It's a tricky balancing act. They need to recover the investment as early as possible to pay less in capital costs but doing that will mean that later on when the product is sub-par it will cause problems and extra work.

Since the engine, game logic, art, story, testing is so heavily coupled together changing the engine a little bit could cause a month of work down the line.

I think personally the best way is to start by making an engine or taking one off the shelf and then write a mini version of the game with shit art that has a lot of bugs.

At the same time making models with hitboxes that all have the same physical properties otherwise, dialog content and recordings and all other content that can be done separately.

Once that is fun to play then you can start working creating a slightly bigger system with a single short storyline to have a cohesive experience and will have the genaral feel of the game.

Once everything above is done setting up a closed beta is the way to go. Take some feedback, add features and redo the small story to be more fun.

Then once everything is a fun experience but people just want more you do the whole everything.