To map a particular point (like a house) with latitude and longitude, you have to be accurate to 5 or 6 decimal points. If you have the right data, then yes you absolutely can map this.
But if you're taking, say a telemetry point from a truck moving across the city and trying to map it to an address, there's going to be some drift in that number. And you're going to use that coordinate to say "what's the nearest house to this" – but due to that potential drift, you might match with a different house.
This is not life or death stuff. But then it becomes harder to "cluster" points together to say "all these datapoint correspond to this house". Having a slightly larger grid system allows you to say "anything in this 100sq foot square will be associated with this house" which allows you to have some margin of error, while still being able to address individual addresses.
So then instead of -72.883322, 38.128312 being treated differently from -72.883320, 38.128317, they all cluster around one "square" which makes it easy to just group them.
But What3Words is trying to give each of those "squares" a unique english word identifier, whereas other coordinate systems probably just use alphanumeric codes. I didn't know their algorithm was proprietary. That's bullshit because there is no "algorithm" per se – it's just mapping numbers to word lists. That's it.
But that's marketing.