A warm-start approach gives a new decision option a head start by borrowing knowledge from similar options, users, or prior models instead of learning from zero. In contextual bandits, it helps new ads or customer segments compete fairly during cold starts without letting established choices bury them. section 5.6 section 5.7
Launching a new ad creative
When a new ad creative has no interaction history, initialize its decision model using knowledge from similar ads or users. This gives the new option a fair chance while the bandit continues exploring and learning its actual performance. section 5.6

