This direct correlation may not be causal. When we change the input we see either no change in our output or, worse, a change in the wrong direction. By analysing the whole system we better understand the real causal links and change causal inputs rather than correlated inputs.