Shrinkage, and designing a mould bigger than the part
How much plastic contracts, and how tools compensate for it.
A mould has to be designed bigger than the finished part because plastic contracts as it cools from its molten fill temperature down to room temperature, and if the cavity were cut to the part's exact final dimensions, every part that came out of it would be smaller than intended, so the cavity is deliberately cut oversized by the amount shrinkage is expected to remove.
Knitting for the size after the first wash
A hand-knitted woollen jumper relaxes and shrinks a little the first time it is washed, and an experienced knitter allows for that from the very first row, knitting for the size the jumper will settle into rather than the size it measures straight off the needles. A knitter who ignored the shrinkage and knitted to the exact final size would end up with a jumper too tight after one wash, the woollen version of a mould cut to a part's exact dimension that turns out undersized parts once the plastic has done what it was always going to do.
Scaling the cavity by the material's shrinkage rate
Every plastic used in injection moulding has a characteristic shrinkage rate, the proportion by which it contracts cooling from injection temperature to room temperature, and that rate is well known and fairly consistent for a given material under given processing conditions. A mould designer takes each required dimension and scales the cavity up by that rate before the tool is cut, so the moulded plastic arrives, ideally, at the wanted size once it has cooled and contracted.
The rates differ a lot from one plastic to the next. ABS shrinks by roughly half a percent, while polypropylene shrinks by something closer to one and a half to two percent, so a 100 mm part needs a cavity about 100.5 mm long in one material and about 102 mm in the other. A tool cut for one therefore cannot simply run the other, and a supplier who swaps to a grade that looks similar on a datasheet but shrinks differently can hand back parts that no longer fit their mating components without the mould having changed at all.
Thick sections pull in harder
A single percentage stops being enough once wall thickness varies across a part. A thick section shrinks more in absolute terms than a thin one of the same material, and it goes on shrinking after the thin walls around it have set, which is why sink marks and warping appear at thickness transitions instead of evenly across a part. A designer faced with uneven walls has to allow for how much more a locally thick feature will pull in, and this is one reason well-designed moulded parts keep their walls consistent wherever the function allows.
It is also why the first trial shots from a new tool are treated as a measurement exercise, with parts measured against target dimensions before the tool is signed off. The shrinkage allowance is a prediction from typical behaviour, and a slightly undersized first-off part may simply mean this batch of material behaved a little differently. The usual first response is to adjust cooling time, packing pressure or cycle timing, and only after that to consider recutting the tool's steel.