1) Specify the requirements, like accuracy, bandwidth, resolution, etc
2) Consider how you are going to interface the device to the real world
3) Consider cost. In reality, if you are doing it just as a hobby project you can ignore this one, but for anything you want to sell, this part IS the most critical part. Generally, you want an electrical architecture that juuuust meets your requirements set in 1). In particular you need to know your expected volumes
In every project i've ever done on a commercial basis, a significant of the "Nice to have" specification has been nullified by cost implications! Saving £2 on the BOM might not sound like much, as might saving 5min on the assy time, but make anything more than about 100 of them and it starts to make a BIG difference!
Also, for industrial usage, generally, getting minimum power consumption isn't as important as keeping costs down (unless you have some very specific energy supply constraints). Especially true for low power devices like a load cell amplifier, generally a basic, cheap (and quiet) linear power supply section, which is admittedly wasteful, is often the correct choice (you need to do the power budget and thermal calcs to validate this decision)
For example, you might find the 12b ADC on the STM32 is actually sufficient, especially as if you have a low bandwidth requirement you can do significant oversampling as the ADC is fairly fast on that chip.
IMO, spend the money on decent analogue front end components and pcb design, and minimise the digital side of things using the KISS principal.
Of course the other critical bit of research is to investigate what is already in the market, how much it costs, and what specifications it has. Use this information to taylor your device, either to offer a better functionality or accuracy, or a similar capability but at a lower price.