None.
They fail because they didn't talk to customers, didn't do sales, didn't do marketing, or failed to find a viable service model after all that.
MVP is shorthand for "do the least work customers will pay you for." Customers are shorthand for "people you can find who will pay you."
MVP is quite often abused to mean "here's a thing I made." If you don't have customers, and you can't find customers ... then it's not viable yet, is it?
Here's a thing you can do:
You have a hypothesis that [service you can provide] will be valuable to [people you can find]. Okay, so go check that you can actually find those people. Then talk to them about what they do.
You are not allowed to sell, or ask "would you buy this." Facebook surveys are also right out. You need targeted customers, not a bunch of random people with few common characteristics besides being bored enough to respond to Facebook surveys.
Your objective is to learn about these people. How do you find them? How do they talk about the problem? How do you tell them apart from people who really don't give a shit about the problem?
Once you've got a target customer validated at the talking-to-people level, then you can try building stuff. Start with the least work that you think will get people to pay you. It's probably much smaller than you think, and may involve no code.
You don't need to solve everything that anyone told you about the problem, you just need to do enough to make their corner of the world a little better.
Then, go back and try to do sales. Face-to-face sales is the highest conversion rate condition you're ever going to have access to. It doesn't have to scale to infinity million dollars of revenue. It's validation. (Although it's also quite possible to get to stability on boot sales alone, if you charge enough. Which is the only time I ever want to hear that "with no marketing" meme.)
If you can't sell when the deck is stacked in your favor, you don't understand your customers, or you're working on something they don't actually care about.
Treat stuff as a hypothesis. Then test it. Don't just throw code at the wall either. Code without a customer is not a product, it's just some thing you made. Figure out who the customers are and aren't. Then try to find them & talk to them. Learn about them. Then build.
Nobody failed because their MVP was too minimal. They failed because they didn't have customers & didn't do the work to fix their understanding of the world.
I should add you can still fail even if you have customers. In someways having customers can be worse than having none. If you have no customers you know what they problem is, but if you have customers and no growth then it can be hard to work out what to do.
Yes, you can be short of desirable features for your customers and fail; but the real failure is that you didn't talk to your (potential) customers to identify this gap. In other words, you shouldn't be adding more features if you think your MVP is too minimum - you should be talking to customers to really find out.
You want to study the people & problems you're targeting to the point that you become the expert they're trusting for advice about the problem. This is hard, but not that hard - minimum viable expertise is a thing as well.
For example, not everyone who uses email as part of their hiring process has spent an enormous amount of time studying how to improve contact-to-interview conversion rates. Specializing on this & studying across N people who are a close match for your target customer can quickly give you working expertise on the problem, because everyone else is worrying about 10 other things at the same time when working with it, while you can go deep on that one thing.
Then based on your understanding of their workflow & what ought to generate value (as obtained by talking to those people in a learner mode vs. a sales mode), you generate a hypothesis and test it. If it actually helps deliver value, and the problem actually matters, and you actually understand these people well enough to get them to trust you...congratulations, you might have yourself a viable product.
For the above example, you need exactly zero unique software to pull this off. You can start as a completely manual service with direct sales only against the people who you think will be most likely to buy. Give yourself the easiest win conditions you possibly can, requiring the least time and investment. Then, build on that, add automation as appropriate and test that the value is still delivered, try to scale, etc. - but start with the scaleless version, it ships easily and then you have actual customers who are paying you.
Unfortunately not all problems are amenable to “stubbing out” with human labor. With my product every new feature has to be built first and then introduced to the market - sometime they stick and other times they fail to get used despite being really great (customers aren’t always rational).
You layout a glorious strategy for proving/disproving a hypothesis. As Steve Blank said, "get out of the building [and talk to customers]!"
A bigger risk is failing because you've built the wrong features. It's harder to change or remove features than it is to build them, and you also don't get the time back for misplaced features. Hence the bias toward starting small.