ShowHide categories

AI and payment systems: how they make sense

Unlocking value

It may be a buzzword to some, but in the world of payments AI could have an important part to play

It’s not a marriage made in heaven: not one of those instant matches that makes you wonder how you got by without the cross-product of the two technologies in the olden days. That kind of strange conceptual marriage is, however, where businesses are increasingly looking for extra value: in this case, tools that help an administrator work with their payments data.

Everyone will probably be thinking about fraud prevention at this point. The burgeoning wave of AI lead applications are coming from the security sector, the network monitoring sector, and other hot fields of computing where unattended, dark systems churn away without benefit of error messages or user-friendly scanners. Two objections arise to this bit of early adopter pigeonholing: one is that there is definitely a trend to make tenuous links to AI to sell something people don’t need. The other is that AI based services don’t need to take over the world in order to actually be helpful.

Earlier on we said, with some care, that the area where advantages can be gained is in working with payments data – not payments software. Neither developers nor users, nor yet entrepreneurs, want to see an artificially intelligent program spitting out payments according to rules about debtor priority, which it has cooked up itself, sight unseen. There are those who believe they can spot the digital footprints of automated share trading programs in the stock markets around the last few crashes and recessions. This puts managers and administrators understandably on guard when it comes to selecting and using an AI tool.

Spotting abnormal patterns

However, there is a perspective from which it makes a lot of sense to call in a smart machine, and that is when your payments data is about to be submitted. Most payment systems still like to work in a pseudo-batch manner, if only to create audit trails that make human-readable reports. The batches may be short and frequent by contrast with the days when BACS would accept lists of payees and transactions on a floppy disk. The point here is that it takes a particularly eagle-eyed admin to spot transactions that are out of the normal pattern.

This is AI’s bread-and-butter work. It is actually better for this kind of search robot to be given unstructured information, with as few presumptions inherent to the information as possible. That way, the AI can bring up the odd patterns it spots, without the data architect’s priorities acting to hide the more obscure (and more interesting) spikes or repetitions or typos.

This is why really large corporates make data lakes - repositories of information with as little imposed structure as possible, so that smart software can tease out trends that are unexpected by the database designer or the project sponsor. Are the programs that tease out these hidden values strictly AI or is it just very large scale statistical analysis?

Old school artificial intelligence specialists will line up with new-school game developers and give that a resounding answer, that this isn’t AI. But from the perspective of someone logging in to a payment management portal, being shown that there’s been a 200% increase in spending on five accounts all in the same rough area, month-on-month, is exactly the kind of helpful analytical presentation that they need. This may still be slightly in the future when it comes to the end-users of payment systems, but it’s certainly in the pipeline, because the benefits of forewarning or an unbiased look for unusual patterns are undeniable.

If you're interested in reading more about the technologies that will be disrupting banking in 2017 click here.