An independent study reference written by Dr Phuc V. Nguyen. It is not official subject material — for assessment requirements always follow your subject outline and vUWS.
Roles in an analytics team
Analytics is done by a team with distinct jobs, not by one person who does everything. A data engineer builds and maintains the pipelines that move and shape data. An analyst answers business questions from that data and gets the answer acted on. A data scientist builds models where a query will not do. A machine-learning engineer takes a working model and makes it run reliably in production. A data steward owns definitions, quality and access for a domain. Davenport and Patil, revisiting their 2012 argument ten years later, describe the shift from hunting for one exceptional individual to building this division of labour.
Why it matters
A restaurant is not one brilliant cook. Somebody buys and preps the ingredients, somebody cooks, somebody plates the dish and explains it to the table, somebody keeps the kitchen to standard. You can find a person who does all four, and they will do none of them at full capacity. Analytics teams learned the same lesson. The useful question for a student is not which title is best, it is which part of that kitchen you want to be good at.
A company's dashboards are frequently wrong for a day after month end, and different reports disagree on what counts as an active customer. Which capability is most obviously missing?
Worked examples
A retailer hires one senior data scientist to do analytics. Eighteen months later there are three impressive models and none of them is in use.
Look at what a single role could not cover. The models ran on extracts pulled by hand, so there was no pipeline to feed them daily. Nothing was instrumented, so nobody could tell whether a model still worked. The definition of an active customer differed between the model and the finance report, and no one owned that difference. Those three gaps belong to engineering, to production ownership and to stewardship. Hiring the modelling role first is common, and it can produce exactly this outcome.
Two graduates apply for the same opening. One writes solid pipelines and dislikes presenting. The other reads a business problem quickly, writes plain SQL, and explains results well.
They are not really competing for the same job, even if the advertisement says they are. The first will grow into engineering work, where correctness under load matters and the audience is other technical people. The second will grow into an analyst or product-facing role, where the binding constraint is framing the question and moving a decision. A team wants both. The mistake is ranking them on a single scale of technical difficulty, because the two absences break an organisation in different ways.
Common mistakes
- ✗Data scientist is the senior version of analyst. They are different jobs, not two levels of one job. An analyst who is excellent at framing questions and shifting a decision is not an apprentice modeller, and a strong modeller is not automatically able to run a stakeholder conversation.
- ✗The engineering roles are support work and the modelling is the real analytics. Many analytics failures are pipeline failures and definition failures. The model is often the smallest and best-understood part of the whole system.
- ✗Job titles map cleanly onto tasks across organisations. The same title covers very different work in different places. Read the description of the day-to-day work and ask what the role is accountable for, rather than trusting the label.
- ✗A team needs one person who can do everything. Someone who spans several roles is genuinely valuable as a translator between them, but a single such person is a bottleneck and a single point of failure, not a team.
Revision bullets
- •Distinct jobs: data engineer, analyst, data scientist, ML engineer, steward
- •Engineering moves and shapes data; analysis answers questions and persuades
- •Stewardship owns definitions, quality and access for a domain
- •The single all-rounder hire is a bottleneck, not a team
- •Many failures are pipeline and definition failures, not modelling failures
- •Read the accountabilities of a role rather than its title
Quick check
A company's dashboards are frequently wrong for a day after month end, and different reports disagree on what counts as an active customer. Which capability is most obviously missing?
A graduate is told to choose between deepening their statistics and learning to write clearly for non-technical readers. What is the sound way to think about this?
Connected topics
More in The Analyst and What Comes Next
Sources
- Davenport, T. H., & Patil, D. J. "Data Scientist: The Sexiest Job of the 21st Century." Harvard Business Review, 90(10), 70-76, October 2012.The article that established the data scientist role in business language, and the source of the all-rounder expectation that later had to be unwound.
- Davenport, T. H., & Patil, D. J. "Is Data Scientist Still the Sexiest Job of the 21st Century?" Harvard Business Review, July 2022.The ten-year revisit: the role became institutionalised and split, and organisations now build mixed teams rather than search for one exceptional individual.