Gollwitzer and Sheeran's review, Implementation intentions and goal achievement: a meta-analysis of effects and processes (2006), reports findings from 94 independent tests. The effect of an if-then plan on goal attainment was positive and of medium-to-large size, d = .65. The NCI page credits the same review with d = .61 for getting started on a goal and d = .77 for preventing derailment.
The abstract says the plans helped people start pursuing a goal, shield it from unwanted influences, disengage from failing courses of action and conserve capability for later goal striving. It reports support for two processes: a plan makes the specified cue easier to notice, and it makes the response more automatic.
How big is d = .65? Our arithmetic, not the paper's: if scores are roughly normally distributed, there is about a 68 percent chance that a random person who made a plan did better than a random person who did not. It is a clear effect, not a guarantee.
When it works best. The NCI page says the effects are stronger when the person has a strong, active goal and when there is a real self-regulatory problem to overcome. When there are few barriers, a plan may be unnecessary. It also says effectiveness depends on how precise and workable the plan is.