Facebook – The long hard road

Signs of AI progress. More needed.

  • Facebook’s ability to spot fake accounts is improving, but it is still making much longer than it should to detect these bad actors and remove them.
  • Facebook has produced its transparency report for 2019 (see here) which reveals a massive jump in the number of fake accounts being registered but also a good improvement in Facebook’s ability to quickly find them and delete them.
  • Despite the improvement, it still takes Facebook far too long to detect these accounts meaning that more work needs to be done.
  • During Q1 2018, Facebook deactivated (and I presume deleted) 583m fake accounts and it took on average 11.7 days to detect and delete each account.
  • During Q1 2019, Facebook detected and deleted an astonishing 2.2bn accounts (up 3.8x YoY) which were deleted at a rate of 24.7m accounts per day.
  • However, the total number of accounts that were thought to be fake and still present on the system has continued to creep up and currently stands at 5% or 119m accounts.
  • Facebook claims that the majority (which means >50%) of these accounts are detected and deactivated in minutes which, if true, means that the other 50% or so are much harder to get rid of.
  • If Facebook is detecting and deleting 24.7m accounts per day and there are 119m fake accounts on the system at any one time. it implies that the average detect and delete time is 4.8 days.
  • This is a big improvement on Q1 2018 where the average detect and delete time was 11.7 days, but it is still far from good enough.
  • The vast majority (almost 100%) of these accounts are detected and deleted before users report them but I suspect that a large number of these accounts are still being vetted by humans.
  • This would explain the long average detect and delete time with the obvious ones being spotted by the machines and the more subtle ones being picked up by humans.
  • Facebook now has tens of thousands of humans working on this which explains why EBIT margins have fallen by 500bp and will fall by another 500bp during the course of this year.
  • However, to be completely fair, progress is being made as the time to delete reduction is a good indication that the machines are handling more of the detections than last year.
  • Consequently, while Facebook is making progress with its AI, it clearly remains a weakness which is likely to continue for some time to come.
  • I continue to think that a company with declining earnings caused by recruiting humans to do the work of the machines should not be commanding a valuation this high and so I continue to prefer Google for 2019.

RICHARD WINDSOR

Richard is founder, owner of research company, Radio Free Mobile. He has 16 years of experience working in sell side equity research. During his 11 year tenure at Nomura Securities, he focused on the equity coverage of the Global Technology sector.