Big Data Analytics

In circumstances where companies have masses of data on customers, it can be feasible to run a segmentation without the need for primary market research. Sometimes the data are housed in different platforms, so they first need merging and cleaning, along with analytics to incorporate hypothesized data into the missing cells. A range of statistics can then be used to arrive at a segmentation. Latent class clustering is usually the more sophisticated solution as it detects latent variables in the data (like factor analysis) to easily identify an appropriate number of segments, and it can work on almost all types of data (unlike K-means which is limited to scale data).

Through this method, each customer on the database can be assigned a segment, and a tool can be developed to distinguish the segments that prospective or new customers would fall into. Additional data is sometimes appended to existing databases – whether purchased or acquired through primary research – and this can feed into the segmentation model. Whatever the method, big data analytics can be used for more optimal customer targeting, and therefore more tailored products / services and efficient communications to different audiences.

A common use of big data analytics is for interrogating and modeling data to better understand customers, markets or business performance. For instance, big data can be used to forecast demand and profitability of specific products over a given time period, to help write business plans and better plan manufacturing and inventory management. Big data can also provide insights into the competitive landscape, especially in markets where data are readily available such as pharmaceuticals and securities – for example, the anticipated impact on revenue from a rival launching a new competing drug. Big data analytics are also of value following mergers and acquisitions, when a company likely has a myriad of data on various brands, housed in different data platforms.

Substantial insights can be obtained by exploring synergy between the datasets, identifying patterns in the data, and inferring the opportunity based on factors of relevance to business success. A range of statistical methods are used depending on the amount and type of data available. Multinomial logistic regression models are used, for example, to predict outcomes (such as the product bundle most likely to enjoy the greatest price uplift based on a combination of factors such as demand, manufacturing costs, the competitive landscape, etc.) Monte Carlo simulation is another type of modeling technique used, especially in more risky business situations as the tool explores not only what the outcomes could be, but also the actual likelihood of each outcome occurring.