PegaWorld iNspire 2024: Managing Sanctions Alerts with AI at Scale

Welcome everybody. I hope everybody had their coffee here at the last session of PegaWorld 2024. And you're here talking to a money center bank about sanctions. What could be more compelling? What could be more exciting? Right. My name is Brian Gashi. I'm a managing director and the global head of sanctions and client screening technology at JP Morgan Chase. I've been with the bank for about 18 years, and before that I spent some time at HSBC and Royal Bank of Canada.

Um, I've been a career technologist in the payments space and in banking, and have spent the last 4 or 5 years of my illustrious career focused on sanctions screening and client list screening. Um, I'm joined today by my product partner, Kate Stern Jones Kate. So I'm not going to date myself by telling you how long I've been at J.P. Morgan because I'm a lifer, so it's longer than Brian. Let's just leave it at that. I'm an operations background person. I'm located in Tampa. My most recent circle circuit has been in product management, but we're really focused on what I call franchise screening. So solutions for fraud detection and prevention at J.P.

Morgan, as well as the screening associated to client onboarding. So politically exposed persons, negative media, terminated relationships and sanctions for client relationships as well as sanctions screening transactions in flight. So if you think about the compliance program there's really three there's onboarding what your clients are doing at the time. And then after the fact modeling associated to AML. So we are the first two of those three things. So we're excited to have you here today. We do not have any fancy cartoons. I was at Rabobank presentation today really impressed by their cartoons. We didn't bring any cartoons.

We do not sing, nor do we dance. So you won't see any musical numbers? No. Um. And then I saw some guys in the ready room who were wearing hard hats with, um, with miners. Um, I don't know what that was all about, but I feel like. I feel like they're having an exciting presentation today. We're going to talk to you about sanctions screening and sanctions screening at scale. When we do things at JPMorgan Chase, we tend not to do them small.

And our sanctions program is no exception. It is, um, something that we have had to deal with some tremendous challenges with scalability and volume. And today we're going to talk to you about kind of the business context of our problem, what it is we had to solve. And we'll talk to you about an application we built called Global sanctions manager, which we built in about 2014, and we are now looking at modernizing and taking to the next iteration of what we want to do with managing our sanctions environment. We're going to talk about why we chose Pega, our feature solution and integration, our modernization journey and what we think is next and what we see out of out of today's session that we find compelling and and useful for our future. And we'll take some questions. We'll let I think about 15 minutes of questions we'll take at the end. So in terms of our business context, what is it that we do? Basically we have the client screening component.

We want to take care of our AML and KYC obligations with regard to our clients. We want to take care of our transaction screening obligations. What is it we do to manage the transactions? Make sure that we're meeting the obligations of our sanctions program and our AML, KYC solutions. Um, we need to deploy very controlled and very controllable um, applications and models, and our solutions must be scalable for the millions and millions of events that we manage on a daily basis. Um, our filters are the first stop for any of our clients, for our transactions. Our filters run up against lists, uh, external regulatory lists, internal lists indicating our risk tolerance for various partners. And, um, pass it on. Pass on the the hits to our our sanctions manager application.

Kate. Anything more on our business context? Um, I think I would just add, you know, we've had a strong partnership with, um, Pega in terms of case management and workflow. And we started down that journey. You said 2014. I think it was. 13. When we were doing a firm, wide event to rationalize all of the places where sanctions screening and client list screening was being done so that we were centralizing it into a firm wide utility using prescribed filters. So when the client list screening side, we used an application called Bridger, which we're actually in the process of decommissioning, which should happen late this summer.

And then on the transaction side, we use an application called Focus Soft. But what we wanted to do was at least give the user experience associated to alert, review and disposition some consistency. The control that that Brian referenced and impose um. Requirements associated to when you needed to have two people look at something. When you had one person look at something when you wanted to know if something was a high risk alert, you routed it directly to, um, a more senior operator so that you eliminate errors. So that was really one of the reasons why Prpc and Global Sanctions Manager, as we call it, was incorporated into our end to end workflow. So let's talk a little bit about the scale that's required at JPMorgan Chase. Um, in the sanctions environment over the past several years and really the past decade, we faced exponential growth in, um, the transactions that we need to screen on a daily basis. Um, we're close to 10 million a day now in terms of what we're screening.

