Much has been written about shortage of skills of analytical practitioners, unfortunately part of the problem is that there are many definitions of what constitute analytical set of skills needed to be able to provide an answer to analytical question. Some have expectancy that all it takes is knowledge of few analytical techniques, some level of domain expertize, knowing some software or programming language and that is it. Truth is, it takes all that and much more. It is different to put together a formula for exact set of skills needed to successfully answer specific analytical question – since experience, intuition and creativity play as important part as raw technical skills. However, below is my view of different knowledge areas that any analytical practitioner needs to have in order to be successful in analytical projects
Knowledge of methodologies and best practices
This knowledge gives practitioner helicopter view of the landscape that one needs to navigate starting from problem definition and ending with problem resolution. Knowing of what are the steps, where do you start, where do you end? What are the outputs of specific steps, what are things to be aware of, pitfalls. And there is different types of methodologies – analytical project methodology or analytical lifecycle. But there is also model building method. Since every analytical project has something similar to any other analytical project and also something unique - knowing methodologies case-studies of both success and failure, and learning what are the best practices in analytical applications in specific industry and specific areas are vitally important for successful deployment of analytics.
Knowledge of applications of analytics
Analytical projects do not start with the data, and certainly not with some data mining method technique! It starts with a specific (business) question. Analytical technology is the mere tool that is used on medium of corporate data for the purpose of addressing this challenge, and building specific business application of analytics. Since, purpose of analytical practitioner is to translate business question into a set of analytical tasks that produce an output that be moved back to business processes for the purpose of solving or reducing business problem – it is imperative to have knowledge of the problem area where analytics is required to assist. Building segmentation models, up-selling models, fraud-detection models, policy lapse prediction models, churn models, or any other specific business model – requires specific knowledge of these application areas, and so far this knowledge is only gained by field experience.
Knowledge of how to prepare data for analytics
Analytical tool’s purpose is to learn relationships that exist between the variables, in the data set. Preparation of the data set is designed to make the information enfolded in data set as accessible as possible to the modeling tool. Proper preparation is a crucial step in knowledge extraction process, yet - in data mining literature, it is often ignored, and in a shadow of more glamorous ‘analytical’ phase. Mining raw data is not advisable, and will not produce good results. Very rarely, if ever analytics happen on raw, transactional level of data. Most often data needs to be rolled up, relativized, derived and transformed in order for algorithm to be able to easily compare customer with himself at different times points (create temporal views of the data), to be able to compare customer with rest of the population, and to be extract key behavioral characteristics. Imagine, customer walk in retail store and purchase some product in single transaction. Very little can be learned from that. But if that transaction is expressed as relative measure of total transactional value in specific period of time then – one can say whether this specific transaction is once of, or part of deeper customer product affinity. Point is – data preparation skills is key in producing good models and it is absolute must in skill set of analytical practitioner.
This involves knowing basic methods of knowledge extraction from data, and knowing at least conceptually wide array of methods and techniques that address most fundamental analytical tasks, such as description, prediction, classification, clustering, association, sequencing, optimization, and time series analysis.
What “slightly” complicate matters is the fact that these techniques come from all over the place – statistics, applied mathematics, computer science, field of machine learning and artificial intelligence. Also, old analytical techniques are constantly evolving, while new techniques are being invented which means that for analytical practitioner learning never stops. Fundamentally, building an analytical model is re-formulating business question into an language of quantitative disciplines (statistics, applied maths, and after such model is built - bringing it back into the area that originated in first place, which is business. Since, everything we do on technical side, such as preparing data, building the model, etc - is a function of precisely following business requirements - technical part cannot be viewed in isolation.
Communication and “soft” skills
As it was already said technical part - purpose of analytical practitioner is to translate business question into a set of analytical tasks that produce an output that be moved back to business processes for the purpose of solving or reducing business problem. Before that happens, business sponsors and end-consumer of analytics need to be convinced or fairly certain that this would indeed happen. Since there is intrinsic disconnect between technical people with statistical background who use analytics to create analytical business solutions and business consumers for whom these solutions are created – this “convincing” becomes critically important. Practically speaking, that means that analytical practitioners need to be able to explain what is intrinsically technical in non-technical and in business terms to audiences that don’t know much about analytics. I have seen very good analytical projects with solid results never being implemented because analytic practitioners have done poor job of explaining functioning of the model and why should model be trusted in production, and I have also seen some mediocre results being presented in an articulate and concise way that made them stand out and capture full attention of business audiences, appearing to be stronger and bigger than what they really were.
And then there are soft skills. Analytical projects go through its life-cycle are seldom without challenges and obstacles that can be either technical or people-related. It greatly helps to be humble, likeable, honest, diplomatic, supportive, understanding, helpful and co-operative because you are very likely to get same attitude back from your clients or business sponsors, and that is exactly what you may need in times when things are not happening as planned.