The data science team does not build a machine learning algorithm for the sake of building it. It is done to solve a specific business problem. And this business problem governs a lot of things in a data science team–be it the team structure, roles or deciding whether to have more data generalists or data specialists.

“Data science itself is a specialized job. But we have diverse specialisations within the domain too. A specialist would have a mastery in one specific kind of algorithm and look for their applicability. Generalists are competent on many of these algorithms and can look at the suitability of these according to the business problems,” said, Dr Allen Roy, Head of Analytics, Mashreq Bangalore.

But the question is how do you decide if you need to hire generalists or specialists?

The answer to this question depends on multiple factors like the business problem, structure of the team and the scale at which something has to be rolled out.

“If a problem is very clearly articulated and the expectations from that particular project are well defined, the specialist will work best on it. Let’s say we are running a credit card portfolio and we want to build an application risk score for our credit card applications which will determine if the credit card will be approved or not. In this case, a specialist in the credit card data science team is a must,” Roy explained.

However, if the business ask was to ascertain what should be the next product that we should be offering to our customers, it would need a generalist. Because in such situations, the team needs someone not only with algorithmic knowledge but also with business acumen.

“In open-ended business problems, we need to bring in a bunch of different things together to solve that problem so a generalist works much better,” he added.

A function like risk analytics will always have specialists but marketing analytics will most likely have generalists. The number however, depends on the scale of the business problem we are trying to solve.

For a bank with operations in 120 countries, 5 centralised teams spread across the world, with massive requirements for data science for countries, business problems, geographies etc. and at any point in time would need generalists. In this case, the whole work of generalists can be further divided into multiple categories of specialists because the scale is massive.

“The third aspect to look at is the operating model of the team. Is it a centralised data science team which works in a little bit of siloed manner from the business or is it embedded together. If it is embedded, generalists will do a much better job since they are more likely to understand the business domain. But if you are working in an agile model, a specialist would do better.” he said.

But no team works either only with generalists or specialists. You may need a healthy mix of both to stay afloat.

Shishir Thakur, Co-founder, Cranberry Analytics, feels that there must be a balanced mix of both in any project. In some projects the ratio may be skewed heavily in one direction, for example if the project demands more business domain knowledge compared to hard science, it may require several layers of generalists and a few specialists. On the other hand if the project is very clear in objective and is a narrow yet hard problem to solve, it may need more specialists than generalists.

“Complexity plays a significant role – companies utilizing exponential technology or complex use cases may benefit from specialists. Whereas, in a cross functional environment with a diagnostic / prescriptive scope, it is good to have most of the team specialized in one of the skills, but with a bent of generalization to be savvy enough in others to have conversations within and outside the team,” said Prashanth Kaddi, Partner, Deloitte India.





Source link