Would you use an NRF chip today, or is there another you would recommend? I'm working on a board with an NRF52840 and have been annoyed with their API. There are so many versions of their API and software, some are marked deprecated but the interface is still used in their latest code/examples. Nordic also doesn't provide a code generator to configure the hardware. I started with them since their chips seemed popular, they provided the BLE profile I needed, and come with a Segger debug license.

There is not a huge range of choices. I considered TI's chips and worked with a student who had some measure of success using BT on a dev board. I would recommend looking over the support forums for any candidate chips, you will quickly get an idea of what the development experience is like. If the functionality you need is simple (e.g., just sending periodic data from a sensor) then something that wraps the SDK's API and provides a higher level simplified interface would be faster and easier to use (Arduino, MBed). If you need low level access, complete control, and special functionality there are few choices, namely the big chip companies that make BT chips and they all seem pretty bad. My theory is that Bluetooth is so hugely complex as a barrier to market entry (also companies like Nordic want to sell you consulting services). We wanted simultaneous Central and multiple Peripheral channels (a connection to a phone and to multiple sensors) so our only choice at the time was use two chips or use Nordic, and a smartwatch has tight space requirements. In general if you just need radio and it does not have to be Bluetooth I would avoid Bluetooth. ANT was a real pleasure to work with and has some nice development and testing tools and there are some phones with ANT radios (Sony if I remember correctly). Zigbee might be another alternative.