Listen: AI helps banks speed maritime trade compliance

With all the ships at sea at any given time, banks involved in maritime trade transactions find it especially time consuming to pore over shipping logs and confirm whether vessels carrying goods are on a sanction list. For this episode of “The Buzz,” Bank Automation News spoke with Ami Daniel, chief executive at Windward, a maritime intelligence company that builds artificial intelligence (AI) models to help banks automate due diligence. Windward provides banks with a risk rating and a “go, no-go status,” depending on the sanction status of vessels and the financial institution’s risk appetite. Subscribe to The Buzz Podcast on iTunes , Spotify , or download the episode .

“You're not allowed to do business with people on a blacklist. Having said that, ship’s behaviors are not binary. So, you need to go back and analyze the last 12 months or so of a vessel's behavior,” Daniel told BAN . “That is, maybe, simple if you are a small operation. If you are a big operation, it might take two and a half days to do that. We automate that with AI,” he said. Windward counts a total of 12 financial institutions among its clients, including European banks such as the $1.8 trillion Societe Generale, $660 billion Danske Bank and Frontex, the European Union’s border and coast guard agency. Listen as Daniel also discusses in this podcast the challenges of doing such compliance work manually, the problem of false positives and why he believes AI is here to stay in the compliance arm of financial operations. Bank Automation News will host a webinar on automation technology for better risk management and security on Tuesday, June 15, at 11:30 a.m. ET. Register here. The following is a transcript generated by AI technology that has been lightly edited and may contain errors. Jaspreet Kalra Hello, and welcome to The Buzz. I'm your host Jaspreet Kalra, and in this week's edition we explore how artificial intelligence can help banks and financial institutions automate their due diligence work to ensure that their business dealings do not involve sanctioned vessels, companies or countries. I spoke with Amie Daniel, CEO of windward, a predictive maritime intelligence firm that believes AI platforms are key for banks and financial institutions, looking to make data bank decisions and avoid onerous fines. We spoke about how the data that goes into building these models can be hard to get, how time consuming manual checks can be, and whether AI is likely to expand its footprint in servicing this need in the future. Jaspreet Kalra So before we get started, if you could just give me a brief introduction of what Windward is, what's your focus and the sort of, you know, headline items for AI applications and how they intersect with sanction management and risk management for how it works for financial institutions. Ami Daniels Were maritime AI company, we build one AI decision support platform to help our customers take the right decisions for them on security issues, environmental issues, compliance issues, safety issues, as well as logistics issues insofar as compliance, they think it's very clear, the bar has been raised on last year, since May 2020, the US and UK has have both requested an enhanced level of due diligence for shipping and trading decisions, which means that every bank, every energy company, every shipping company, every insurance company, and so on so forth, would need to screen for vessels behaviors, not just for black lists. Because a different ballgame, black lists are as like their name black and white. So there's a blacklist, you're not allowed to do business with people on a blacklist. Alright, having said that ships behaviors are not binary. So you need to go back and analyze the last 12 months or so of vessel's behaviors. Has this vessel disappeared from the screen? For too long? Has it been a ship to ship transfer for too long? and so on and so forth. And that is maybe simple. If you are a small operation. If you're a big operation, you know, it might take two and a half days to do that. We automate that with AI and give you a go immediately go no go recommendation. Jaspreet Kalra So just to sort of, you know, dive into how you build your models, and what sort of data that goes into it. Do you use like public source open intelligence that talks about ship behaviors? And how do you sort of expand on that? Because your recommendations also carry with them sort of like a risk appetite and carry with them? possibility that if that recommendation is wrong, banks could be liable for a fee. So how do you navigate that sort of liability and what sort of data goes into building Ami Daniels that model is primarily commercial and public information is permanently commercial. So source globally, ships transmissions are satellites, picking up the vessels, locations, optical images at a global scale, on ownership database that we've launched in FEHB. Basically, a mature a lot of companies built that since 1760. So our ownership database is based on a UI and a dynamic approach to the data, weather information, Port information, and so on. These Who are these go through a rigorous filter and fusion mechanism. There's a lot of problems with data as well as as intentional spoofing of the information once that once that gets fused together, then we have a set of machine learning and deep learning algorithms to calculate the vessels will be called activities. For instance, every time a vessel as a shifter should transfer we use a deep learning algorithm to model for the vessels behaviors. Specifically, if you just measure proximity, you get whatever 80 90% false positives, right? You don't want that thing at scale, you want to take the false positives to a low as possible level. The same goes for the dark activity if an investor disappears. The question is did it really disappear or not? You want to every third vessel disappears for more than eight hours in the Gulf, in the Arabian Gulf. So you also don't want to flag every third vessel, right? Probably you want to zero that down to the minimum possible quantity. So that's how we use AI on multiple levers levels. The top line risk score is actually rule based. And the reason is that advisor regulated, so the feedback we've received is if you just give a deep learning score, it will not want to see it fly in all regimes. So it will you For instance, it probably will not work. The banks there are obliged to have things that are explainable and auditable, and trackable, if that makes sense. So that's an approach we call a glass box, and not a black box. Last point you mentioned is, is risk appetite. So we've built and launched a few months ago, our risk customization capability. So every customer actually, every subsidiary can customize Click, click, click their risk appetite, insofar as regimes look back, and all kinds of parameters. And we can simulate what does that mean for their business. But basically, it means no two people would get the same answer in the world. They think that's a very big difference in some of the data vendors in the space that offer you data, but not just data for everybody the same data, Jaspreet Kalra and just sort of walk back a little help. Because this is a transition from very manual very sort of paperwork based processes to more software based processes. Could you give me an inkling of the time difference that your product is able to bring to the table for any back? Ami Daniels Sure. So we provide a any go noble recommendation instant, instantaneously, Quick, get it also very be ipi. Even if you have 10 million API requests a month, while you're still getting a second. A person might take two and a half days to do complicated cases and two hours to do simple cases. So I guess you can multiply that by the answer is, it's much more effective. But I think even more than that, specifically, with f5, what we're seeing is there's a lack of competency and a lack of talent for the second and third lines of defense. Jaspreet Kalra Okay, and I don't mean by that, Ami Daniels sure, of course. So yeah, the first line of defense which says go No, go, escalate, don't escalate, approved on a proof. And this is just the basic structure of how banks work. And it's not just for shipping, trading, it's true for everything. So these guys would look like millions of transactions. And then they if something has a red flag, or an amber flag, they would need to escalate it to a