Many organizations are drowning in the rising tide of fast-moving, high-volume, diverse social media data, from Tweets to blog comments to YouTube videos. According to the IBM whitepaper: “Social Media Analytics- Making Customer Insights Actionable,” consumers share 500 billion impressions online annually, and that’s not ebbing.
Mined intelligently, this data offers a treasure trove of consumer insight to marketing, sales, business development, customer service and even product development.
As Forrester Research reported in its 2013 “The State of Digital Customer Experience Technology,” 56 percent of surveyed professionals are making analytics strategies top priorities for the next 24 months, yet Forrester also found that 37 percent of analytics decision-makers say that finding adequate analytics talent is a major hurdle, according to their report “Global Digital Experience Delivery Online Survey,” from March 2013. Trying to apply old solutions to new problems with inadequate resources can create an expensive, time-intensive burden.
There’s no silver bullet for the social media data analytics challenge, but with the ever increasing data volumes, companies should re-evaluate their approach to analytics. By exposing the analytics function to more people within the organization, more decision-makers will be able to probe into data for new insights that will propel their businesses further.
The Analytics Continuum
In the recent TDWI ebook, "Investigative Analytics: The New BI Frontier" (June 2013), author Stephen Swoyer describes the bookends of the analytic continuum as follows:
Traditional Analytics. On one end of the continuum is traditional analytics, which puts questions into historical context, includes common BI activities (e.g., reports, dashboards, scorecards) and is mostly SQL-driven. Swoyer writes: “Traditional analytics doesn’t attempt to answer a question; this is the responsibility of the business user, who must base a decision on the preponderance of the available data. This is classic decision support.”
Predictive Analytics. On the other end of the continuum is predictive analytics (PA), which uses data mining or statistical algorithms to score data with models and forecasts. Swoyer explains: “PA is great for answering questions or anticipating events. [The] line of business doesn’t want its service technicians making judgment calls about whether or not a particular device will fail based on a probability score; it wants to alert a tech that this device is going to fail.”
Investigative Analytics. In between lives “investigative analytics,” which Swoyer describes as “an open-ended activity that looks for patterns, anomalies, and clusters (i.e., for clues) that can be used to formulate questions or which can be correlated with events, conditions, or phenomena.” He compares investigative analytics to PA, in that they both involve the analysis of structured and semi-structured data, and are highly iterative. Unlike PA, however, investigative analytics is a more open-ended process.
The sheer volume of social media data lends itself to the more open-ended root-cause analysis of investigative analytics. Investigative analytics allows users to ask a series of quickly changing, iterative questions to figure out why something did or did not happen. Consider, for example, an online retail aggregator that mines data to provide consumers with one-stop shopping. Applying different types of analytics to results in a variety of questions to be asked of social media data:
- Operational: How many Callaway golf clubs did we sell via Pinterest leads last week?
- Investigative: Why are we seeing so many leads from Pinterest this week? “Are there 10 boards giving the most leads? Are we selling more to the 18-25 demographic when the clubs are pictured being used on a golf course vs. in a static, indoor image?
- Predictive: How many clubs will we sell to users in the 18-25 demographic this month through Pinterest leads?
Social media analytics has previously focused on the content of posts e.g., text of a Tweet to measure consumer sentiment. However, to get actionable insight, companies need to take analysis further. Though it’s not the only step, investigative analytics can be a great first step for more complex analysis at massive scale. It allows non-data scientist users to “play” with social media data by asking iterative questions in near real time, regardless of data volume. Maybe marketing is monitoring Facebook and, thanks to a new query, they’ve decided to serve up a location-based coupon. Or, sales wants to take advantage of a trending topic with an immediate promotion, but they first need to ask new questions of the data that aren’t accounted for with BI’s predefined queries.
The emergence of investigative analytics complements existing approaches, creating new opportunities to find causation by the continual evolution of ad-hoc queries. That learning can then be put into predictive analytics models to inform even further action.
The Right Tool for the Right Job
When considering analytics options, it’s critical to leverage all of the structured social media data (e.g., location and time stamp of a Tweet) that accompanies unstructured text. In addition, by enriching social media data with insight from other systems such as CRM, financials and supply chain management, companies can get the most strategic view of customers and optimize activities for a particular outcome.
Operational analytics will always have a place for immediate alerts and KPIs, but it falls short in allowing for real-time analysis. Operational analytics is also typically limited in processing pre-determined queries. Likewise, predictive analytics works well in scenarios such as forecasting or planning (such as calculations to evaluate risk/opportunity and guide a decision) but requires correlations to be built into the statistical models. Not all companies have the talent or time to do that modeling.
Imagine a scenario where there’s a rush of customer complaint Tweets about a brand name product. Operational analytics can alert the company of a predefined level of negative posts over a certain period of time. Using investigative analytics’ iterative, real-time querying against the structured data about the Tweets, plus enriching it with CRM system data, the company could determine the issue and take action. Is it a shipping or inventory issue? Does this activity coincide with a recent local promotion? Predictive analytics could then leverage that learning, ensuring that future promotional activities trigger downstream supply chain activity to avoid future problems.
Investigative analytics could also be applied to more finite instances, such as a previously loyal mobile phone user who is now frequently complaining on Twitter. With investigative analytics, companies could move from network management to customer insight, determining that all calls on the customer’s mobile phone are now going through tower A instead of tower B. By combining social media analysis with network logs and call data records, it can then be determined that the customer has moved to another city. With the right information, companies can head off potential problems with appropriate intervention.
In order to capitalize on social media data, companies should weave together a tailored tapestry of analytics approaches that allows users across different roles and with different skill sets, to ask new questions of their data questions they may not have yet dreamed up, but that will inevitably lead to new insight. In the end, organizations will want the tools to not only answer “What?”, but to also ask “Why?”