Since Energy Conservation efforts are so important, it's paramount that we join in by ensuring our BXDFs do not waste any energy, we don't want anything to go unused.
On that vein, there's a lot of work that has been done to keep PBR materials energy preserving. Work that usually involves metals and dielectrics.

But what about the in-between? Can those materials be energy preserving? Probably the solution is widely known but I couldn't find it, so here we go.
Let's step back a little and see what happens with those two types of materials and think on how we can come up with a feasible option.
For dielectrics we can think of a clear material that has all these particles inside packed together closely. We like to think that they are so close that we model them as a diffuse lobe and the outer part as a specular lobe. So when light hits the surface, some of it will interact with the surface and other is refracted to the diffuse lobe. [1][2]
Metals (conductors) on the other hand have this property where all the light interactions happen on the surface, light is reflected or absorbed, nothing goes below so no diffuse lobe. [1][2]
As for the in-between materials, that I'll call Semi-metals from now on, we introduce a new parameter called metalness that goes from 0 (dielectric) to 1 (metal) and it is used to change the material in the following way [2]:

F0 is plugged into Fresnel and you can notice that as the material gets more and more metallic the diffuse lobe is scaled down as more absorption happens at the specular lobe. How does this look?

So if we want physically plausible materials, one requirement is that energy that comes in should come out, no more, no less. We can think about that interaction using the following equation:

How can we check for something like this? We have the trusty white furnace, in it some geometry with a perfectly reflective material surrounded by a white environment should completely disappear, think about it, no energy is absorbed or emitted by the material, so all of the light bounces and it will equal the color of the environment.[2]

Left image shows an energy preserving material (I swear there is a sphere in there), middle one is emitting more light since there is a lighter halo and the right one is absorbing light as can be seen by the darkening of the sphere.
Unfortunately by themselves specular lobe and diffuse lobe when added naively are not energy preserving, either they absorb energy or emit energy, that's not good. Fortunately a lot of work has been done to tackle this issue, like Kulla and Conty[4] or Turquin[3]. I'll be basing the following nonsense on Turquin's work since those equations are less involved and the results are quite good.
Here's the equation we can use to have energy conserving metals[3][5]:

Where ρss is a single scattering of our specular BRDF. Ess is defined as all the energy we'd get from all the single scattering interactions on the BRDF, F is set to 1 since we don't want any absorption, light is also 1 in all directions (basically a white furnace) yielding [5]:

Where G is the shadowing-masking term and D is the Normal Distribution Function.
How do we deal with dielectrics? Remember or energy conserving equation? Well, shuffling around some terms, we can come up with [5]:

(1-Espec) scales down our diffuse lobe based on how much energy interacted on the specular lobe. Here Espec is slightly more complicated to compute than the previous Ess since we need Fresnel to be involved, and we get [5]:

Now that we know how energy conservation can be attained for both metals and dielectrics, let's go back to the topic that brought us here, semi-metals. How do those fare in the white furnace with no energy compensation?

Erm, not good. Notice that on low roughness we have extra light coming from the edge and on high roughess light is absorbed. Now the question is, how can we use the aforementioned equations for semi-metals?
If we just plug the energy conserving equations naively, say we do this:

Then we get this:

This is bad!! At least dielectrics and metals at each end of the graphic look good, but what happened in the middle? Why is it darker? Well, we are scaling down the diffuse lobe by both (1-metalness) and (1-Espec).
So what we need to do is think of what we are trying to accomplish. As we decrease metalness, we are letting more and more energy to go to the diffuse lobe, indeed there is some scaling happening, how can we properly scale the diffuse lobe considering both metalness and energy?
Accounting for how much energy is reflected or refracted sounds a whole lot like Fresnel, so let's exploit that fact and introduce a new parameter that we will call F0metalness and is calculated as :

Which we can use when calculating Espec in our Energy Conserving equations and using the resulting term as the scaling factor for our diffuse lobe instead of (1-metalness) . Notice that we are not using reflectance/albedo yet, we'll come back to it later. And the working equation looks like:

Interestingly when metalness is 0, the equation is the same we saw earlier and when metalness is 1, F0metalness = 1 yielding Espec = 1 , leaving us with the known equation for metals:

Tying the whole math model together.
How do we account for absorption at the specular lobe? In other words, how do we plug back the color of the material? We can do the following:

How does this fare in the white furnace?

And how does this look like? aka pretty pictures plz

When we compare methods we can see how the energy compensation takes care of the over brightened edges on low roughness and brings back energy on higher roughness:

And now here's how energy compensation looks like for roughness and metalness:

As for implementation, there are two equations for which we need either a LUT or a numeric approximation, namely Ess for which parameters are roughness ∈ [0,1] and wo·n ∈ [0,1]. For Espec we need a 3D LUT which requires roughness, wo·n and also Fresnel ∈ [0,1]. [3][5]
Footnotes
[1] Pete Shirley's Graphics Blog - Fresnel Equations, Schlick Approximation, Metals, and Dielectrics
[2] Crash Course in BRDF Implementation, Jakub Boksansky
[3] Practical multiple scattering compensation for microfacet models, Emmanuel Turquin
[4] Revisiting Physically Based Shading at Imageworks, Christopher Kulla & Alejandro Conty
[5] Enforcing Energy Preservation in Microfacet Models, Davide Sforza and Fabio Pellacini