Charlie Delingpole, founder and CEO of regtech startup ComplyAdvantage, said big tech firms like Google, Apple, Facebook and Amazon face no shortage of regulatory headaches by venturing into financial services. Facebook, with its Libra cryptocurrency, is in for an exceptionally bumpy ride, he predicted.

ComplyAdvantage, a New York City-based firm that uses machine learning to provide banks and other organizations with real-time anti-money laundering and compliance screening, is growing at a rapid clip. Coming off a $30 million Series B funding round led by Index Ventures in January, the company has swelled to 210 employees and 350 clients in 45 countries. Revenues have climbed from $3 million to $15 million over the past 18 months.

In an interview with Bank Innovation, Delingpole discussed the regulatory roadblocks big tech firms face in expanding their reach into financial services and why banks are resistant to automation, as well as machine learning’s place in regtech. An edited version of that conversation follows.

From a compliance perspective, what do you make of Facebook’s Libra?
If there’s one definite, great use case for crypto, it’s laundering money. Libra will do for money laundering what Facebook did for fake news. If they’re incapable of understanding who’s on the platform, how are they going to understand who’s transacting? Can a social media firm really do finance?

Of the $10 million that all the partners promised in escrow, nothing has actually been wired, nothing is actually live and the regulatory backlash is already huge. However, given the scale, resources and reach of the business, I wouldn’t necessarily bet against them. In this specific incarnation, it looks like Facebook is thinking out loud. They’ve hired some great people, but I think what they’ve done is hugely ambitious and has many barriers. They have to systematically eliminate each barrier in order to succeed.

What are the regulatory implications for big tech players drifting further into financial services?
Generally, the big techs are not perceived in a very welcoming way by financial regulators. There’s already a line item on Facebook’s statements every year for $2 billion or $3 billion in fines for shopping violations. If you think that’s bad, imagine how much they’ll be fined for laundering money or financing terrorism.

The big techs all recognize that they have amazing resources, scope and distribution ability, and they see the banks as weak and fragmented. That said, it’ll be interesting to see how they cope with regulation. Do they actually want a million compliance things to do? I’m thinking probably not.

Do you face resistance from banks over automation?
I think that, institutionally, you’re going to see that people who have big empires will fight to protect that empire. It’s like the Meiji Restoration and the destruction of the samurai. Whichever class of person that’s being wiped out obviously is going to fight against that.

If people judge themselves by the size of their headcount, they will not want the tool to come in that’s going to reduce that headcount. However, if compliance is 10% of the cost base in most banks, imagine if that was removed. Every bank in the world would be more profitable.

Overall, what kind of reception do you typically get from banks?
We are making good progress, and we think that our clients like the solution. The whole theory behind what we’re doing is, if every day we can make a 1% improvement, that’s going to have an exponential impact on quality [over time]. At some point, there’s an inflection point. When we were in my garage five years ago, the data was garbage, but you constantly refine it, improve it and, eventually, you have the best product.

How do you use machine learning to get a better understanding of risk?
We’ll take structured and unstructured data, then do the extraction, classification and resolution of entities, automatically, in the pipeline. You also can use training data to inform those choices. For instance, with classification, the more examples you give the platform of what a terrorist looks like, the more accurate it is in terms of classifying those entities. You want to catch more bad people, but you also want more of the extractions to be correct, meaning no false positives but not missing any entities either.