Big Data, Big Mistakes

Big Data, AI – Buzzwords?
Big data, and even Artificial Intelligence, are more like buzzwords to many within Pharma, and the understanding of these is very limited. This is totally understandable as the skill set to deal with these is very different from the traditional analytics team skill sets. Attempting to think of it as they did for traditional statistics is explicable but detrimental to real results. It is important for both marketing and analytics teams to get a greater understanding of all of these to ensure that the insights gained from the analytics are optimal to provide strongest results for the brands using them. A common mistake we see is teams expecting random bits of data uncover amazing insights without initial planning around the questions that need answering and the data required to support it for analysis. Big data, after applying AI to it, can uncover very strong insights but the process should be methodical and planned.

In our projects we tend to start with the problem or challenges being faced. The next step would be to look at the data available within the company. We do a GAP analysis on the questions that need answering and the data available to analyze. Once we know what data we have and what we need, the two can be matched and any data gaps uncovered and sources for that data found. Following that, the appropriate Artificial Intelligence techniques should be applied to the dataset. You will certainly find additional insights you were not looking for in the data, and that is an added benefit. However, do not start with just randomly looking at any data and crossing your fingers, hoping for something spectacular. Random data can provide random answers but it might not be necessarily what you want answered.

So, how should you go about the process?

Simply put, you need to plan. It sounds obvious but sometimes needs restating. Without the initial planning and scoping, you are setting yourself up for an expensive and potentially time wasting exercise. By developing a plan there will be a common language for the data scientists, the technology team, and the marketers to decide where the most fruitful place to start will be and what results are required.

Think of it like you do marketing planning. Would you plan your marketing by saying, “Don’t worry about the objectives or planning the segmentation, or specific messaging, positioning, channels, or budgeting. Let’s just do some advertising and see what happens.” You would never do that, would you?

You start with your objectives and what you are trying to accomplish; then you look at the data you have, and what you know. You supplement it to get the answers you need to complete your marketing plan, and then you plan the appropriate marketing strategy and tactics. It is not haphazard and unplanned, and neither should your big data analytics plan be. So, how do you plan a big data strategy for sales and marketing? Begin by considering what your key challenges are and what data you have available. Sales and marketing teams are critical to the process from the beginning. They are the teams that will implement the results, so they must be involved in the planning for setting the objectives, agreeing the dataset content used, and agreeing how it will be utilized in the real world. When you have the objectives from sales and marketing and the datasets available, examine the data and ensure you have sourced all relevant data (both internal and external) that is required to apply the analytics to get the answers needed.

Thereafter, try to follow this mathematical process:
    1.    Match the appropriate Artificial Intelligence techniques to the datasets and endpoints to determine the appropriate techniques and algorithms needed. For example, you may choose Random Forests and Deep Learning for one type of data, and maybe you are planning a Support Vector Machine (SVM) approach for another.


    2.    Once the data and techniques are planned and executed, you start a process of dimensionality reduction. This is an important step as you will be getting hundreds of millions to billions of data points from the process and you will need to pull out the ones that will impact what you are trying to achieve.


    3.    The model must then be trained to learn from the data put into it. Often R.NET and Rserve are used for this process.


    4.    Once that is completed, the model must be tested on the data for each analysis to predict elements in the data and be refined accordingly for more precise results. By doing this you can be confident that the approach is robust and you can anticipate what will happen.


    5.    Once these steps are concluded, the fun begins. You then begin applying data exploration techniques to each interesting dataset uncovered in the previous steps. Again, the techniques used depend entirely upon the data and facts found.


    6.    Once all of the data is analyzed, the commercial and strategy teams get to work, turning the results into insights that the marketing team can use and act upon to impact their results.

Eularis take this further and also program all the data sources and algorithms into simple-to-use dashboards so that marketing teams can plan various scenarios based on the data. This programming also requires numerous steps and typically involves millions of lines of code, which includes writing apps so that constant ongoing data enrichment can take place. This is good to include because the data will only be valuable if the marketing teams understand it, can use it, and can convert it into tangible business actions.