Exponential backoff is a retry strategy where the wait between attempts grows exponentially — 1s, 2s, 4s, 8s — instead of retrying immediately. Adding random jitter spreads retries out so many clients do not retry in lockstep.
Worked example: on a 503 or timeout, a client waits base times 2^attempt with a random jitter added, giving the overloaded server room to recover instead of a retry storm. Gotcha: naive fixed-interval retries from many clients recreate the thundering-herd problem and can keep a struggling service down — jitter is not optional; also cap the retry count and the max delay, or a permanently-failing call retries forever.