Crawl budget is not something to minimise at all costs: a blanket Crawl-delay for every user-agent also slows legitimate shopping and comparison agents, so products get discovered later. A steadier approach is to group requests by user-agent, see which paths are expensive, and throttle only those paths while leaving product pages reachable.
Keep the rules explainable in robots.txt: which paths each user-agent group may take, why a limit exists, and when it will be reviewed all belong in a change log. After each edit, re-scan the home page and a few key product pages with JavaScript disabled to confirm products are still in the crawl result.
This article covers crawl reachability and rate configuration only; it makes no compliance judgement about any measured traffic figure. Rate changes are cheap config edits, and precisely because they are easy to change, each one should be dated and traceable.