if your algorithm guarantees a confirmation within the next block, your algorithm should suggest a fee which does exactly that. Slowly incrementing the fees for the algorithm simply won't work and would probably result in far longer wait before a confirmation for most. There should not be a scenario where the algorithm suggests a fee that risks not hitting the target, some services doesn't really have a choice; CPFP makes the TX unnecessarily expensive, RBF is not an option.
I disagree on the CPFP part. If - say - Binance now pays 100 sat/byte for a transaction with 100 outputs, they could easily to for 10 sat/byte and still get into the next block
about 90% of the time. For the 10% they don't make it, they could instantly do CPFP onces fees in mempool go up. If they quickly bump the average fee to 25 sat/byte, they can still make it into the next block, and depending on the state of mempool, they can increase fees by as much as needed.
The cost of an incidental CPFP is tiny compared to the amount they save on total fees.