I have no idea idea either, but if I were running an outfit similar to Amex, here's what I'd probably consider:

1. Quality of charges - Discretionary income on luxury items are more valuable than business charges on inventory that falls under cost of goods sold. Leverage to charge higher merchant fees are driven mostly by the promise of delivering high quality big spenders. There's a world of difference between charging, say, $500K in post tax luxury travel related goods than $5 Million worth of vendor inventory in a low gross margin business.

2. Zip code - Especially in today's big data actionable world, it's not difficult to get an idea of a person's net worth, and where one lives is a good start. For instance, anyone who lives in a zip with high home prices with no or low mortgage will excellent collateral, and will likely have a higher real avg income than what's often on the current year's AGI line the tax form.

3. Profession - Certain professions exclude those with bad credit, history of having filing bankruptcy and exhaustive background checks. Those in these professions in combination of the above bear very little reputational risks--they generally always pay their bills and never run afoul of laws, especially financial ones.

4. Prior history - Amex with their closed loop system will mean they have the most granular data at their disposal and combined with a cardmember's history should allow them to, more or less, devise the secret sauce on how best to maximize revenue and enhance the "mythical appeal" that has always been associated with this card.

5. Assets known to them - Some individuals and businesses, for whatever reason, require limits higher than Amex would be comfortable, and have submitted documentation buttressing the case for higher charge capacity. In this cases, Amex has intimate understanding of assets, and may thus make decisions accordingly.

I'm sure there's much more, but those seem pretty common sense and easy enough to implement in a model.