The move that works, and it is not leaving.
Build the direct channel first, then reduce the platform gradually. Leaving abruptly loses the volume before you have anywhere to put it, and that is the failure everyone has watched a neighbour go through.
What actually shifts people, in rough order of effectiveness:
Something in the bag, every single order. A card with your own ordering link and a reason to use it: a small discount, a free item on the first direct order. This is the highest-return thing available to you because it reaches exactly the right people at exactly the right moment, and it costs pennies.
A direct price that is visibly better. If your platform menu is priced to cover commission, you can be genuinely cheaper direct and say so. Customers understand this now.
Make direct ordering easy. A simple ordering page and a phone number that gets answered. Most people will use it if it exists and is not painful.
Keep the platform for discovery. That is what it is genuinely good at: new customers who do not know you exist. Treat it as an acquisition cost for those, not as your ordering system.
Give it six months and measure the split again. The goal is not zero platform orders; it is that the platform stops being where your regulars order from.