Warfare is very much about information asymmetry. A small informational advantage can often be levered into a large tactical advantage. This is nowhere more obvious than real-time strategic games like Blizzard's Starcraft. Although Actions-Per-Minute (APM) is one of the primary metrics used to measure the strength of a player, arbitrarily fast clicking will not overcome a tactical mistake based on poor information. In the game, one must weigh whether to move forces to raid an opponent's base, or hold them back to mount a successful defense. Whether 'tis better to attack or defend depends on what the enemy is doing and planning on doing in the near future, so scouting is critical in the game, as in real life. Moving your forces takes time, so abandoning your base to go on offense at the wrong time might mean losing an outpost when the enemy is able to also retreat and defend because they know you are coming.
Ultimately, there are 3 levels of analysis which may be considered in a game like Starcraft. The tactical situation involves the question of: "What is the enemy doing right now, when he is in my view?" During a pitched battle, the enemy may deploy various tactics like activating special weapons, focusing fire on various targets, feinting, making surprise attacks from different directions, etc. Except for cloaked units, the information at this level is complete, and standard game theory can be applied to predict what an optimal enemy will do, and the optimal response. While humans develop an extremely high ability to execute well in this environment, a computer can likely do better because of "infinite APM", etc. A real army will not move as precisely as a Starcraft army, and thus, a real general will have more difficulty executing a plan. So knowing your own capabilities is almost as important as knowing the enemy.
Above the tactical level, we have "general strategy", or "the meta-game". This involves questions like: "How quickly should I develop technology vs. build units?" "When should I expand?" "How forcefully should I defend expansions?" These questions depend on what the enemy is doing, which requires scouting. And scouting cannot gather complete information, so predicting what the enemy is doing at this level is much more difficult. However, there are several "standard" strategies, from rushing to turtling which can be recognized by the choice of various actions (which buildings did the enemy construct, where did they place them, etc.). Which of these strategies is popular in any given month changes as top-level players try out new techniques and counter old ones. This shifting of high-level strategy is called "the meta-game". This is analogous to generals knowing about trench warfare or siege warfare or combined arms tactics, etc. They are high-level techniques which can be recognized on the battlefield (or the build-up to war) if you know what to look for. Although there are an infinite number of high-level strategies, the vast majority of them are bad ones (do nothing for X minutes/days/years, etc.). The few good ones are quickly discovered, learned, executed, countered, and evolved. Any good analysis needs to keep up with the meta-game against potential adversaries, or it will not be able to anticipate anything. The weakness of the meta-game is that the best strategies involve tight constraints. You need to build certain things in a certain order to obtain the optimal outcome. So they have a signature appearance which can be detected by the informed observer. There are branches in the strategies which give some flexibility, but all choices commit you to a dwindling number of final paths.
Finally, the best players in the world have personal styles which compete, merge, mutate, and become the meta-game. But because they are the major innovators, they are the ones to watch to find out what players in the lower leagues will be doing a month from now. So we move from anticipating enemy actions based on well-known strategies to making predictions based on the observed actions of a single well-known adversary. This is the top level of analysis. It is both the strongest and the weakest level of prediction. On one hand, players are victims of their own habits. This makes certain things they do predictable. This is as true of entire nations and armies as it is individuals. On the other hand, the best players are always trying something new. This makes them inherently unpredictable, which tends to defeat predictive analysis.
Now, how a "computational stratego" might aid a future military depends on who the enemy is. If the enemy is unknown and makes a surprise attack, all bets are off. If the enemy knows you and your weaknesses, they are likely to prevail (or they would probably not have attacked if they knew they had an informational advantage but would still predict loss). If the enemy is known but makes a surprise attack, then the tactical predictor can still help anticipate small-scale actions and advise on best counter-measures (retreat, fire weapons, maneuver, counter-attack, etc.). If the enemy is known but conflict is not started with a surprise attack, and disposition of forces is mostly known, then the strategic predictor can probably anticipate most likely actions (because again, there are only a few optimal strategies, for most forces). But if the enemy is known intimately, and a conventional war breaks out, then the enemy can probably be modeled with high precision, especially if they have attacked other adversaries with a consistent strategy.
At the end of the day, every human being is a predictive modeler. One could argue that this is, in fact, the very definition of intelligence: to predict the near-future and optimize one's behavior accordingly. So a science or a system to do this is not going to do fundamentally better than a good military leader, except that they may do so with less bias and perform precise calculations more quickly. It will not become emotional or stressed, or make impulsive decisions. Ideally, it will actually do one thing better than humans could: react to a surprise. After all, an enemy is only going to attack because they have overwhelming advantage, and it doesn't matter what you do, because they are forced into this position and have no better alternative, or because they can exploit an information asymmetry and hope you do not respond quickly enough.
If you can predict overwhelming enemy advantage, then you have no counter-move, so psychohistory is useless. If you predict that the enemy is cornered into a particular move, then further prediction is trivial. If the enemy surprises you with an informational advantage, then by definition, you cannot predict this move. Thus, fast reaction is better than prediction.
But at the end of the day, information is what wins wars. This is why the "shock and awe" strategy of modern US military doctrine calls for disabling command-and-control capability first. This is why an unarmed drone with a camera is a big force multiplier for ground-based fire teams. This is why a strong intelligence network is better than a bunch of ballistic missiles in silos. Israel has wreaked more havoc with spies and assassins than any country has launching ICBMS. Everyone talks about China's million-man army, but the country's hackers report to an army general. Which do you think is more powerful today? Prediction cannot discover the secrets that an adversary is actively guarding and developing; only espionage can. Psycho-history is really only useful for slow-moving, non-adversarial targets, like civilian populations.
But to the extent that the enemy is purely human-powered, a tactical predictor can probably anticipate faster than a human tactician, in the same way that computer-controlled anti-aircraft guns are better than human-controlled ones. But this is a very trivial level of prediction, and not really what we mean by psycho-history.