DATA DRIVES ANALYTICS
Analytics practitioners know this: familiarity with the data is the single most important factor in being able to produce quality results. Without an understanding of where the data comes from, how it is processed, and why it exists in the first place there is no hope for being able to use it effectively and repeatedly. If you are analyzing a marketing campaign, you need to know the nuances of your metrics—that means, for example, identifying the value in email versus display versus SEM based on what the campaign was originally meant to accomplish.
WHAT DOES YOUR DATA LOOK LIKE?
Take this a step further and think about how you will model your data for an advanced analytic effort using machine learning. Based on what you know about the data, how should you organize it so that your data scientists can make the data sing? Should their input tables be wide or narrow? What level of detail is relevant for the analytic?
And when you finally have the data ready for consumption, how do you continually prove that what has been readied is reliable and clean? Here is where we will ceaselessly bang the drum on implementing robust quality control and sanity checks on the data. This is done using repeatable data profiling techniques and an exploratory data analysis (EDA), all of which are shared across the enterprise so that all of the stakeholders can view and approve the data. Reliance on your subject matter experts to spot obvious trends or anomalies begins with creating easy-to-use interfaces for them to interact, filter, aggregate, and drill into the output.
BAD DATA = BAD RESULTS
We have worked with clients to resurrect unhealthy data ecosystems, but sometimes the damage has already been done. For example, several years ago we inherited a pricing analytic. The team that had created it had been let go and disbanded because they simply failed to curate the data that was used. The litany of crimes included having incomplete data, not using enough history, and biases were introduced when input values were based on gut and not evidence. Bad data yielded bad results as “good” customers were priced out and “bad” customers were drawn in.
THE POWER OF GOOD DATA
When the sources are highly validated and there is a high level of confidence in the data good things can happen. This proved true for a pharmaceutical company that had a hunch their drug reps were over-detailing (spending too much time on specific doctors). An evidence-based analysis replete with response curves and predictive models convinced the client to reduce the detailing budget. There was no loss in brand volume six months later and the exercise became an internal case study.
Beautiful analytics start with beautiful data. If you can’t point to what you have curated and state with certainty that it is correct you are better off doing nothing until you get there. Wrong conclusions will inevitably cause harm, but a system that is built on rigorous data acquisition controls and profiling will succeed.
