In my academic/public service career, I've had what seem to me like a bunch of good ideas and a few important insights. But I've never had what I thought was a major conceptual breakthrough—until now.
Now, I think I've figured out how to have half as much crime and half as many people behind bars, and to do it over years rather than decades, using techniques we already understand and spending less money on police, courts, and corrections than we spend now. The trick is not to substitute social services for law enforcement, but to get punishment right.
At first blush, there would appear to be a trade off between crime and punishment: if punishment deters offending, then less punishment means more offending. But it's possible to have both less offending and less punishment by substituting convincing threats for the actual infliction of penalties.
There are two central principles of smart punishment:
1.) For actual human beings as opposed to economically rational expected-utility maximizers, certainty and swiftness of punishment matter a lot more than severity, in part because of behavioral economics and in part because a system that is certain and swift can also be fair, which randomized draconianism isn't, and people are more willing to adjust their behavior if they think the system that's pressing on them is reasonably fair. Moreover, severity is the enemy of certainty and swiftness, because severe punishments chew up scarce capacity and require a lot of time-consuming due process. That's textbook criminology going back to Beccaria, but putting it into practice is tricky.
2.) The more effective a deterrent threat is, the less often is has to be acted on. Because the risk of punishment faced by an offender depends not only on the capacity of the authorities to punish but also on the rate of offending, there is a natural positive feedback in offending rates. More offending means less risk for each offender; that's the logical structure of a riot. By the same token, as violation rates fall the risk of being punished for those who do offend rises. That creates the possibility of "tipping" effects, situations which have both high-violation and low-violation equilibria. In such a "tipping" situation, even a temporary increase in enforcement can lead to a lasting decrease in violation rates, and - this is the key point—that decrease can be sustained even after the additional enforcement resources are withdrawn.
Principle #2 leads to a strategic proposal: when there's not enough capacity to deter every possible violator of every rule—as there generally isn't—then it makes sense to concentrate threats on some subset of offenses or offenders or areas, "tip" that subset to the low-offending equilibrium, and then move on. Because enforcement is costly and enforcement capacity is scarce, enforcers and potential violators actually share a common interest in having actual punishment not happen; therefore, authorities ought in general to "telegraph their punches," issuing specific warnings to specific people in order to reduce the cost of the transition from high to low offending. And Principle #1 means that, if you can deliver on the threat of punishment quickly and reliably, you don't need to threaten years in prison to change behavior—days in jail can be perfectly adequate.