if they try to use only a part of the analysis results to support a pre-determined conclusion, rather than making business decisions based on data, they are very likely to face correlation and causality traps.

 To give you an example, let's consider hot days on which ice cream sales increase but so too do drownings. What then would the correlation be between the sales of ice creams and the number of drownings? Of course, following statistics, the correlation would be quite high. However, can we realistically conclude that drownings and ice cream sales increase are truly related?

 This is perhaps an extreme example, and we do often make the mistake of misinterpreting correlation as causality based on stereotypes or experiences. The reality though is that causality should only be established with a very detailed basis and statistical method through which no conclusions can be reached so simply. The easiest way to avoid this kind of error is thinking that there is no causality from the start.

 The conclusion should realistically be scientifically validated in a more objective manner when analyzing business attempts especially with intended goals and efforts, and their results, beyond just a simple analysis of some basic phenomenon. If you interpret the overall flow and correlation based on your intuition or experience, or try to interpret them in a positive way for the rewards you can receive based on your effort, what then is the difference between decision-making based on data and traditional decision-making based on intuition?

 For big data analysis, many experts say we need to measure everything that can be measured and transform the things that can't be measured into something that can be. The reasoning behind this is that acquiring data is essential for any type of analysis. So, what comes next after the analysis? Would it be utilizing the analysis results in a useful way? Then what would the objective be? Let's look at an example of how Google is utilizing big data with clear and effective objectives.

 In 2012, Google faced a problem with M&Ms. This odd combination of Google and M&Ms resulted from the snacks provided in the snack bar being arranged in the office according to the directions of Sergey Brin, a co-founder of Google. Google thought that if employees excessively eat snacks with high levels of calories and sugar, it will be harmful for their health. Following this train of thought, Google organized a team for 'Project M&M' to analyze the pattern of snack consumption of its employees.

 After analyzing the data based on employees' eating habits, Google decided to provide unhealthy snacks, such as chocolates, in opaque containers, and healthier snacks, such as nuts, in clear containers. The unhealthy soda drinks were placed at the bottom of the fridge where they would be difficult to reach, and bottled water was placed at eye level. These little changes led to an increase in bottled water consumption by 50% and a decrease in sugar consumption. As a result, the employees in the New York City office could reduce their overall calorie intake by a total of 3.1 million calories in just seven weeks.

 While this project created some controversy, I'd like to focus on the project's objective, which was to increase the health and happiness of the employees. What if the objective had been to reduce sugar consumption by 20%? Or, what if the objective had involved reducing the number chocolate containers being opened by 50%? Could they have still achieved these objectives? Even if they could achieve the objectives, would the employees have felt happy about the surprising result of the project that they had involuntarily been subjected to?

 Experts continually emphasize that the objective of utilizing big data should be to provide customers with benefits if a company wants to achieve successful business results through the use of big data. Companies, however, often set measurable short-term goals while utilizing big data and are really just using big data as a tool to improve traditional business performance indicators, such as the number of visitors, the number of customers who actually purchased, and sales. Of course, it is necessary to set and manage goals, but it should be noted that companies need to avoid being trapped by these goals and instead aim for something more meaningful.

 Many experts in fact warn that using big data as part of a short-term strategy to seduce customers will eventually result in decreased customer loyalty over the long term. You may be tempted to focus on achieving short-term goals to improve performance in a short period of time, but just as a runner needs to pace themselves to finish a marathon, you may need to focus on the goal of providing more benefits to more customers while maintaining a steady pace to go faster over the same distance and go further for the same time.

 In this section now, I'd like to discuss online A/B testing, which is a tool that combines the best of both of big data and small data. You may have heard of it also referred to as split testing, which is a way of comparing the effects of two items on a website or in a mobile application. This can also be extended to more than two variations, in which case it is called A/B/n testing.

 To give a more technical description, each of the users are shown one of the variations when they access a website or launch an application. The user responses to each variation are then measured to determine whether there is any statistically significant difference between the variations. An example of A/B testing with evenly distributed variations A and B will be performed as shown below.

 The advantage of A/B testing is that the results are easy to understand because the preferences of real users are directly shown. Everything, including the structure, flow, design, and text of a website, which has been determined by a marketer or a designer based on intuition or experience, could be subject to some form of A/B testing as this can lead to changes being applied in accordance with the preferences of the customers who actually use the website or the application. As such, A/B testing can be said to be in line with the 'Customers Decide' philosophy.

 Google, first introduced A/B testing to determine the optimal way of providing search results by performing more than 7,000 tests in 2011 alone. Amazon, eBay, Facebook, and most of the leading companies have conducted and are conducting a lot of tests to determine the best direction for their own businesses to take. Companies with a pervasive culture of A/B testing are performing the tests not only for simple webpages, but also across the entire business to move forward.

 Big data enables better observations to be made of the customers and the market as described in the book Do Not Imagine, which I introduced back in the first chapter. In line with this, there has been an increase in the importance of A/B testing as a tool for observation and selection. Amorepacific similarly introduced an A/B testing solution, called Optimizely, this year to promote the culture of A/B testing and experiments and next year, an event will be held where the company will share the cases of their A/B testing, so make sure your voice is heard at the event.

 This now brings me to the end of what is also the last chapter of my 'Big insight Big !nspiration' column. Looking back, I can't be confident or sure that I tried my best as I had first determined to, and even though I am not certain that I met the goal of introducing big data in a light and easy way, I'd like to take this opportunity to say thank you for reading and supporting my column.

 Once I said, 'We are not trying to determine whether to use big data or not. We just can't choose not to use it.' I'd like to introduce a story that left a big impression on me. I think this story is well in line with what I've said previously and the situation of the company regarding the adoption of big data. You may think you're too late to recognize the starting signal, but you can still start. Remember, this race is a marathon, not a 50-meter sprint.

A 50-year-old middle-aged American guy talks about his worries to his close friend.
"I want to be a lawyer,
but if I enter a law school now, I'll be 55 after completing the five-year course."
And his friend asks back.
"If you don't enter a law school now, how old do you think you will be in five years?"

 Thank you once again for reading my column thus far.