Interface conventions exist because enough products used the same pattern that people learned it. Every departure asks the user to spend attention learning something before they can do what they came for.
Sometimes that is worth it. Most of the time it is a cost paid for the designer's satisfaction rather than the user's benefit.
Familiarity is transferred learning
A user arrives already knowing how a checkbox, a search field, and a navigation bar behave, having learned it elsewhere. Matching those expectations means your product is partly learned before it is opened.
Reinventing them discards that entirely, and the user has no reason to believe your version is better until they have already invested effort finding out.
Internal consistency matters just as much
Three date pickers, two ways to save, and buttons that sit left in one screen and right in the next all impose the same cost repeatedly, within a single product.
This is usually not a decision but an accumulation — different people, different months, no shared component. A design system is mostly a mechanism for preventing it.
Spend novelty where it is the product
If your differentiator is a genuinely new interaction — a novel editor, a new way of visualising something — invest there and be conventional around it.
A creative login form has no upside. A creative core interaction might be the entire reason someone chooses you.
Test the departure specifically
Where you do break a convention, watch real users meet it cold. Designers and internal staff have learned it and can no longer see the difficulty.
If people hesitate, the pattern is costing more than it returns, however good it looks in a portfolio.
The best interface is frequently the one nobody remarks on, because nothing about it required attention. That is a difficult thing to present as achievement and it is usually the achievement.
