The first step of the Wittig Reaction – attack of the phosphonium ylid on the carbonyl – can be reversible or irreversible depending on the nature of the reactants, and this can have an important effect on the stereoselectivity of the process (if both reactants are asymmetrical).
If the ylid is stabilised in some way – usually by an Electron Withdrawing Group as a substituent, then the first step will be reversible because the ylid is a good leaving group – e.g.;

Note the orientation of attack, with the phosphor and oxygen away from each other – this minimises steric interactions.

If the ylid is not stabilised, or even destabilised (by an Electron Donating Group), the first step is irreversible because the ylid is not a good leaving group;

As both the ketone and ylid are asymmetrical, there are two possible orientations of approach and attack, i.e.;

These two approaches are energetically inequivalent – the first one has the least steric interactions between all substituents (it puts the small H next to the Phenyl, and the small-ish Methyl next to the PhNH2), so this will be the most likely mode of approach. However, once the two reactants have ‘stuck together’, there must be a rotation about the central bond so that the O can attack the +PPh3. This is shown below;

Now the bulky substituents (Ph and Ph-R , where in this example R=NH2 or NO2) are directly next to each other – clearly an unfavourable conformation. Because this intermediate conformation is so unfavourable, the betaine with the stabilised ylid is more likely to dissociate than to adopt it.

However, the betaine with the destabilised ylid cannot dissociate and so must continue through the disfavoured conformation to the kinetic product.

Even though a large proportion of the stabilised ylid will attack in the more favoured manner, some of it will attack the other way – and when the central bond rotates the conformation will be much less crowded;

Thus the stabilised ylid will react to give the less crowded transalkene – the thermodynamic product.