Just imagine. You use Duolingo to learn a language and it shuts down. What do you do? What's the export format. Will the new product have the same levels / trees/ etc? Very unlikely. You'll have to start all over again. It's not about the raw data, it's about the data that your work has generated.
If you're doing machine learning, then there's the raw data, but there's also all the other data resulting from your hard work: training jobs, experiment tracking, parameters and artifacts and metrics, models.
The question becomes... First, how important is your product in the user's life? If it's important, they'll rely on it a lot and use it a lot to do things that are important for them, or for their organization.
Then, the service needs to provide export options and send you a shutdown notification. These are not always true but, assuming they are, you'll have to scramble to exfiltrate your data.
Then when you exfiltrate your data, it's an export. You'll have to find another product or service, and then somehow hope you can import that data to the new product, and then set up another workflow.
And here I'm talking about consumer products. For enterprise, there are many parameters including who the vendor is, procurement, team size. Meaning if you are Google/Amazon/Microsoft/etc, then a lot of people will treat you like a utility that will continue to be there, even with the track record of some of these of shutting down products and services (how many products has Google shut down over the years).
Look at this thread[0] I posted about our machine learning platform and the conversation that followed. The person objected that as long as it wasn't open source, they wouldn't use it. But digging a bit deeper, it turns out what they wanted wasn't really open source, but open source was one solution/implementation to their underlying problem. Trust you'll be there or control.
- [0]: https://old.reddit.com/r/MachineLearning/comments/kolobf/p_c...