Whilst Mauro's answer details helpfully how Anscombe derived the definitions she did, it does not perhaps answer the second part of your question, which is the "point". The reason why Anscombe made such a distinction is that she was trying to reject determinism to an extent and considered the intentional reason to be a thing which is known without observation. Thus the distinction is required to support her position.
There are flaws in the argument, however, which may explain your uncertainty. To take Anscombe's example of the man raising his arm up and down. If an outside observer had access to the pre-motor cortex of the agent's brain, they would be able to deduce with absolute certainty that the agent was about to raise their arm. We could extend this beyond current neuroscience, but still within the realm of possibility, to Anscombe's shopping list example. If an outside observer had access to the neurons which represented the agent's image of the state of their store cupboard, and those containing the concept of the desired state, then they could derive a shopping list of intentions in no different a way to that in which they derived their shopping list of actions. We then end up with a description of an identical item of knowledge which is different depending on whose brain it is in, which is un-generalisable.