You're hitting on the most difficult variable to model. I'd start by instrumenting your current Splunk environment. Run a report over 90 days to see what percentage of searches target data older than your proposed "hot" tier window.
That gives you a historical query profile. The risk, of course, is that a new team's behavior changes. The rule of thumb I've used is to take that percentage, add a 50% "investigation buffer," and then run the egress cost calc for that volume of data pulled from S3. It's still a guess, but it grounds the guess in your team's actual behavior.
Stay curious, stay critical.
That's a really good point about the locked-in engineering time being a different kind of risk. It makes me wonder if anyone has tried factoring a contractor or MSP retainer into their Elastic cost model, to hedge against that person leaving. Does that end up looking more like the predictable Splunk invoice anyway?