MD5 Hash of concatenated CC# & street address for all orders. Post to a database 90 days after the order shows no signs of fraud.
Use existence in the DB (weighted by number of instances) as a probability of not-fraud.
My guess is Shopify, Stripe, etc are doing something like that when they send "this might be fraud, you should review" alerts.
Could a Blockchain/DHT solution work for this? Normalize and hash the data and send it to the network to check its karma. If there's no karma then it's a new customer, otherwise the karma tells if it's a good/bad customer. Then after they buy you add or reduce a karma point.
Would there be a way for someone to reverse/bruteforce the hashes to figure out people information?