Why Your Photos Know It Was a Beach Day
Sinead opens the photo app on her own phone and traces how it labels a beach, a dog and a birthday, with help from a computer vision researcher in Cork who builds these systems.
The Card Decline That Thought It Knew You
0:00 / 0:44
Sample the first 60 seconds free. Sign up free or log in to hear the whole episode.
Show notes
Inside a fraud score
A declined card is one of the most common encounters people have with a machine learning model, and almost nobody is told why. Sinead walks through her own decline, then sits down with Ruairi Devlin, who spent six years building fraud detection for a European payments firm.
- The dozens of signals a fraud score weighs in under a second
- Why a new location plus a new merchant is enough to trip it
- What happens to your flagged transaction after the text message
- How to reduce false alarms before a trip
- The human review queue nobody sees
Next week: the spam filter that reads your email before you do.
Transcript
Welcome back to Autocomplete Lives, I am Sinead Fallon. Two weeks ago I stood at a petrol pump outside Letterkenny with a queue behind me and a card that would not work. Thirty seconds later my bank texted to ask if that was really me. It was. So what did the machine think it saw? Today I am joined by Ruairi Devlin, who spent six years building exactly these fraud models, and he is going to tell me honestly what tipped the balance: a new town, a new merchant, an unusual hour, and a dozen other things I had never considered. We will also cover what you can do before a trip so it happens less, and who actually looks at your transaction once it is flagged. You are listening on Witspeak Podcasts, and this one might save you a queue.
Open as text