This seems to have the same issue as Pirsch[0] in that (i) a mobile user will have multiple IPs during a day as they roam between wifi networks and mobile networks - which will incorrectly over-report unique visitors and (ii) NAT gateways with similar devices behind (a home with 4 iPhones, a workplace with 5000 identically imaged desktops, all similar devices at a starbucks, an entire lab of computers at a school) will incorrectly under-report unique users.
An IPv4 address is an unreliable way to count "a visitor". At best it represents "at least one network (~building) of 1-n visitors". You could argue 0-n visitors (some sort of automation/crawler). It could also represent more than one network (a temporary IP obtained for less than a day, a IP obtained for a day that changes during your definition of a day).
User-agent reduces the scope, but doesn't eliminate (between recommended evergreen updates and identically imaged/uncustomized use of devices).