In my experience BAW earns its cost when three things line up at once: a steep skirt, a tight power and temperature budget, and a frequency where SAW is running out of lithography headroom.
The scaling argument is the clearest one. A SAW device works with surface waves, so its finger pitch shrinks with wavelength; above about 6 GHz the fingers get very fine, power density rises, and insertion loss and power handling both get worse. A BAW resonator works in thickness mode, so its critical dimension scales differently and it keeps a higher Q at the same frequency. That higher Q is what buys the steep transition band and the low loss in a narrow channel, for example in Wi-Fi 6E/7 front ends and in n77/n79 duplexers where the transmit and receive bands sit close together.
The counter-case is just as real. If the fractional bandwidth is wide, the rejection requirement is moderate, or the blocker level is low, a temperature-compensated SAW or a hybrid topology often meets the system mask at a fraction of the cost. I have also seen designs where a SAW filter plus a notch or a better LNA linearity budget was cheaper than moving the whole filter to BAW.
My practical rule is to write down the numbers before choosing a technology: required insertion loss, rejection at the nearest blocker offset, transmit power, and drift over temperature. If the skirt needs to fall tens of dB within a few percent of the centre frequency, or the SAW option is already at its power or temperature limit, BAW is the honest answer. If none of those limits is binding, the cost premium is hard to justify, and I would spend the money on the surrounding matching and shielding instead.