Commentary

Building The Ocean With Big Data

While working at an agency with a robust analytics group is exciting, it can also be frustrating at times. Clients challenge us with questions that are often difficult to answer with a simple data pull/request. For example, an auto client may ask how digital media is driving auto sales for a specific model in a specific location.  Another client may like to better understand how much they need to spend on digital media, and to that end, which media sequencing is most effective (e.g. search -> display -> search -> social, etc.).   Questions like these require multiple large sets of data, often in varying formats and time ranges.  So the question becomes, with data collection and aggregation more important than ever, what steps can we take to make sure we analyze Big Data in a meaningful way?

The Steps to Solve for Complex Analysis

Big Data promises to deliver a better understanding of your campaigns, a more precise ability to target, and ultimately provide grander insights about your customers.  But going after atomic-level data is daunting, so where do you start? Here are four steps to get you started to manipulate Big Data in a meaningful way:

1.  Define your objectives. What are you trying to answer using this data?   You will save yourself time and heartache if you simply ask the right questions and define your objectives early on.

       a.  State your goals.  Is it to produce more detailed reports, or deliver campaign insights?  Or is it to define your most valued customers for targeting?  Targeting to help inform: creative design, message placement, frequency capping, etc?

      b. Don't shy away from complication.  Ask complicated questions of the data.  You would be surprised how pliable and telling your data can be.  You and/or your analysts will determine what can/cannot be answered with your data. 

2.  Design a Measurement Plan.  This is a critical step which lays out your objectives in a document.

     a.  Be specific.  Be concise in defining your measurement plan.  Structure it so that your analysts and/or IT can quickly grasp the premise of what you are trying to achieve.

     b. Be detailed.  In your plan you should include defined measures of success, timing, assumed data source(s), and even a mock version of what the output will look like.

3.  Design Answer Data Cubes.  Often times answering one question leads to a series of follow-up questions.  You are bound to need more from your data.  The best approach to address variations of your initial request is to structure your data and/or reporting in a dynamic way.  Develop data cubes (dynamic reports) to allow for multiple pivots of the data.

     a.  Design cubes (i.e. aggregate datasets).  Develop aggregate data structures versus atomic-level data to allow for manageable and quick manipulation of data and allow sharing of data with others as you work through your requests. 

     b. Ensure cubes have sufficient and relevant information. The cubes need to be constructed to contain elements and data dimensions per the Measurement Plan (defined in step 2).   It is easy to over-populate your reports with information, so stick to only the relevant elements aligned to your objectives.  This is a delicate balance and often considered the 'art' in designing a report.

     c.   Ensure simple yet effective navigation.  Think of data cubes as dynamic reports; though be careful not to over-complicate your reporting.  Step 3b. describes the need to focus only on the relevant data elements.

4.  Tell a Story Through Data. Convert the results into a business story.

     a.  Have a beginning and an end to your story.  Besides being dynamic, your reporting should aim to tell a progressive story about your campaign.  Focus on:

         &n bsp;                                &nbs p;  i.     What happened: start with looking at what happened prior to the launch of a campaign or media activity?

       ;                                 & nbsp;  ii.     Trends: how are things trending?

                                         iii.     Issues: does the data point to any issues?

                            &nb sp;           iv.     Predict: what will likely happen if we stay the course?

    &nb sp;                                  ;     v.     Optimize: provide recommendations for improvement.

     b. Only report on what is relevant.  Do not over-produce metrics or mix various metrics, which is bound add confusion.  Separate metrics into:

         &n bsp;                                &nbs p;  i.     Diagnostic metrics:  meant to monitor a campaign's operational efficiency. Is the tracking and data collection working? Etc.

 &nb sp;                                  ;       ii.     Key Performance Indicators (KPIs):  designed to directly address the business questions and connect to your objectives (when possible).  Without a doubt connecting media metrics to business metrics is the most difficult combination to define.  When direct KPIs are not available, use Leading Indicators (KLIs) or metrics which are statistically correlated to your business metrics as a proxy.

By following these steps, we can help ensure that we collect the right data and analyze it effectively to solve even the most complex analytics challenges.

1 comment about "Building The Ocean With Big Data".
Check to receive email when comments are posted.
  1. Mark Hughes from C3 Metrics, July 27, 2011 at 6:01 p.m.

    Michael, great template for any marketer for the rest of their career!

    Especially liked "only report what is relevant" often we can find interesting added-value tidbits which get presented, but dilute the relevancy of addressing the objective.

    Excellent insight.

Next story loading loading..