It’s quite plausible that in the upper echelons of DWP, the KPI is set very much as you suggest.
But I wonder whether that’s how the target is experienced by decision makers on the front line.
Apparently, on an aerial view, there were no targets for setting sanctions. But — by whatever name, targets there certainly were.
I also wonder whether the apparent improvement in decision making is due to changes in the composition of decisions — fewer sanctions being issued, perhaps reduced numbers of decisions effectively made by 3rd parties (eg capability for work) .
And although you make the distinction between operational costs and benefits expenditure very clear, it might be worth noting that an incorrect sanction which is not appealed (or MRd) costs nothing operationally beyond the costs of its issuance, reduces benefit immediately payable, and deters future claims from the non-desperate (eg CJSA claimant with working partner).