Apply good data science to outthink competitors’ marketing

Chief Data Scientist, Analytics Services, IBM

Marketing departments quickly adopted big data analytics and obtained good results. Many companies—such as Amazon and its noteworthy and effective personalized marketing powered by big data analytics—use big data–based marketing analytics to outthink their competitors. However, according to a recent survey by Kantar TNS, one of the largest research firms in Europe, delivering meaningful personalized marketing is still a big challenge for organizations.

Nevertheless, utilizing big data definitely is the first step for smart marketing. Sometimes, even a little insight such as the best timing for sending emails exposed with big data analytics can lead to some significant improvements. And, having all marketing being data driven is always a good thing, but it can also be dangerous if only incomplete data and some simple analyses are used. Incomplete data and simple analyses can create biased estimations or even lead to disasters. Many bad examples were identified in 2016 as the year of bad analytics by some people, including a FICO partner.

Three steps for data science work

To go beyond incomplete data and questionable analyses, at least, three steps of good data science work need to be performed. First, we need to create more complete data. Big data does not equal complete data. Data completeness helps us, at least, to avoid misrepresentation and model misspecification biases. The marketing data set from the past for any company is always a good place to start, but we need to merge it with other data sources. We need to merge the readily available internal data with some external data sets such as census data and open data, which are publicly available and easily accessible. With time and location data added in, we then need to go further by merging weather data and social media data, and applying text mining to generate new features.

Second, with more complete data in hand, we can then apply modern data analytics to derive causal structures from our data. Correlation is not causality, so using correlation to infer causality is misleading. For example, better marketing results may exist in certain areas with high-income customers, which truly deserves further analysis, but we cannot infer that income is the cause of making marketing more receptive with this piece of correlation only. Further, for this type of research, we need to consider utilizing deep learning because many layers of relationship to explore are available. Only with a solid understanding and discovery of some actionable causal insights can our data analytics be meaningful. third, after some actionable insights are derived, our next step needs to be designing solutions, rather than using analytical results just for recommendations. Good solutions developed from scientific analytics can be extremely powerful. In the case of marketing, they can potentially help prevent mistakes and increase sales. One way of utilizing this insight can be to develop a system of delivering personalized marketing or optimizing offerings.

As we learned from many successful data science projects, in this era of cognitive computing turning data into actionable insights requires always staying away from simple and easy analysis that often produces misleading results. Instead, we need to follow some good data science principles. For marketing analytics, we first need to create more complete data by merging and cleaning it, and then applying some suitable analytical methods rather than easily accessible methods such as simple correlation. This practice enables us to produce an implementable solution, rather than some questionable results to support misleading recommendations.

Two events for data science enrichment

To further discuss this important data science work to help outthink conventional marketing with big data, don’t miss IBM outthink tour 2016. The next event will be held 28 September 2016, in Los Angeles, California. And if you’re a working data scientist, data engineer or data application developer, register to attend the IBM DataFirst Launch Event, 27 September 2016, in New York, New York.