And, um, over that period of time, um, we've also seen a tremendous increase in the use of sanctions programs globally. Take a look at what's happening with the Russian sanctions program, the Iranian weapons sanctions program. Our governments are using sanctions as a tool more and more now than they ever have before. Um, so this double edged sword of increased volume with increased number of sanctions entities, has resulted in a tremendous increase in the net we cast and the initial set of fish that come back in that net as we pull all of our transactions through our filters. So the graph here shows that kind of life cycle that we've had to manage in terms of scalability and how we have utilized Global Sanctions Manager, the Pega workflow tool, and all of our internal automation capabilities to help manage this increased volume and increased, um, hit profile over the years. That initial that initial net is, do we have a hit? Do we have a match against a name, a location, a vessel, an entity, a good service? Um, and that's going to pull back a pretty large proportion of, uh, hits initially on the transactions that we send. The first thing we want to do is say, can we clear this based on something we've seen before?

Does this look like something that we've passed, um, maybe through a human decision in the past? If we can determine that, we can filter that out as a false positive. Get rid of that, reduce that noise. The second step in the automation that we've implemented is the use of artificial intelligence and machine learning in a supervised decision tree that is extremely, um, audited and, uh, tested through a quality assurance program. Um, this will determine if certain categories of transactions are suitable for pass or for perhaps a lower level of human due diligence to pass that particular transaction. We found that applying these two concepts to our transactional volumes can get us into a very high percentage of automated clearing versus what is presented by that initial net that is cast. And we're presenting very few subsequent alerts to our actual user community, to to action in the Global Sanctions Manager tool. Our final step in the automation journey is to present them with as much information in context with regard to the transaction they're looking at and the relative hit that they are matched against. For them to make an intelligent and informed and consistent decision every step of the way as they go through their user journey of escalation.

Things that we are integrating and growing are our use of integration to external data stores and risk assessment utilities to assess the relative risk of what it is folks are looking at. And we would like, as you'll see in our future journey story to integrate more information and more contextual instruction into a user's journey in the Global Sanctions Manager application. Anything to add on this slide? I think one of the places that Pega has been really useful for anyone who's been on the machine learning or AI journey, cataloging and tagging of data is really valuable labeling. Um, and the way the workflow enforces that every decision has a decision and a rationale associated to it has really helped us in terms of our machine learning journey, because it then helps us train the model more effectively. So Pega has been very useful in terms of forcing the user into a very structured output. Yeah. So why global Sanctions Manager, you've heard a little bit about it. Um, in 2013, 2014.

Um, anybody from the keynote yesterday remember some of those system context diagrams that were blurred out? I swear they pulled them him from JP Morgan Chase. Right. We have a Constellation, a plethora of systems that, uh, historically have been engaged in what it is we do in 2013, 2014, we really set out on a mission to make a much more consistent user experience and a much more consistent controls environment for interacting with our outsourced from our sanctions screening and client screening portfolio. And we built global sanctions manager globally consistent processes integration. So if you think of that Center-out diagram that um, that they were sharing at the keynote yesterday, that is really what Global Sanctions Manager represents to us, is it is our center for our operations users. It's the hub with which we connect a lot of disparate line of business systems, a lot of disparate business processes, including Other, Pega, prpc processes around the bank, of which there are many, um, a very robust controls environment, consistent and scalable interfaces, and a very good integration touchpoint for our automation and our AIML implementations. So why Pega? Well, first and foremost, um, Pega has a very long story with JP Morgan Chase, and we have a very robust service and investigations platform that is implemented on the Pega RPC platform and has been there and was there in 2013 and 2014 when we were considering options for where to centralize our sanctions hit management process.

Peggy's reputation Pega, uh, ease of use and time to market is a very compelling argument. We also had a pretty good developer base, right? So we had a good group of folks upon which to build Pega Platform, and we actually carved out some folks from our service and investigations had them help build our global sanctions Manager application. Obviously, the capabilities of Pega, when you're talking about workflow management, when you're talking about being that Center-out hub, integrating multiple facets of your line of business ecosystem, Pega is a pretty obvious architectural choice along with some others. But we found it to be very compelling in terms of that, um, hub to establish a consistent business process management workflow and controls environment and fit our strategic direction. It was scalable. It was fitting the need and comfortable for our user base as well as our developers, and we felt it had a good and promising future. And I think that's been proven out as here we are in 2024, ten years later, and we're talking about global sanctions manager and what's next for it. So obviously we think we made a proper choice and a good choice there.

