How do you reduce EC2 costs for a service with predictable traffic vs spiky traffic?

It is also worth mentioning that when it comes to infrequent traffic spikes after long idle period and/or short lightweight API requests, it might be feasible to replace EC2 instances with Lambda functions

For predictable traffic you can follow below

Baseline covered with Savings Plans + Auto Scaling for minor variation.

For Spiky Traffic you can follow below

Small on-demand baseline + Auto Scaling + Spot capacity for bursts.