Decision paralysis is a big problem. One might expect that the area of data science would be free of this issue, but having more information and sophisticated tools doesn’t seem to always do the trick. Over-thinking and confirmation bias kick in. All evidence contradicting what the project sponsor/owner would like to see is rejected. Teams get flooded with more and more requests for detailed data and analysis. In the end, no decision is made.
To avoid locking yourself in the endless “we need more data” loop follow these tips:
- Keep it simple. Whenever a report or analysis has more than 3 dimensions, pause and think if they are really necessary. Too much detail obscures trends and makes things difficult to understand and visualize. Use simple labels on reports and charts. Make sure all metrics have well known definitions.
- Transform questions into a yes-no form. There is nothing worse in Data Science than an open ended question. “Who is the customer?”, “What are the key strengths of our brand?” etc. These can send you on a year long journey into minute detail, definitions and exploration without finding a satisfying response. Remember a lot of “basic” concepts in business and marketing are quite fuzzy. BTW: marketing researchers are more and more convinced that customers exhibit a continuum of behaviors instead of neatly separated clusters.
- Guess if needed. Data Science is often like a game of golf. You rarely get the ball into a hole with the first swing. As long as you know the direction – you will eventually get there. We are often perplexed by the complexity of the task before us and choose the safe path by trying to account for all uncertainties and unknowns. It is natural to fear failure, but in reality – the loser is the one that does nothing. Do! Guess, review and finally refine your model.
- Think within a framework. Use mental models to guide your analysis. Define a conversion funnel, standard customer lifecycle, acquisition and retention model etc. Analyze each piece separately and then again as a whole. Work top-down: define the general business problem and then identify which information and insight is needed. Start with a very basic tool (linear models are often enough) and review-refine as often as needed. Try to action from the very beginning and remember – as long as you know where the target is, you will hit it eventually.
- PAUSE – THINK – CRAWL – WALK – RUN. It is easy to be distracted by the “new cool” on the block. Are you eager to jump straight in into the Big Data world? Before you do it, stop and review your goals, resources and processes. Do you really need a Hadoop cluster? Have you already implemented all of the simple models before jumping into deep learning? Create a list of your problems and matching algorithms/models. Arrange it in the order of increasing sophistication (e.g. linear/logistic regression, association rules, SVM, Collaborative Filtering etc.). Start with the simple ones and learn more about your data. Action immediately and increase sophistication as long as you see business benefits.
- Accept results even if they don’t match with what you expected to see. Business environments (especially at the beginning of their analytic journey) are full of preconceptions and biases. Get to know your data. Thread carefully and don’t bully others into accepting your results. Have patience and people will eventually change their opinions.
- Give yourself limited time to make a decision. Agree to a rough plan of how long it will take you to do the first analysis. Admit there will be noise and uncertainty in the result and account for it in the implementation plan. The more noise you have, the more flexible, simple and easy to test should the solution be. If you are unsure which option to choose – just toss a coin. If you are unhappy with the result, select the opposite.