So what have we accomplished. What what did the Pega JPMorgan Chase partnership. Deliver? Um, also at the keynote, we had somebody who was it who was saying, uh GenAI Gen AI. Gen AI GenAI. That's the that's the thing. Well, I want to say for us, it's data, data, data data because generative AI is nothing without data. The data set that we have built since 2014 of curated human decisions with consistent reasons for why those decisions were made, has allowed us to implement that decision tree. That second step of automation with over 2 million human decisions, training that decision tree and allowing allowing us to automate a lot of what a human being does over and over and over to determine this is a valid name match.

This is a name without proper location. This is not a vessel, this is a person. This is an entity, not a person. These types of reasoned decisions inform our data set in a way that allows us to really supercharge the automation after so many years of building this data set. So I'd say that is probably one of the greatest accomplishments and things that we have been able to pull in. I think we've also been able to build a very. Robust and compelling controls argument for our regulators and our auditors. When they look at our sanctions program, they see global, consistent, controlled and measured risk taking and processing. And I think it's been a very rewarding experience to Have the Pega Prpc solution for GSM.

Anything to add on the outcomes? Got it. So what are we doing now ten years later? Well, I was at Rabobank's presentation this morning and similar to them, we're all on a cloud journey, right? And right now we're on a what eight for one on BSI. Yeah. Um, and we're going to a Pega Infinity 23. And we're doing it on a J.P. Morgan hybrid cloud implementation.

So the J.P. Morgan hybrid cloud, we do some stuff with AWS on public cloud in a curated environment there. We also have an internal cloud, more, um, more of a Cloud Foundry implementation that has been customized and wrapped within the Jpmc controls architecture. So we are we're delivering to that containerized, Kubernetes based Cloud Foundry internal cloud. Not without challenge. Um, you know, the Pega Infinity 23. Certainly the Pega Cloud is a compelling spot to be didn't quite fit our risk appetite and controls environment. So we went with our internal cloud. But we've been partnering with Pega on overcoming some of those challenges with implementing to our hybrid cloud solutions, um, solution, especially around security and controls, making sure that it's safe and fit for purpose for what we're doing.

Obviously, the benefits of cloud have been well expanded in terms of scalability, maintainability, uh, resiliency, recovery, flexibility. Um, and we hope to make this then the platform for the future and how we can get to that next step of where we want to be, which is really the the reason we're all here today. And we saw some really compelling stuff in the innovation Lab that we would like to see in the vision that we have for global sanctions manager in the next 1 to 3 years. Um, GenAI is very compelling, and we see a really good use case that Coach process. And what was the other one? Buddy, buddy, Coach and buddy working together, we think is a super compelling vision for how our users are going to be working with Global Sanctions Manager in the future. Today, it's very static, right? They see the transaction, they see the list, they see the structured data. They see some added data, but they don't get a lot of insight into what's the right thing to do.

What is the likelihood based on what we've seen before, that this is really a true hit? Or is this more likely a false positive? Because we've seen this go to investigation for this same person three times before, and it always came back with this birth date and this full name. And it wasn't the sanctioned bad guy. I think that the partnership that we see with Pega delivering the next generation and iteration of global sanctions manager is really going to take us to a world where our operational users are going to have a very different, compelling and refreshing experience in working what we hope is fewer manual cases and working them much more efficiently and with much greater insight and expert level information at their fingertips. Anything to add on our vision and where you want to be from a product perspective? Okay. So when we think about the business side of it, we've gone through an organizational shift over the course of the last two years. And we're really hoping that our journey with Prpc or with Pega more broadly, is going to help bring those two different organizations together, normalize a lot of what they do, what they do, and make it more consistent and also, frankly, enhance the user experience, right.

So today, if you're a client list screening person, that's really all you do. You look at pep, negative media and terminated relationships, some sanctions alerts. And if you're on the transaction side, all you're doing is churning through the work. On the transaction side, I think it should make life and the and the user journey more, more interesting. and hopefully not that we have a user retention problem because we don't. But you know, certainly allow for more, um, career progression in terms of the user base, which is something that we'd also like to focus on. So to expand a little bit on the kind of the technical future and a few more of the compelling features of global Sanctions Manager that we want to expand upon and explore more. Um, there's a few key integrations that we've done. One is with a symphony chatbot, and it allows a user to enter a reference to a transaction that's being screened and gives them some feedback on where that transaction is at.

You know what? What's going on? Is it in level one review with operations? Is it in investigation with compliance? Where when can I expect this transaction to clear the chatbot experience today is extremely static and extremely limited in terms of how we can implement it. When you plug that into the generative AI capabilities, and you start to leverage other data sets within the bank to give context to what the user is asking, it can turn into a much more conversational, much more useful tool for our users to understand what's going on, not just in the sanctions ecosystem, but through plugging different data sets together. We can expand that view to give them context of the entire life cycle of a payment or transaction that may be in our bank. So we see this as something that we can use as a lever to start to build more consistent and common communication across the bank with regard to the context of the transaction that a client or an operations or a sales associate may be interested in. Another compelling integration touchpoint that we want to expand upon is something we built around the Pega framework called Smart work.

Smart work. We layer on top of the Pega framework. It's a Java application that helps us interpret the skills, the abilities and the background of an operator and customize how casework is assigned to them, what they get, at what level, in what language, with what SLA. Now, since we built it, Pega has expanded a lot of the predictive analytics capabilities within the Pega RBC tool itself and Infinity. And what we'd like to do is start to plug those two things together and start to leverage a little bit more of what Pega has to offer in terms of what's happening in our case, population, The telemetry and metrics around the case and what's going on with it and the user community and what they're doing with the cases. And then we want to plug that into some of the smart work attribution that we've built out and enrich and enhance that turn it into a true product capability that will help us more. It would be nice to make it more responsive to. So if someone's inquired in the chatbot over a specific transaction or a client list screening record as an example, it would be nice that the system would understand that someone's interested in this. Therefore, it has to be important, give it more priority, and bump it up to the next eligible operator who would be able to potentially action that ahead of other things that were already in the stack.

Right. So more responsive in terms of the client outcome as well. Another compelling thing we'd like to plug in, this is where the buddy is really compelling is give some feedback Back to the user. Give some feedback to the person working the case with regard to, you know, throw them a practice case, throw them a scenario and give them some feedback about what did you get right, about your decision about the questions you asked? What did you miss about your decision? A little feedback mechanism. We may even turn it into some gamification to help. Maybe make it a little more fun. You know, how many cases have you done in the last hour?

How many did you do last week? Who's got the record? Chase them, but do it right. How many errors? Right. So we think there's a lot of potential and a lot of things that we have not even considered about what's possible now with the Pega Platform. So anything before we close, Kate. All right. Well, we've left almost 19 minutes for questions, so we will be able to entertain some before we go to questions.

I do want to call out some of the contributors, two of whom are in the room here, and if you want to know more about the Jpmc ecosystem and what it is we do with global Sanctions manager, um, our developer community is very strong. We're based in Tampa, Florida. Jerry Russo and Campbell Ben, not Ben Campbell. Campbell, Ben, um, Campbell and Jerry are in the audience today and can answer any questions you may have, as well as Kate and myself. Um, if you have other questions after the session. With that, I'll turn it loose for questions you may have on JPMorgan Chase and sanctions automation at scale. Yeah, there are, um, microphones in the room, and I think there's a walkie somewhere. So I've got a three part question. Is that.

Oh, my. Three. Is that cheating? Listen, we left lots of time. I was only going to leave 15. I left 18, so. I lost them one at a time. Though. You can't expect us to remember all these.

You know, I was going to say these three list things are a nightmare because I'll remember two of them and I'll forget the third by the time I come to the third. But the questions I've got are against the backdrop of the evolving landscape of payments more generally. The first question is, um, around the springing up of new payment schemes all over the place. So RTP schemes are springing up, Artiguez schemes are evolving. Typically sanctions has been more focused on B2B type infrastructure. Are you applying GSM to all of these other schemes that are springing up, including all the RTP schemes and the peer to peer type schemes? Yes. Is the short answer. Um, and it depends on the jurisdiction.

