The great divide: how many jobs will be lost to automation?

How many jobs will be lost to automation by 2035? That’s a pretty loaded question, and every forecasting and consulting firm has a different answer.


  • Frey and Osborne (2013) forecast that 47% of total U.S. employment is at high risk.
  • OECD researchers (2016) forecast that only 9% of all jobs in the U.S. are at high risk.
  • PWC researchers (2017) forecast that the correct number is actually 38%.
  • ... and McKinsey (2017) thinks it's actually 46%.
  • </ul>

    So - who do you think got it right? And why?



    Osborne & Frey -

    OECD -

    PWC -

    McKinsey -

    Thanks for the nice and interesting recommendations @Roey!

    Overall, my impression is that the lower estimates of the OECD are based on the observations that it is not whole occupations that get automated but tasks within automation. The possibility of employees to adjust to automation by putting more emphasis on those tasks that cannot yet be automated well are the mechanism that decreases the percentage numbers as compared to the other studies. My feeling is that many economists view this refinement of the estimates as reasonable.

    Another important aspect related to these numbers is that they refer to the pure technological feasibility of the replacement of workers by robots. However, it will not always be economically meaningful to replace a worker by a robot even if that is technologically feasible. E.g. robots often have high maintenance/electricity requirements and are still quite expensive. Replacing a worker with a relatively moderate salary by an expensive robot with high maintenance and operating costs will often not pay off. So I would regard the percentages estimated in the mentioned contributions as an upper bound (referring only to technological possibilities and not to potential economic/legal/ethical considerations that companies might also have in mind).

    I forgot to post the links to two empirical studies that are interesting in this context. The first is on the effects of automation on jobs in the US and the second uses the same methods to estimate the effects on jobs in Germany

    The numbers estimated in these papers take economic considerations into account and they also include the effects of some compensation meachanisms (e.g., that goods become more affordable when produced with robots, which raises purchasing power and demand. This in turn can raise employment, etc).

    Thank you, @Klaus !
    I agree with both your points. I will say, though, that the PWC researchers utilized the same task-focused approach the the OECD researchers used, and yet got to very different conclusions.

    Thank you, @Roey, for pointing this out! I will try to find out possible reasons for that.