second level of defense, second line of defense or third line of defense. The second line of defense, we'll take a look at these and actually screen some of them and say you have, you know, these 10 are false positives. But these 10 are extra suspicious. The third line of defense does complex investigations. I think what we're seeing in the one of the unintended consequences of this regulation, enhanced level of due diligence is simply a crunch, a talent crunch, the banks don't really have enough shipping and trading experts to hire to do the investigations. So what means may be more natural to a commodity trader, because they're used to looking at chips all day isn't necessarily as natural to a bank. So I think Well, probably trying to provide is not just a go, No Go recommendation. But the expertise baked into the product for non experts. And therefore that helps to deal with that second line of defense a crunch. Jaspreet Kalra Right, okay. And something that very often comes up in AI related discussions, as well as that building that base data set that you sort of expand on and do your recommendations on can be a tricky process. Now, Has that ever been your experience that sometimes the case went in and the machine, the machine, or the machine learning model, just said there was insufficient data or insufficient pre existing information? Now, how do you deal with those cases? Ami Daniels Yeah, of course, specifically, within the shipping space, it's, I think, even more complicated because you, we predict some of the cases which have a very low, low low level of amount of labeled datasets. So so if you want to frame it as giving a noncompliance example, we we predict the probability of a vessel to have a collision, there are about in about 200 collisions a year in the world. So how do you predict to build a model without having overfitting? Because, you know, you play with it too much. With 200 labels, I think that's that's quite a challenge, for example, and by definition, the more you break it down into what you're you're predicting, the more accurate you are. So I think that's, that's a challenge. What we do is we use our own domain analysts to tag a lot of cases and use that as a label data set to train the models. So you can actually know how good you're improving or not. That's one answer. The second answer, I think, by now we have enough user base to provide us a lot of feedback. So we have a lot of ship owners reaching out to say, hey, my vessel is high risk. Could you help me understand why and obviously we can. We have very rigorous SLA on this and I think it's very important to action. Having said that, not necessarily, not necessarily all requests are clear. Because in some cases, you know, what we see, as some people say, Hey, you know, you're saying that my ship went off screen, but actually it didn't, or I had a malfunction. Which happens, we asked for supportive data where where you can get yourself a phone location, and clear up that. And other instances, hey, while I'm already flagged because, you know, I sold this vessel yesterday and still not updated in your database. So you know, that can happen. So we just say quick, great, could you please provide us documentation and we'll update it clear you. But there are other cases in which people principally do not agree with a concept of having a platform that says your high risk or not. So it certainly don't agree with the facts. They agree with effects. They do not agree with a concept of being rated for, for risk, which, which I think by now is, you know, in the US, basically, everybody's been getting rated for credit risk, when you get a loan, you the price the loan on your past performance. So does that make sense? Jaspreet Kalra Yeah, that makes sense. And sort of just adding on a question to that is when you are doing that risk analysis, when you're looking at some suspicious elements that you might be thinking of this might be fishy, this might not be fishy. How often? Is it that you also see state actors involved in that? And how do you deal when a state actor sends your request to flag and flag a certain vessel that might be also dealing with multiple financial institutions? Ami Daniels Yeah, first of all, obviously, silly activists have that as well. I think the recent example is, so the US the us remove the sanctions from the company building nordstream to their CEO, but it added 13 Russian vessels who are dealing with that project. Okay. Okay into one of the sub lists of OFAC and it was as recent as a few days ago. So, obviously, there is absolutely connection to state actors. Now, the example cineq there is very specific restrictions on ship the shoe transfer with transfers within the South China Sea, I believe, with scenic, that's there's another example, which is obviously Chinese National Oil Company pivetta. There is sanctions and divison. Venezuela, so you're actually allowed to come in trade into Venezuela, you're not allowed to trade with the devesa. So, so absolutely state actors, and obviously, they're more more examples, right. The Emerald, which did an oil spill, probably, of Israel, was that Iranian flag, but it was raining operated in the Panamanian flag. We actually haven't gotten into a situation until now that the state actually reached out to us and say, could you please change this? We did have flags reached out to us with our big ship or is it small ship? Or is a big charters? reach out to us, but you know, I look forward to the day that this happens. For us, I think we welcome and most will soon have an online. We have something similar, we're gonna expand online the mechanism allowing ship owners to reach out to us and provide us their information for screening. Because for us, it serves, you know, it's a win win situation. The banks and the energy care companies and insurance companies get better accuracy and less false positives. And the ship owners do to know the business because you're not wrongly flagged, which could happen, right? Somebody told me a guy in shipping. very senior guy told me, you might be 99%. Right? But that means you're 100% wrong to 1% of the people. Jaspreet Kalra Right? Yeah. Okay. So if you could just talk to me a little bit about the scope of the data you have, because you mentioned you have a lot of clients, you also monitor a lot of vessels, if you could put a number on that for me. And also, if you could put a number on how many financial institutions slash insurance companies you're currently working with? Ami Daniels Yeah, so we're working with a couple of dozen Finnish institutions. Not all of them have been published, both on for for insurance, marine insurance, as well as ship finance, correspondent banking and trade finance. We have all these types of customers, the different the screen for different transactions. So marine insurance companies primarily screen a pre pre renewal. It depends if it's p&i, or how machinery p&i is on February 20 18, or 18. As well as on an ongoing basis, ship finance banks use this for quarterly basis and a quarter review of their loans, trade finance banks, obviously screening against every trade transaction of correspondent banks when they used to use this for correspondent banking. That's a different use cases in terms of data that we cover We're all IMO vessels. So which means about 100,000 vessels globally. We're also covering cargo information. So we're also screening for oil information. So if you're taking suspicious cargoes, which is now being, it's mandatory to check that. So I believe we have the only solution in the world right now doing that. So others talk about owning the vessel, we also help you screen the cargo. And we're going to expand that really soon. And whatever, 35 days again, keep your ears open for that, I think it will be very exciting, because basically, I think what we're seeing is that oftentimes, the banks don't necessarily have all the information, they would have just bills of lading, or the name of a ship or name of a company. So just the entity matching part is quite challenging. Jaspreet Kalra So if you ever give me like a five year down the line sort of prediction or sort of, even though estimation, how deeply integrated Do you think AI based systems will be in this space? Ami Daniels The deepest it's possible, nobody will stay behind. I think I think COVID really proved to everybody, you cannot not go digital. So and at scale, you can't go digital without AI. So

