Hi Korma, I have had the best results "building" a light fixture that mimics the way actual spotlight fixtures work: they consist of a globe (light source), a dish reflector behind the globe, a simple cylinder for the overall housing, and an (optional) fresnel lens in the opening for the light to exit. If you make the interior of the housing white, you can get away without the reflector. The key is the distance of the light source globe from the opening of the housing. Closer to the front opening the spread of the light will be broad; the closer to the back of the housing the beam will be more focused. I rigged up some simple Xpresso to control this relationship (spot to flood). You can also go crazy and fashion bard doors or use digital versions of flags, scrims and so on (it helps if you have real-world lighting experience.) For the light emitter, you can use an area light (circle-shape) or fashion your own emitter polygon, which some argue can be tuned a little more than the standard Octane lights.
Of course, if you want to "see" the beam, you need a volume medium of some kind. Elsksa described the Octane "spotlight" and how it works and you could do the same. It is common to put the volume around the scene components, and others have done so just around the lights. If the overall illumination is dark, you might be able to get away with that, but if you have a lot of bounce lighting and your overall illumination levels are brighter, you will need an encompassing volume.
Note that the volume medium will be the noisiest in the brightest part of the cone. In fact, most lighters will not use that portion of the cone for illumination in the real world, as the center area of the cone tends to be the sweet-spot. A good read up on set lighting technique is a good idea. Octane is physically-based, so mimicking real-world techniques is the best way to go. This technique will also play nice with other lights that you might have in the scene.
In terms of sampling, you will have noise with the volume medium. That will take samples to deal with. I have found that that PMC kernel is the best for a pleasing spotlight cone, but it is also by far the slowest of all of the kernels. If you are animating, the PT kernel would be better for time (though noisier) and you could work with the denoiser settings to get a look. You are going to need a good amount of samples to resolve the noise to the point that the denoiser can give you a cleaner result.
Here is the resulting image:
