A global retail application uses Spanner for inventory reservations. Most writes originate in North America, but European shoppers must see low-latency reads. The team chose a multi-region instance and asks how to reduce write latency without violating consistency. What should you evaluate?