Recovering a late project
When a project is late the instinct is to accelerate whatever looks most troubled. That work is frequently not on the critical path.
Most recovery effort is spent in the wrong place
Shortening work that is not on the critical path changes the finish date by nothing at all. It costs real money and produces real activity, and the date does not move.
This is invisible without the arithmetic, which is why it keeps happening. Measure every proposed action against a recomputed critical path and the wasted ones identify themselves immediately.
The four levers and their real costs
Crash — buy time with money. Predictable, and the marginal day gets steadily dearer as the cheap options run out.
Fast-track — overlap work that was sequential. Often cheap in money and genuinely risky, because the overlapped work proceeds on assumptions the earlier work was going to confirm.
Cut scope — the only lever that reliably works, and the only one needing someone with authority to say no. It is also the one people reach for last.
Move the date — always available, rarely popular, and frequently the cheapest option once the alternatives are costed honestly.
Adding people to a late project
Adding people mid-project costs weeks of onboarding, drawn from exactly the people who are already the bottleneck. Productive capacity often falls before it rises.
Any model will happily show the schedule gain and none of that second-order cost. Treat a recovery plan resting on a substantial team increase as optimistic by default.
Sustained overtime is borrowing
Overtime buys real days in the first fortnight. Beyond that, output per hour falls, defect rates rise, and the recovery plan starts generating the rework it was meant to avoid.
A plan resting on months of overtime is not a recovery, it is a deferral — usually into a worse position than the one it was fixing.
Do not let a tool score the risk
Fast-tracking rework, overtime fatigue and onboarding drag are real costs that arithmetic cannot weigh. A tool that folded them into a single score would be inventing numbers, and a recovery decision made on invented numbers is worse than one made on none.
Record them as notes against the option and let the people who know the work weigh them.