Right. So, um, if you are a UK bank located in the United Kingdom and you're doing a domestic payment, you don't have an obligation to screen. We as a US bank doing payment activity in real time payments in the UK do have an obligation to screen because we're US regulated and it's by definition, a cross-border payment, even though it's just going to two people in the UK. So we do have sanctions screening there. Um, I don't know that we've necessarily cracked the nut in terms of the end to end operating model, nor, frankly, the end to end infrastructure. Um, but we're working on it. Each market has different requirements in terms of service level agreements, response times, outcomes and responses back to the consumer. There's data privacy things. So there's a whole host of complexities there that we are working with our payments product folks to normalize as much as possible and influence the market participants so that we can make it the best possible, both for the actual the participant themselves in terms of who's actually trying to move the money, but also in terms of the user community that actually have to disposition those alerts.

I guess 11A is does that also apply to things like crypto and things like that as well? J.P. Morgan does not participate in crypto. No kind of distributed ledger type stuff. Um, we do internal distributed ledger and we do apply global sanctions manager and focus soft in that context. And we do screen digital currency addresses. However, we do it in the context of cash payments and do not participate in a lot of the crypto staking and clearing market activities that maybe some of our competitors may. Okay. Part two is around, um, customer experience.

So the expectation now of payments is more real time, real time, faster transparency and all that good stuff. Obviously, sanctions is a huge roadblock when it comes to making payments. What are you doing to kind of get ahead of that, um, and stop that friction and make things more seamless for a customer. So they're not thinking, well, hold on, why is my payment not completed? Um, oh, you gotta turn loose my product partner. So there's a there's a series of things, um, you know, first of all, it's around education, right? Um, there are things in a US context that you don't want to include in a payment. As an example, I don't care what type of sandwich you're ordering. I don't care if you're having a Greek sandwich or a Cuban sandwich.

I just care that you're having a sandwich. So don't tell me what type of sandwich it is. So that's one thing. Also, educating our user, our consumer base, to say you need to articulate things on the payment. Right. So don't include acronyms. Don't include you know, don't tell me that this is not a Cuban something because that's going to stop it anyway. Right. This is not related to Iran is just going to stop it because you put Iran on the message.

Um, a lot of the AI and the automation that Brian has talked about has really helped drive better service level, um, delivery in terms of our end to end turnaround time. So, you know, I don't know how long it's taken us, but as Brian said, we're at nearly 10 million messages screened. And that a significant portion, probably about 80% of that is payments. Um, we automate about 88 to 90% of those alerts that are generated so that a human never has to touch them. Right. So we are operating at about 75% lower than our prescribed SLA, which from a client experience perspective means in a real time payment context, the client hits, you know, pay Kate and they expect to get a response back immediately if there is a fuzzy alert on that. Typically, you know, they may not get a response back in, you know, several days in the old, old world. Now, we usually are able to respond in less than 30 minutes in many cases. And the final part, you almost kind of answered it in your answer there.

But it relates to non payment messaging. So obviously the crux of sanction screening is really to stop the financial transaction from occurring. Um and what we are seeing is obviously there's a huge volume that sits on this side of that, which is the non payment messaging that might be to relate it to the exception, perhaps, but those messages typically are getting screened as well. Is that really necessary, given that it's that given the fact that, for example, modifying a payment, you'd want to kind of control that, to make sure that somebody's not sending a nefarious payment and able to use nonpayment messaging to modify the payment. Given that that kind of behavior is disappearing out of the industry, why are we screening every single message and then potentially causing holdups in that process? And is there an opportunity for the industry maybe to eradicate that or solve that challenge somehow? Yeah. So there's lots of discussion in the Swift community. There's lots of discussion with Wolfsburg and others.

You know, Other, we're talking to other banks about opportunities to rationalize some of the messaging and some of the content frankly, associated to the back and forth when we're asking and answering questions to one another. Um, some of that tends to be very flowery. When you put it in a Swift message, it kind of wraps around to another line. And does the filter pick it up properly and all of that kind of stuff. You know, there's not a simple answer to that question. It's every financial institution has their own risk tolerance and policy that we have to adhere to. We work very closely with our line of business and compliance partners in terms of defining what those internal policies are, and then obviously, we have the regulatory expectations that we have to meet. And depending on how your flow works end to end, if you have an amendment on a transaction that's still sitting in the payment engine, you know, if the payment engine is going to resend it with the amended information, then obviously we don't have to rescreen the the amended the amendment instruction. Right.

