Self-esteem is an evaluation: how good I think I am. The structural problem is that any evaluation needs evidence, and evidence fluctuates. Esteem built on achievement rises and falls with it, which makes it an unstable foundation at exactly the moments a foundation is needed.
Research has also complicated the “low self-esteem causes problems” story. Roy Baumeister's 2003 review found that high self-esteem does not produce better performance, better relationships or fewer risky behaviours — the correlations often run opposite to the assumed direction.
Hence the shift toward self-compassion in Kristin Neff's work: instead of “I am worthy”, something closer to “I am human, and humans fail”. Self-compassion needs no comparison with others and does not collapse at the first failure, because failure is precisely the occasion on which it applies.
Practically: low self-esteem is not repaired by positive affirmations. Someone who does not believe they are worth much rejects the affirmation as false, and the exercise confirms the discrepancy instead of closing it.