Very informative article from The Atlantic. TL;DR:

• The spread of the virus obeys the Pareto principle: something like 80% of the contaminations comes from 20% of the infected. In other words, if in average one infected person contaminates x other people, that x is the average between many 0s and a few big numbers. In other words, the virus spreads in clusters.
• Such dispersion seems to be the result of properties of events rather than properties of the infected people themselves (we talk about super-spreaders when we should talk about super-spreading events). Events to avoid are described by the 3 C's: Crowds in Closed spaces in Close contact.
• Backward tracing is preferable to forward tracing. Forward tracing won't yield interesting findings in most cases because an infected person is unlikely to contaminate someone else. Backward tracing is crucial to identify clusters because an infected person likely belongs to a cluster.
• Unreliable tests are valuable, as long as they yield no false positives (false negatives are okay). If there is a suspicion of cluster, then everyone needs to be tested cheaply and rapidly. If there is anyone who got the virus, then odds are that many other people got it too. So the test only needs to find a single one of them to raise the alert (some false negatives are okay). However, the test should not yield false positives (or it will raise alerts everytime).
• Japan understood all of this in February 2020 and contained the virus without any lockdown despite having some of the densest cities in the world.
• Many of the regions most critically hit by the virus around the world might simply be bad luck of having had a big cluster in this specific region.