It’s no secret that data science is one of the most in-demand jobs in the IT world right now. The number of people listing their profession as ‘data scientist’ on LinkedIn was 6.5 times higher in 2017 than it was in 2012, and data from Dice suggests this has grown more since then, with members of this profession able to command a median salary of $95,404 per year.
However, just because data science is becoming a crucial element of doing business doesn’t mean it’s something companies are succeeding at across the board. In fact, a Gartner analyst recently estimated that as many as 85% of big data projects fail. If that sounds familiar to you, it’s a good idea to assess why that is.
Quite often the technical aspects of data science projects go smoothly, but the work doesn’t produce anything of value to the business. This is a common issue, and can cause projects to underperform even if the science side of things was flawless. If you’re not seeing any value from data science, there are a few possible reasons why.
1. Value is an afterthought
MIT research scientist Kalyan Veeramachaneni writes in the Harvard Business Review of a panel he hosted in which he polled the 150-strong audience asking how many of them had built a machine learning model. About a third had done so, but when he asked if they had used their models to generate value, it turns out none had.
The problem Veeramachaneni sees is that data scientists are great at creating models, but aren’t best equipped to look at the bigger picture. The value of a data science project needs to be baked into it from the start, otherwise it will become an afterthought once the models have been created and completed.
Zank Bennett, CEO at Bennett Data Science, recommends starting all data science projects with a “product-first mindset”. In other words, before any work begins, it must be linked to a method of creating value for the company. Start by asking for a result, such as a forecast for consumer behavior over the next year, and then set your data science team to work trying to achieve it.
2. Your scientists are lone wolves
Linked to the above point is the problem of isolation. So much of data science requires taking a broad view and looking at the big picture, yet teams are often kept separate from the rest of the company. If value is to be achieved, data scientists need to be integrated with the wider business to ensure everyone is working towards the same objective.
This is because they’re able to look at different facets of the company and see where data can help. "Often they'll surprise the people in the business. The value wasn't where people thought it was at first."
Often, the best thing is to get your data scientists’ hands dirty by experiencing a problem firsthand. For example, New York City’s analytics department was sent out into the metropolis to look at correlations between fires in buildings and several other factors. By heading out with building inspectors, they were able to discover that the level of landlord maintenance was a major predictor of fire likelihood.
3. Your team isn’t right for the job
Oracle vice president of machine learning and artificial intelligence product development Ian Swanson puts it best:
In order to do data science right, it has to be a team sport.
Unfortunately, not all teams are created equally, and several factors could be causing yours to underperform.
Experience is a major factor. Recruiting data scientists early in their career will be cheaper, but without veterans at the helm the team might find itself floundering. Equally, team members need to be bought into the overall goal of providing business value rather than simply doing data science for its own sake.
Another important factor is your data scientists’ interests. If you have a team of analytics-focused professionals who are heavily interested in reporting but not in the engineering side of things, your overall results will suffer. Ensure your team has a diverse range of interests and experiences to boost value.