Webinar: Collaborating with AI Tools in Financial Services

Risks and regulations dramatically slow the use of AI tools in financial services, especially if the use of AI involves private data.

In this webinar, TripleBlind and UnionBank of the Philippines will discuss why enterprises need to be prioritizing collaboration around AI tools and how to do it securely and privately.

The topics to be covered include:

  • Challenges in collaborating with AI in financial services
  • Opportunities in operationalizing AI in private financial data
  • How to securely and privately collaborate around private data using AI tools

 

Begin Transcript:

Chris Barnett:

All right, everybody. Welcome and thank you for joining today. We appreciate you joining this session with UnionBank of the Philippines and TripleBlind on Collaborating with AI Tools in Financial Services. We’ve got a great panel with a couple of short presentations and then we’re going to have a panel discussion and I would just encourage everybody in the audience, feel free to put questions into the Q&A box and we’ll give questions to the panelists as we go. So I’m Chris Barnett and I’m your host today, and I’m going to introduce our panelists here and then we’re just going to get right into it.

Chris Barnett:

Really appreciate everybody joining. Dr. David Hardoon he’s the head of the AI Center of Excellence at UnionBank of the Philippines. Welcome David. Thank you for participating. We also have Dr. Adrienne Heinrich who’s the Chief Data and AI officer also at the UnionBank of the Philippines. Adrienne, thank you as well. Also we have my good friend and colleague, Samir Mohan, who’s the partnership engineer at TripleBlind. He’s a senior technical resource to TripleBlind and all of our customers and partners.

Chris Barnett:

So, Samir, thank you for being here as our subject matter expert. So we’re going to kick off with David and an intro to UnionBank of the Philippines. David, the floor is yours.

David Hardoon:

Thank you very much. Just to check if we’re … Oh, perfect. Just a brief one, so Adrienne’s the Head of the AI COE. I have the pleasure of [inaudible 00:01:40] also on the government side of things. But anyway, thank you very much for the invitation and for the opportunity to speak. Very briefly in terms, and I think if you could just go to the next slide, UnionBank of the Philippines, it’s a Philippines-based financial institution operating … In fact, we’re celebrating tomorrow, no today, our 40th year. So happy anniversary to us, we’re 40 years old now which is coincidental in terms of aspect.

David Hardoon:

What I wanted to actually introduce in terms of the bank is more specifically the journey that it’s taken, and very briefly if I may, over the last almost six to seven years, starting off as it a local bank, really in seeing how can it leverage and digitize and leverage in technology in terms of providing financial services and products to the Filipinos? As you may or may not be aware, we’re talking about 50 to 60, and back then it was even higher percent of the population is essentially either unbanked or underbanked. As you can see, the ability to tap on nonconventional traditional means of financial services which would be usually your agents, relationship managers, branches and so forth, which an organization at [inaudible 00:02:57] scale back then is extremely and absolutely important.

David Hardoon:

With that, our continuous growth and leveraging on data science and AI is really that next step of as those digital assets are being created, as those different means of engaging how do we tackle it? How do we go beyond and provide that degree of service that we are holding ourselves accountable? Whether or not there’s a relationship manager to know your [inaudible 00:03:24] person name. Now we can have other means of doing that.. So very briefly on UnionBank of the Philippines, and as I’ll just give the catchphrase of banking starts with you. If we go to the next slide, I wanted to also maybe spend … as a segue of that, a couple of minutes as to the importance of this topic and what we’re doing and that’s why it’s important to us.

David Hardoon:

So I’ve hopefully made a compelling argument as to why we’ve gone down that path of using technology and using alternative means that may be available disposable, at the disposable arms of the bank. Therefore also how we want to go about leveraging, be it computer vision, be the other array of data science, machine learning and AI techniques. With that, as you may be aware, financial institutions and financial sector is … I would argue if not the most regulated industries, one of the most regulated industries, and as it should be with respect to the governance and fiduciary responsibility, because at the end of the day, we are the custodians of you effectively, of your data.

David Hardoon:

So how do we balance those challenges of on the one hand, security governance, controls, which should and must be in fact, be there as a hygiene with those opportunities to flourish and exploring new avenues and new solutions? It at times a bit of a paradox, because the traditional sense of how development will be done is say, “Okay, let’s just all jump in and brainstorm.” But what happens if we can’t share the data? “Oh, let’s brainstorm.” But I can’t [inaudible 00:05:06] I can’t do exploring, how to extend collaboration, specifically within AI data sites, with AI tools is such a manner that allows us on the one hand to maintain and uphold privacy, security and governance as I mentioned.

David Hardoon:

Because that is an absolute hygiene and responsible element, with the opportunities that quite frankly, we didn’t know, we couldn’t have explored because we didn’t have the opportunity of working with this research institution, with this other partner, or this other organization. And how do we go about and finding those underlying solutions that can effectively improve and increase customer specificity and personalization, operational efficiency and also risk management and control? So this is really, really important. Now, just as a final point, it’s not that it doesn’t happen today, it does.

