In general Vietnam is seen as a total defeat of the McNamaran approach to statistical approach. Things like war do not follow simple mathematical models (you can count casualties, tonnage destroyed, etc., but it turns out "winning" is more complicated than that), and many of the things you think you can count turn out to get "juked" when you alter the incentive structure (if you make everything about "body count" then you will start seeing juked statistics very quickly — this happens in pretty much every realm of human activity where simple metrics are incentivized and it is hard to tell whether they are "real" or not).

Paul Edwards calls the McNamaran fantasy the "Closed World," a world in which massive amounts of data, coupled with good models and strong computing capability, can translate into control. It doesn't work that well: there is "pushback" from the "Green World" (as McNamara puts it), which is to say, the people who live outside of the "Closed World," and the model of the "Closed World," like all models, is only a partial representation of the world. The system ends up measuring what is easily measured, and ignoring what cannot; this leads to massive systemic errors over time. When it came to Vietnam, again, by the McNamaran metrics the US was vastly "winning" the war. But in reality these approaches led to deep alienation of the South Vietnamese population, entrenched will by the North Vietnamese, and a gradually increasing sense of futility among the American public — a recipe for total failure.

For a really nice discussion of the role of computers in particular in the construction of this idea of the Closed World, and its relationship to McNamara's policies in the DOD in general (not just Vietnam), and the ways in which these approaches led to situations that are obviously not that desirable in retrospect, see Paul Edwards, The Closed World: Computers and the Politics of Discourse in Cold War America (MIT Press, 1996). It gets into the appeal — and failure — of the McNamaran approach quite well. The main problem with mathematical modeling of complicated social problems is that they rarely are capable of dealing with the complexity of said problems, and your model is only as good as the modelers. The track record, while not impossibly poor, is not that great.