I think we might have a chicken vs egg question here. Let's try few different ideas that I think might be interesting to contemplate.
1. Seven Kingdoms had an interesting model where each city had a limited population. This population could be tasked to one of several different tasks, they would gain experience in those tasks making them more capable. As an aside, espionage meant sending "immigrants" to the opposing village and having them become part of their population. Amusingly, appointing a spy to a general or a governor could turn over control of the fort or the city.
The practice would be to model several different skills for each "Charismatic Leader". Each city has a few leaders that "learn" skills. I can build a unit of soldiers with one leader. As they train at facilities, the leader and his troops becomes better skilled. (I'd think of them as the captains of a company.) If he trains with Infantry, his troops get better with fighting as infantry. If he is trained as a archer, his company will become better at fighting from a distance. History and fiction are full of sending out barely trained people to die on the battlefield. The inherent balance is that you need to have time and food to feed troops.
As an aside, you set up skills for: Farming, to give boosts to food production for that city/town. Trade, to give boosts to economy. Woodworking/stoneworking to give boosts to construction or durability. Espionage/ Counterespionage would end up being a cat/mouse game. Do you trust your new immigrant with maintaining your walls without a keeper?
Max skill for anything could be a result of tech, or maybe mana.
One interesting factor for this mechanic leads to skilled labor sending caravans of products, rather than materials, to less developed 'Frontier' states. Instead of sending metal, your now sending swords and armor that can be raided and amusingly used against you. You can setup assassinations and CRIPPLE a city or productions by killing the smiths and bowwrights.
2. The Sims MMO had an interesting balance mechanic. Basically, no one could master everything. Ypu had 10 (or 15) points to spend on skills. After you used up your points, trying to improve one stat caused something else to degrade, usually your best skill.
The mechanics behind this idea is that you set 100 (or 1000) points of focus on a set of sliders reflecting different social agenda such as infantry focus or trade. You can set your sliders to however you want in the first turn but, each turn afterwards can only alter a particular focus by 1 percent. Figure weekly turns and you're now trying to anticipate what you'll need months down the road. To possibly complicate matters slightly, once you get past a certain amount of focus, you might unlock "hidden" sliders. For example, if you improve your focus on infantry past 50%, you can now dedicate yourself to specializing specific troop types such as light infantry, heavy infantry, or Ninjas.
Note that this idea is global and applies a constant pressure on the troops in the field. All Infantry will become less effective as the focus moves away from them to Archers. This tends to model the idea that individual troops require more than a single sword, or pair of boots.
3. Since each faction will have unique trees and spells, simply make different troops require or benefit from different raw resources on the map. To produce troops for the Fallen Juggernaughts, they REQUIRE a source of Bronze. To produce Human Britons, you need lots of wood. Give Britons a food bonus from Forests and they'll be spreading trees across the map. This is the MOM approach.
