Getting data analytics right

Analytics as an art

The science of analytics is well understood, but organisations should also see analytics as an art “because art is a lens for interpreting the world in different and often insightful ways”, said Parker.

Parker said that requires analytics teams to identify the right questions, and not simply the right answer, because one purpose of art is to provoke and challenge, which is generally achieved better with questions than answers.

“Good analysis, like good teachers, answers questions, but just like great teachers, great analysis can prompt us to ask new and better questions,” she added.

For example, Parker said while “Who is buying a product?” is a question that can be easily answered with a maximising strategy, a great question would be: “Who is buying our product that looks nothing like the customers we know and understand?”, because the answer would help an organisation identify new market segments.

“Data artists create great questions, just like data scientists create great answers, and we need both,” she added.

That said, analytics practitioners will also need to know when and if an answer is good enough.

“For a decision-maker, an approximation in the next five minutes could be more valuable than the exact right answer in five hours,” Krensky said. “And remember, every question answered potentially leaves a different question unanswered, so we have to constantly ask if we want to spend another hour answering this question or spend that hour investigating a different question.”

In some cases, organisations might not even find an answer to their questions. Take, for example, the Brussels Intercommunal Transport Company, which had hypothesised that its buses were breaking down unpredictably due to the bad driving techniques of its drivers.

Krensky said the company’s business intelligence (BI) team sought to solve the problem and applied the Scrum technique of time-boxing, spending no more than 10 days on data preparation and no more than 10 days on model development.

The plan was to use the buses’ telemetry data to predict breakdowns by applying deep learning techniques, but the team could not find any pattern at all.

“Because they had time-boxed their project and accepted the risk of failure associated with this kind of innovation, they were able to move on instead of getting bogged down by trying to prove a point.”

Krensky said the good thing that came out of the project was that the BI team had built a trusted relationship with their operations colleagues who could see that they had the integrity to not force a result when they couldn’t find one.

In addition, the BI team had identified a list of buses that run in the hilliest parts of Brussels, and by rotating those buses to run in every part of the city, the company was able to save €40m a year.