But if we're not going to rescreen the in-flight transaction, we would have to screen the amendment. So it's a little bit complicated, and it's not necessarily as black and white of an answer as you would think. I hope it does get a lot clearer with the end of the CBR plus coexistence period, right? When we're all Amex and we all, as financial institutions, are comfortable with the ISO 2002 two context, and how to appropriately join the different data sets around a KMT versus a PACs or a pain or another payment type message. Once we can understand, oh, we've already seen this and this is what is new, then that's the sweet spot for getting some some traction on the quote unquote, over screening or rescreening that occurs multiple times in today's environment. Yeah. And it's not just payments, right? I mean, we screen asset movements. So we're screening securities instructions, trade finance.

You know, it covers the entire gamut. So it's not just physically around payments. Although as I said it's about 80% of what we do. You always wonder where it's going to stop. Is it just with messages. Or they can start screening emails next. And that is my biggest nightmare. And I plan to retire before that happens. I do believe they were all the correct answers, so thank you very much.

Any others? Other questions today. About ten minutes left if you've got more to answer. Thank you, Ryan and Kate, for the presentation. Um, I'm Gopal from Citi. Um, this use case is, uh, one of the projects that we are currently working on. It closely correlates with what we are doing, some of the, um, challenges, what challenges you faced when you try to auto detect the, um, the AML block, for example, if the client has already AML block exists and you got the feed from outside and you are trying to sanction that client with a transaction override or, um, to allow to sanction the to to to approve the transaction to go through. Is that everything going to be is everything done through the STP process? Like how much percentage of that STP has gone through and how much manual review that he control with the legal, compliance and risk approvals with manual routing and taking action.

Do you have the balance, like how do you control the real time alerts coming in, and how do you have that consistent? It shouldn't be false positive, but it has legal risk, um, complications when you try to leverage STP versus manual review. So how do you control that. So we draw some lines around what the AI is able to auto decision. So there are some things sectoral sanctions is a good example right. Where purpose of payment is key. As a driver, excuse me in terms of whether or not something falls under sanctions jurisdiction or not. So we typically have the AI skip that type of decision. Um, we also have logic built into GSM and into the screening solution that we have that actually then auto routes that to a tenured agent so that we don't hand it to someone who just started sanction screening and expect them to make a decision on a highly complex item.

So I don't know if that answers your question or not. Do you know how much percentage you currently do SDP rules versus how much percentage you do manual? Do you have a. Sense of I know exactly how many we do, but that's not typically something that we communicate in a public forum. Okay. Thanks. And I know he touched based the cross-border policy and PII assessment. And in terms of the region cross-region accessing the transactions for the client, do you implement any data visibility rules within the Pega Platform? Absolutely.

Yeah. You do. Yeah. There are certain jurisdictions for which we will firewall or isolate data around the world. And it's all Within the policy and the data appetite of the particular regulator and our compliance organization. So obviously China king of Saudi Arabia got it. Some other jurisdictions. Thank you. And you said you are already on Cloud right.

So are you leveraging the features of Customer Decision Hub along with Pega Platform customer, CDH and Pega together, or is it only core Pega Platform? Yeah, not 100% yet. We're still we're still establishing our baseline on the Infinity 23 Platform from there, we'll start to leverage a lot of the capabilities within the most currently available set from Pega. But guys today. No. And I see the interesting aspect of you talked about the skill based routing and all the good features of the Pega Platform. The model is one thing. Interesting you talked about the supervised learning AIML, um, in terms of risk prediction. Um, how how do you basically reduce the false positives, um, from the model perspective and how Pega has been integrated, uh, with your model, is that model on on premise on JP Morgan Chase or is it on cloud?

Our models are currently implemented on prem. On prem, on prem. And we leverage, um, a very, very firmly bounded data set. The data set is not a extremely public data set. Obviously what is done in GSM is used within that context only within the bank. Um, and we don't have any plans to go in anger in anything to public cloud. From a generative AI perspective, we still keep that within the boundaries and context of the internal on premises J.P. Morgan Chase implementation. Does that answer your question?

Yes. How large is your model? Do you deploy for the sanctions? How large. Is it? A it's large. It's been trained. It's been trained on many years, over 2 million human decisions. And the decision tree is quite large.

It's this big. Okay. Thank you. Appreciate it. You're welcome. Any other questions? We have about five minutes left. If anybody else has anything else, or if you just want to get up and stretch your legs after a long day and wake up again after talking about sanctions for 45 minutes. Anything else?

All right. Well, we'll let you out about five minutes early. Thank you.