Many companies today, especially those in the medium and large enterprise categories, are investing in setting up analytics teams and projects. They put together data scientists, give them tools, and provide support and guidance of various kinds from the IT and business functions. Depending on the specifics of the organization’s design, culture and ways of operating the new function is assigned a leader and an owner.

In addition to all of this, however, there are additional considerations that have to be taken into account in order to better improve the chances of success, and also deliver to the expectations of the managers that invest in the project. In no particular order, these are as follows.

  • Have a clear business objective. The analytics team needs to know what they are supposed to achieve in terms of a business objective. This will in turn provide steering and direction to all the rest of their activities.
  • Invest in achieving the right quality of data. In order for analytics results to be reliable the data that is used must be of consistently good quality. Sourcing, cleaning and integrating data to a high level of quality usually requires a lot of time and effort, possibly up to 70% of the entire project. It is a worthwhile investment.
  • Ensure that the data is as fresh as it can be. Once a model goes into operation, at each step of the statistical analysis and modelling stages the data that is used must be as fresh as possible because over a period of time changes in the external business environment may render results obtained with older data invalid. In the early stages of shaping a model it may seem like a lot of effort to keep getting the latest data sets, but being able to refresh these regularly also ensures that the model is validated more rigorously.
  • Have the right business expertise available. When dealing with tens of variables and trying to detect and analyse correlations and other patterns that may be subtle a very intimate knowledge of the business is needed, so that the right interpretations are made of each individual variable and how they usually change values.
  • Select the right tools and understand their capabilities. Once the business objective is clear, and decisions about data requirements are made, the right tools need to be used, from data integration to modelling and visualization, along with a clear understanding of their capabilities and suitability for the tasks at hand. Not having the proper tools can lead to results which are not optimal.
  • Try to produce results early. While it is always possible for a team to be tempted to take several months trying to refine and perfect a predictive model so that it produces the best results it’s probably better to adopt a quicker approach, trying a few variables early on and then adding to them or dropping a few in iterations. Producing results that are adequate, rather than perfect, and being able to produce them quickly also helps give the business and management more confidence in the whole concept of using analytics.