Value of Integrating Facebook Data For Customer Analysis
Unless you recently awoke from a coma, you probably noticed that Facebook filed for IPO earlier this year. The stock opened at $38/share valuing the company at ~$104 billion. Since then, the stock has gotten absolutely crushed as critics everywhere have come crawling out of the wood work with claims about Facebook’s mobile strategy (or lack there of) and attacking the usefulness/value of its mountain of social graph data from an advertising perspective.
Now, I’m not a finance guy and wouldn’t know where to start if I was asked to value the business…$104 billion? $50 billion?…no idea! But I am a data guy – and therefore know with absolute certainty that there is definitely value in Facebook’s social data. Petabytes and petabytes of data about real people: what they “like”, who they like, how they influence others (see Klout).
Quantifying the value of social data is difficult – there are many many factors.
Facebook offers advertisers the ability to slice and dice the social graph so that ads can be targeted at a very specific subset of users. However, in order to be successful, some preliminary analysis is required. For example advertisers need to know ahead of time what their target demographic looks like or they risk wasting money posting ads to users who don’t care. There are advertising methods that can be use to fine tune the demographic subsets (ex. A/B split testing where the same ads are sent to two different user subsets and the responses is compared to see which group had a more favorable response) and there are platforms popping up all over the place that integrate with Facebook and will enable you to employ these techniques…at an additional cost of course.
But, depending on your business model, direct advertising (think: click-through and page impressions) via Facebook may not be the best bang for your buck…as GM was happy to attest to just days before Facebook’s IPO.
What about using social data to increase the “richness” of your existing customer data?
Analyzing existing customers – especially the high profit customers – can be of great value. However, customer analysis is largely constrained by the “richness” of your customer data. By “richness” I’m simply referring to the amount of information you have on each customer.
Most data warehouses have a customer dimension representing all customers and everything known about each customer. It typically looks something like the following customer dimension from AdventureWorksDW2012 database:
As you can see there are the basics (Name, Address, Gender, Email Address) as well as some additional attributes that add some texture to each customer (YearlyIncome, TotalChildren/NumberChildrenAtHome, HouseOwnerFlag, Education, Occupation, CommuteDistance). These “textural” attributes are items identified by the business…and prove useful during post sales analysis – these are the attributes that drive the richness of the customer data-set.
In this case, the company, AdventureWorks, is an online retailer that sells bikes, clothing, and accessories. In this context we can start to see the benefit of these specific “textural” attributes.
For example, looking at the sales by commute distance, a business analyst may uncover that people who live within 1-2 miles from work are 10x more likely to ride a bike to work (than people living 5+ miles from work). Based on this information, the marketing department can target people living closer to work with ad campaigns.
Or perhaps in a more sophisticated (or contrived – depending on your perspective) example, where the team of business analysts mine the customer, product, and sales data and detect a cluster of customers that exhibit similar purchasing behavior: recently married males who just had a child are 20x more likely to purchase bikes and clothing for women and accessories (like a baby seat attachments). Based on this analysis, AdventureWorks can make adjustments to capitalize on this pattern.
Now, to bring things full circle, imagine your business is able to link your existing customers to their Facebook profiles. You are now able to increase the “richness” of your customer data-set by bringing in a host of new textural attributes into your analysis and data mining efforts. And this is where I believe the true power/value of social data lies.
Stay tuned for follow up posts where I’ll cover the obstacles associated with extracting Facebook social graph data, provide some examples of how to do it using SSIS, and discuss some various types of analysis that are possible once Facebook data is linked to your existing customer data-set.