No doubt social media is a valuable source of business-related information. People routinely discuss brands, products, and services in social networks, blogs, microblogs and forums, where other people can read, comment, share, or just "Like," and in this way opinions are spread around the world by word of mouth (WOM). WOM is driven by customer satisfaction, trust and brand commitment and has far-reaching consequences (e.g., affective/emotional, cognitive, and behavioral) for both consumers and organizations. WOM can significantly change results of advertising activities across all other media channels. There can be synergy between WOM and other forms of marketing.
At the same time, word-of-mouth marketing means that an organization has taken active steps to encourage WOM (e.g., by offering rewards to the WOM sender), whereas, in normal WOM, the sender is not rewarded. Individuals are more inclined to believe WOM marketing than more formal promotion methods; the listener tends to believe that the communicator is being honest and does not have an ulterior motive. It is not always easy to separate WOM marketing from public WOM, but some docs cleansing procedures can reduce the portion of WOM marketing in social listening analytics. However, influential effects of such marketing can still be very significant and reflected in public WOM.
How we can read WOM as a big data source? If we read just a few random posts about a subject under our interest, then we have some understanding of what is being discussed and what the opinions are, but such reading can be very biased due to sample size, i.e., the volume of posts that we read personally. To analyze business impact, millions of everyday posts have to be treated as unstructured big data, and we have to use statistics, general math, and even sociophysics. Typical characteristics of social media posts about an entity/category/product include:
(1) Volume (total number of posts); and
(2) Counts of sentiment scores on different scales (positive-neutral-negative, or on a wider range, say, from -5 to +5, where 0 is neutral).
These characteristics allow us to compare volumes of discussions for different brands/products (which relate to awareness) and compare opinions (which relate to preferences). These characteristics can be connected with such business values as market share, sales, stock prices, number of complaints, etc. Understanding such connections is beyond the scope of this post, so here we will restrict ourselves to a starting point for such an understanding – dynamics of WOM.
On the time-series graphics below, as an example, we demonstrate a typical normalized daily volume of docs for a brand (blue line connecting blue dots).
To understand the graphics better, the daily volumes are supported by a moving average (µ) for preceding 14-21 days (red dashed line) and by boundaries (red solid lines) to frame typical volatility of daily volumes with 95% probabilities (µ ± 2σ, where σ stands for standard deviation). If we observe an unusual behavior (e.g., a daily volume outside the 95% probability range), then something unusual is happening in WOM and we have to understand it and, then, make appropriate business decisions. Typically, an unusual volume is associated with corresponding discussion topics and opinions. We can discover these by comparing uniqueness of a words cloud for unusual time interval vs. typical (inside the range) words clouds. This way, early warning is accompanied by information about the corresponding causes.
Many of advances social listening methodologies are incorporated in Genpact Media Interactive (GMI) app.