How to make AI product implementation at large company successful?

As a product person building AI solutions for the last 10 years, I strongly believe a great product implementation of AI beats having the best models.

You can have the best of models, finetuning, and AI engineers but without a great product problem solving, the results don't come.

So here I am asking

What have been your key learning while building and productising AI solution? How to make it successful?

For example, for me, starting with the end-product in mind and working backwards worked best. I start with how end-product would look like, success metric etc and then start building. Have been able to make 5-6 successful transformations using this.

-Ayush