The FBI used phone metadata and face recognition to find who assaulted the U.S. Capitol. The identification of suspects based on mobile phones’ metadata and face recognition is very effective, but can lead to false positives.
I worked on mobile phone localisation, and it is almost like in the movies. It is less precise than GPS, but the error can be as low as few meters in areas with good coverage. However, on large numbers, errors happens. Phones may connect to distant antennas or register a signal with strength lower than the expected, and that confuses the algorithm.
To explain it in simple terms, imagine a one dimension world with just one antenna: it would be possible to estimate the position of a phone depending on the strength of the received signal. However, obstacles could make a close phone register a low signal strength and confuse the estimation.
In the real world, there are multiple antennas, and it is possible to triangulate the position more effectively, but the method is not perfect. On top of that, the American police used face recognition. It is an effective technique, but notoriously gives higher false positives with anyone who is not a white male.
The two strategies combined compensate their vulnerabilities. It is improbable that localisation techniques place you on a crime scene where a camera filmed someone that closely resembles you. However, it may happen, meaning that it will happen on large numbers.