Wednesday, September 3, 2014

Social Media Analytics and Demand Forecasting – Does it really work?

Over the past decade, social media engines like Facebook and Twitter have used up much of our free productive time. So much so that it has now become a huge blob of data, most of which is just sitting on a server, or “the cloud” as most of us fancy it today, providing no additional value in the future. Does this really have to be the case? Can we actively use the data sitting in these servers to affect our business capabilities in the future? Social media analytics is proving to be an effective tool that decision makers across the globe are using increasingly to improve their daily business. And unknowingly, all of us, as social media contributors are taking an active part in this process.

A particular area in which social media is still under utilized is in providing demand forecasting for a product or a service. Two properties of social media specifically can be exploited to our advantage in providing forecasting. Firstly, it is current data that we can analyze in short-term time intervals. Secondly, there is a lot of data available. A lot of data! Both these characteristics make it a perfect platform for us to conduct a forecast. Let us see how we can put it into action.

We are all excited about the new Apple iPhone 6 releasing in a week from now. We are right? Well I am, and so are millions of people across the world. Everyone is tweeting about his or her anticipation and views about the next iPhone. So what happens with all this data about the iPhone that people share? Can Apple use this data to forecast user demand for their product even before the product is launched? If social media analytics gets it right, this is very much a possibility. So how exactly does this information translate into useful data for supply chain forecasters?

The huge amount of data uploaded by the users, in the form of tweets, likes and comments can be analyzed using social media analytics and provide a rating through a process called “Sentiment Analysis”. A rating of “positive”, “negative” or “neutral” is associated with a particular post, comment or tweet depending on the sentiment that it carries relating to that particular product or service. This rating is then turned into a numerical factor and plugged into a formula to forecast the product demand.

The method of sentiment analysis can be classified as a Causal method, as it assumes that a specific variable, in this case the sentiment factor, drives the forecast. Enterprise Resource Planning (ERP) software giant, Oracle, is seeing this method of analytics as a game changer in providing companies insights about product launch strategies, marketing strategies and sales forecast. Oracle, along with companies like IBM has already released social media analytic tools that have the capability of providing sentiment analysis aiding their traditional forecasting capability.

Where else can this concept of analyzing social media trends for forecasting demand be applied? “Google Flu Trends” was one of the initiatives taken up by Google to analyze and forecast flu trends across 25 countries in a time series analysis. The tool collected historical data of the total number of flu cases reported and created a forecasting model that could predict the outbreak of flu before it actually occurred. Applying our knowledge on social media analytics to this model we can analyze how sentiments on having flu can be used to provide additional information and help improve forecasting. Suppose I am catching flu and I tweet "Ugh, I am falling sick". This information will be analyzed and assigned a sentiment rating. Similar to me if there are millions other tweeting about catching the flu, it is not difficult to see how this information will be translated to generate a forecast. This information can come handy for organizations like the Centers for Disease Control and Prevention (CDC) or a pharmaceutical company to stock up enough flu medication supply before the flu season begins.

Critics of using sentiment analysis in forecasting have come down hard on it. The two major drawbacks of this method debated are reliability of data and risk of getting it all wrong. Can we rely of data uploaded by common people? What if people start using it to mislead companies on their forecasting strategies? It definitely needs to be factored in. But which forecasting method does not entail a certain amount of risk? That is why it is very essential to capture as much data as possible and provide forecasts for short-term periods, which can help reduce the disadvantages of this method. Over time, sentiment analysis can also be used in a time series analysis to predict social media trends and anticipate sentiments even before they are posted! The possibilities seem endless. I believe that within the next five years, social media analytics is going to impact traditional forecasting methods in a significant way. Only time can tell how this technology will evolve and leads us into a new era of forecasting.


No comments:

Post a Comment

Note: Only a member of this blog may post a comment.