A probe, independent from vendor and technology, decouples the path basic transmission (baseline, reference) from the mechanics of pipelines and tiny instabilities after "opening and closing" the path at cleaning/referencing causing tilt of the spectra. The reproducibility of path length is not the key problem at advanced spectroscopic technologies!
Compared to simple peak-baseline ratios at photometric measurements, some methods using much more (dozen, hundreds ...) data points of a spectrum. This "band shape" based calculations (for example chemometrics) require extreme stable unchangeable optical benches at the path.
This path length stability is easier to achieve with probes than with any kind of cells and in some application like in vessels or reactors the "probes" are without any alternative.