Listen: AI helps banks speed maritime trade compliance

CEO of maritime intelligence company Windward discusses due diligence automation

Movers and shakers: Tech leaders get boost at JP Morgan, Bank of America

Banks automating risk see benefits, including reduced costs

Related Posts

Bank of America ordered to pay $1.2B penalty for AML violations

Nine large banks engaged in so-called debanking, OCC says

Who’s to blame if AI fails at work? Wharton study shows a split

Banks automating risk see benefits, including reduced costs

EMERGING FINTECH DIRECTORY

FinAi Podcast

SPONSORED

The Next Generation of Cyberthreats: Powered by AI

Build an Antifragile Strategy to Outperform the Market

How AI and Product Experts Turn Fuzzy Requirements Into Focused Dev-ready Roadmaps

Connect

Welcome Back!

Retrieve your password

Continue Reading with FinAi News Premium - Less than $2/Day

Listen: AI helps banks speed maritime trade compliance

CEO of maritime intelligence company Windward discusses due diligence automation

Movers and shakers: Tech leaders get boost at JP Morgan, Bank of America

Banks automating risk see benefits, including reduced costs

Related Posts

FinAi Podcast

Connect

Welcome Back!

Retrieve your password

Unlock This Article

Continue Reading with FinAi News Premium - Less than $2/Day