We propose FCGrad, a conflict-aware gradient adjustment method that improves social welfare while ensuring fairness in multi-agent mixed-motive settings by resolving conflicts between individual and collective objectives.