I applied online through Apple's website and received a callback from a recruiter the same week. We set up a phone meeting with the manager for the same week from there.

The phone interview with the manager lasted on the order of 30 to 60 minutes, where we discussed my background and research, as well as what the Apple Data Group does. We then scheduled a one-day interview with approximately six one-hour interviews with people from the group.

The Apple Data Group appears to be about 8-10 people:

• Team manager
• A sub-manager who directs a more hands-on QA group
• A database specialist they call their statistician
• A half-dozen or so recent PhDs from a CS group in AI from UT Austin (most of the people came from this same group and basically know each other).

The one-hour interviews were all very friendly and non-technical. I was asked maybe two to three fairly "softball" questions. Almost all of them seemed to be questions about how I would handle some problem they've encountered at Apple.

Basically, the group is in two sub-teams (plus the separate QA team).

The first group consists of the database specialist and appears to be focused on QA'ing test results from Apple products made in China. I've done a pretty fair amount of electronics-type QA and I got the impression (taking his comments at face value) that he was kind of winging it and didn't really get his hands dirtier than, say, running an ANOVA or PCA on test results.

The second group works on fraud detection. The bulk of the work is aimed at stopping people who steal credit card information and then use it to buy Apple credits, and then sell those Apple credits. Once again, taking what they said at face value, they essentially maintain a program running Naive Bayes (trained on past found credit violations) to weed out potential new threats. It basically outputs something along the lines of an expected risk, and the riskier transactions are sent to a human-based phone screening. So, definitely nothing super mathematical or sophisticated on the algorithm end. I asked one of the newer team members how he spends his day, and he basically said looking at data, so I'm sure they are doing some other type of screening also.

All in all, the team, especially the fraud side, seemed like a really likeable group. I was a little surprised when I was passed over, especially since the interviews were so easy. But I was given a strong hint when I was informed (I assumed I was lied to) that the manager was out sick that day and couldn't meet with me.

All in all, a good experience, definitely less mathematical/technical than I expected, both in what the group does and in the interviews themselves.

We perform pre-screening on the data to remove fraud threats. How do we find a data sample that can accurately represent fraud events?

Essentially, we use Naive Bayes (although the question wasn't explicitly framed around this method). Sometimes, probabilities do not work out when there is no data in the area we are examining.