By random chance, the Spy has the same attacking power as a Colonel. Either piece can eliminate 13 pieces out of 21, converted to percentage would be around 62% winning against 38% losing. Hence, using a Spy to attack any untested piece at random has the same probability of winning or losing as that of using a Colonel.
The circa 70s and 80s saw spy pieces "embedded" as a defensive cover of pro players whilst the new genre of players with the likes of Riden, has the craftiness to read and detect privates, then moved his spies as though it were a middle official or general. Combining deductive and assumptive analysis, he pinpoints not only the strong pieces but also the most likely position of your privates, then used his spies to take out the supposedly non-privates. Because privates can only eliminate the two spies and flag or splits with the same, the more likely position for these would be supportive than aggressive. Wherever the strong pieces are, you will find privates adjacent. This makes privates easier to detect than others. Wherefore, attacking the non-privates with a spy would confuse the enemy by inducing a misguided assumption of your formation. He may eventually find himself in a spy split after the initial exchange, thereby shifting the vantage of momentum and ratio to your favor.
There are only 6 privates out of the 21 pieces, one can easily use a spy to challenge pieces whether tested or not, then move such piece as if it isn't one. After the challenge, the spy would not retreat but instead get supported by another. Or the spy may move lateral to an adjacent piece provided one reads such to be a non-private. Instead of the usual boldness against another suspected enemy spy, it would feint weakness by moving away to avoid a split. The spy may likewise move and attack in conjunction with the other generals and challenge pieces as if it were a sixth general. When the enemy found it out, a lot of damage must have already been inflicted.