governato's example is a nice one, and bears some truth. Numbers alone don't tell everything in this game, sure, but Axis/Soviets can use them nonetheless as a rough orientation for the choice of strategy. Get the Wehrmacht to 2.8M men or below, and you can start pushing more recklessly. Get the Soviets to 3.8M and risking more by holding ground during blizzard 41 can become a feasible and sensible thing to do.
You could cloud the numbers, but soon a spreadsheet would pop up where players would enter the turn's losses, destroyed counters, and manpower centers, and people would keep track of numbers probably rather well. You'd also need to significantly "FoW" losses, probably all except captives. Which should be, as even today the total losses on both sides are a much-discussed and studied "unknown". Would make sense to me.
Another thing is the historically accurate setup. Just like in AE, it is a beautiful beast to have, and a quality mark, but it leads to predictable optimization of campaigning. After a few man-years of AARs, one knows when and where the units are used to be best effort, just like the Lvov openings. There are alternate starting scenarios for WitE, just like for AE now, that partly address this.
However, one concept that would be even better for "pure land-warfare where is attrition very important", would be some randomization just like in the old days of C64 games. I remember for example Rebel Charge at Chickamauga, where reinforcement times could be randomized by up to 6 hours, which made huge differences for your strategy during the two days of fighting. Similarly manpower pools, centers or the reinforcement/withdrawal schedules (and even amount of withdrawn units) could be randomized a bit. Then there'd be no waiting off and on for the withdrawal of SS units, and more uncertainty as to be (growth) power of the two Armies. Each side would be free to under- or overestimate, and strategies would become much more diverse I think...