Why is Data Science such a bad career?

Dec 16, 2020 201 Comments

Media keeps saying Data Science is the sexiest career of 21st century, turns out it is the worst career choice for many. Every DS position is inundated with applications and it is so hard to get any DS position in the US

comments

Want to comment? LOG IN or SIGN UP
TOP 201 Comments
  • New
    x8kAn6

    New

    x8kAn6
    Pays like shit (80K-150K entry level)

    Lower career ceiling than SWE (DS maxes at 300K)

    Non-transferable domain specializations
    Dec 16, 2020 23
  • Google

    Go to company page Google

    Yeah the media hyped it up too much. DS is just the last mile for optimization. The core businesses of most companies rely on engineering and infrastructure.
    Dec 16, 2020 0
  • Asurion
    tokkale

    Go to company page Asurion

    tokkale
    There are Data science jobs and there are Data science jobs. Former is a bunch of jobs rolled into a sexy term to get on to the big data, data science, AI/ML hype band wagon. Not the fault of employees, it’s the companies that hype and name the data analyst roles to data scientist ones.

    Then there are the true Data science jobs where true ML, math, quant, algorithms are used, outputs are produced, dollars are saved, and ML pipelines are put in PROD. These are very few. This is the sexiest job - not the former one.

    90% of the Data science projects are run on laptops, that’s where they end - only 10% end up in PROD servers. They don’t even make it to dev data platforms.
    Dec 16, 2020 5
    • Stripe
      TYhE02

      Go to company page Stripe

      TYhE02
      I’ve seen people undervalue data analytics and hype the “sexy” ml roles, and most of them don’t really work on any data related stuff. If you look at how LinkedIn spun up their analytics team (the person who built this team actually came from the ml side), it’s obvious what the business needs are.
      Feb 28, 2021
    • L5 IC here with a prior background in academia.

      The thing that gets distorted by ENG's that become DS vs SCI to DS is that there aren't textbook solutions that you can plan to "just work" on the shit data that no one knows is shit, or represents the wrong business concept.

      Keep in mind OP that the 90% failure rate of projects is because you need to have a scientific approach to the fundamental business problem, which PHds and ENGs don't appreciate at first. Also going to a company that has a proper ML platform like metaflow working in their environment makes a huge difference for making ENG to DS happy with trying to match their traditional productivity expectations.
      Mar 6, 2021
  • Top reasons why my DS friends ultimately leave the field:

    1) Being the enemy/Life of disruption. DS is expected to find opportunity from the data. In reality, most change is a win-lose proposition, that usually means exposing something broken. Only the very strongest people can repeatedly message the need for change, creating losers in the company, and survive and it takes masterful communication and stakeholder management skills. It's incredible how change-avoidant most business leaders are and it's for very, very good reason---change turns the boys club where not much gets done, but boy does everyone sure get along, into a battleground with winners and losers. Most mid-senior management consider their #1 job to be "don't rock the boat".

    2) Position of weakness. You are constantly searching for opportunities, but never get to own a problem with pre-existing buyin like a SWE or PM. A DS is expected to find problems, influence others to solve them, and move on---this doesn't scale or imply a need for headcount. On the other hand, a SWE org would expand linearly as each problem needs a SWE owner. Hence product (SWE/PM) empires are easy to create and leaders are hard to remove---they have a position of strength, owning the "means of production" (CODE and ORG) as opposed to the DS position of weakness where you own nothing.

    3) The 'measurement' career is not measurable / you don't keep what you kill. Most SWE at big companies realize sharing ideas is a loser proposition as credit-stealing is pervasive in most corporate environments. They overcome this by only generating enough ideas for their own or team scope---and they own delivery of solutions. They "keep what they kill". As a DS, you have to structure your interaction with teams extremely clearly to leave enough of a paper trail to document influence---think of a million post its saying "I thought of that". A year later, when the solution lands, will anyone remember? You don't "keep what you kill".
    Dec 19, 2020 6
    • Google
      earthangel

      Go to company page Google

      earthangel
      That's what we used to call it cloud!
      Mar 16, 2021
    • OPs points happen but the extent of the issue is highly dependent on corporate structure. We have a couple DS COEs that definitely struggle to even get projects started cause of resistance to change. Change mgmt is critical and may take a while, but good business leaders can get those started. #2 is generally not true for us as there are specifically identified problems only get a DS if there is buy in. Then there's a team with a business PM/product owner to ensure everything gets implemented with the business appropriately and measure business value for the entire solution. The team accrues the credit and model is core
      Mar 31, 2021
  • Google
    rumntomic

    Go to company page Google

    rumntomic
    Data engineers can do most of what DS do. Why do you need a DS? The proper DS are called research scientists
    Dec 16, 2020 9