And how do you know this in advance?

‘Will you pay an additional $x/mo for this feature?’

You can estimate your AWS costs for what the feature would cost for that user.

You can estimate dev and implementation time.

It doesn’t even have to be 100% accurate, but if the numbers come nowhere close to making sense, don’t pursue this further.

Now, there may be a future time where some fundamental cost changes, and it is worth considering again.

Users lie. Not even maliciously.

Clients will say they'd pay for it, then when you deliver it they'll say they realized that your competitor provides that and everything else that is really important to them in the standard fee. The competitor is actually no better, and would require the same amount of development to get to parity in other areas, but they win by claiming that everything will be rainbows.

Clients will say they'd pay for X, but once you deliver it they'll suddenly realize that it won't actually work for them until you also implement Y, but they won't pay any more for X+Y than they would for X alone.

Clients will say they'd pay for X, but once you deliver it they'll have shifted direction or something has changed and they don't need it anymore.

Clients will say they'd pay for X, but once you deliver it and they start using it they'll discover that it's not at all what they actually needed.