Organisations can gain more business value from advanced analytics by recognising and overcoming a number of common obstacles, says Mayur Joshi.
Rhetoric around the potential of emerging technologies––and the professions coalesced around those technologies––is often accompanied by as many myths. The evolution of data science coalesced around artificial intelligence (AI) and other intelligent technologies is no exception.
Many of these myths are simply the result of our infatuation with exciting and new advanced technologies such as AI and machine learning enabled analytics, challenges which I examine through my research on data science.
Data science in practice
To better understand why, my colleagues from both academia and industry and I have been conducting in-depth studies of the data science activities of three of India’s top banks with well-established data science departments (1). And what we found was that organisations can gain more business value from advanced analytics by recognising and overcoming a number of common obstacles.
Indeed banking is a sector where these myths seem particularly prevalent, where data science is closely aligned with the FinTech (financial technology) revolution. Today the digital transformation of almost every aspect of banking means that banks now possess a huge amount of both structured and unstructured data. Yet although they have started putting to use this data in strategic decision-making, the problem remains that many data science projects fail to deliver.
For instance, analytical solutions are likely to work best when they are developed and applied in a way that is sensitive to the business context. Dazzled by the hi-tech aura of analytics, managers can easily lose sight of this context, especially when they see a solution work well elsewhere, or when the solution is accompanied by an intriguing label such as ‘AI’ or ‘machine learning’.
Another problem that can be encountered is unrecognised bias and ensuring that it doesn’t creep into business models. This can be particularly challenging as managers who are not analytics experts cannot easily tell how the ‘black box’ of analytics generates output.
However analytics experts––who do understand the black box––often don’t recognise the biases embedded in the raw data they use, as these raw data are often generated based on the processes and practices devised by business managers and enacted by the front line staff interacting with the customers.
So one answer to this particular problem is for data scientists to become more familiar with the data generating mechanisms, which are at times controlled by the business managers and frontline executives, before using those data in their models.
The obstacles we identified invariably occurred at the interface between the data science function and the business at large. This suggests that leaders should be adopting and promoting a broader conception of the role of data science within their companies, one that includes a higher degree of coordination between data scientists and employees responsible for problem diagnostics, process administration, and solution implementation.
This tighter linkage can be achieved through a variety of means, including training, shadowing, co-locating, and offering formal incentives. Its pay-off will be fewer solution failures, shorter project cycle times, and, ultimately, the attainment of greater business value.
Data science in research
It is also worth mentioning that data science is now influencing the work practices of academic researchers too. Over the last decade we have seen techniques such as topic modelling and sentiment analysis, which facilitate analysing large collections of textual data, making inroads into management scholarship. While applying these tools in academic research the researchers assume the role of data scientists, which my colleague and I uncover (2).
These techniques come with a plethora of opportunities to build novel theories and make the previously unanalysable data more accessible. However they also bring challenges for management scholarship, some of which are similar to what we observed in industry, while some others are more related to the rigour in academic research.
1: These findings are based on my ongoing research project on data science in practice along with Ning Su, Rob Austin and Anand Sundaram. Some of the findings from this study appear as a practitioner oriented article in the latest issue of the MIT Sloan Management Review.
2: These findings are based on my ongoing research project on data science in research along with Wendy Gunther. Some of the findings from this study appeared as a short article in the 2020 International Conference on Information Systems.