David Hardoon:

But we spend what? I think an average of eight to 10 months of signing a piece of paper, then another whole bunch of months to get data across and … sure, it happens. But, if we’re seeing the speed in which adoption and acceleration is happening in the tech world, excuse me, we want to be able to do that in a much, much more compressed and accelerated passion and getting solutions, or if you even just exploring possibilities as fast and underlying possible. So this is freely the underlying core and the importance of how do we extend our possibilities and our capabilities as a financial institution, has some time, the tech companies wouldn’t flip our hats to collaborate, to share hands.

David Hardoon:

If I reference to my … which now seems to be further in the distance past, research and academic career, the true breakthroughs happens through sharing and collaboration. I don’t know if I have a next slide or hand it over to Adrienne. Just go to the next slide. Oh yes. Sorry. So I do have just one more point. I touched on this, but I wanted to give a couple of more examples as to why this underlying collaboration is truly important. As I mentioned earlier, is that customer centricity in terms of how do we make sure and build around various solutions that look at, well, you.

David Hardoon:

And the end of the day, banking, financial services, I always half joke that it’s actually the most boring thing ever. It’s just money. But, it’s the most important one. Everything that we do when you go by you go buy a car, when you deal with insurance, healthcare, thinking of your future house, everything ultimately is situated and substantiated on a financial bus in a lack of a better word. So how do we go about truly personalizing and accommodating one’s need? So really having that centricity. It’s combination of those different worlds. How do we do that? How do we explore that?

David Hardoon:

That has to be done through collaboration. That naturally leads to the multi-sectorial approach, providing better and more specific products, more underlying, relevant solution. How do we enable that? Nowadays, that’s a very high bar to cross, not that it shouldn’t never be a less of a bar from again, governance, privacy and security, but those can be achieved with the right capabilities and the right techniques. Data sharing, like I said, it’s about you. And now obviously with open banking while there’s a legislative aspect that facilitates that, how do we really go beyond and how do we think about it? Funnily, working backwards, creating solutions that are relevant.

David Hardoon:

So, just the tidbit of the iceberg of what’s possible, why it’s important to us and then specifically from UnionBank’s perspective. So, thank you with that. I think hand it over to Adrienne. Oh, back to you Chris.

Adrienne Heinrich:

Thanks David. Yeah, I’m Adrienne. As introduced, I’m heading the AI Center of Excellence. And as part of the AI series role, we are exploring mid and long term innovation opportunities that can be fed into the bank when successful. So for us at UnionBank, it’s very important to put the customer at the center. And therefore we seek to develop customer-centric AI solutions. Now, besides looking at AI solutions for banking-specific features, we are also looking at innovative solutions beyond banking. Since our lives are not centered only around our financials, right? So if we wish to provide more value to a customer’s life, we should also consider value ads in adjacencies.

Adrienne Heinrich:

This will allow us to treat the relationship that we have with the customer in a more holistic way. It’s basically about helping the customer with the area of his or her interest, the area that the customer wants or needs to spend money for. And we can actually see that this is really a customer need, when looking at major topics that come up in personal finance forums on the web, we can see that less 20% of the conversations actually mentioned bank products or bank services. And the majority of topics is actually related to new areas of personal finance. For example, this could be household budgeting or education or car purchases.

Adrienne Heinrich:

Yeah. So now to create more value for our customers, we are exploring partnerships with external players in the industry or academia. This will enable us to be fast with innovative advances. Let me give an example in the health space, chronic diseases are not only health wise, a burden, but they also present a large economic burden. So for the individual, he or she would like to live life at highest possible quality rate at affordable cost. Therefore we can think of a hypothetical use case where we as a bank get access to a large pool of data of patients with certain chronic disease, X, Y, Z.

Adrienne Heinrich:

And by running machine learning models on this data, we can give financial advice for the coming months and years to be able to cover the cost for the treatments and have patients select their preferred life enhancing products or services accordingly. And we can also easily enhance the dataset with our own data of the customer of our customer, who is actually in this use case the patient as well. And that can be shared back to the care provider if the patient agrees. So for example, a patient behavior that would deviate from a predicted trend that could lead to a flag at the care providers.

Adrienne Heinrich:

Now in order to explore this use case with a selected partner, we would need to have agreement from the patients of the care provider that their data can be shared to us, and also from our customers that their data can be shared to the care providers. We would need to have several contractual agreements in place with a care provider company data sharing agreement, data processing agreement, et cetera. And then accordingly the data would have to be de identified or pre-processed in some way so that we will comply to the type of data that is shareable across the borders of all involved organizations.

Adrienne Heinrich:

Now here you can really see that several steps need to happen before we can actually kick off the work. Another example where we may want to collaborate, and what we are actually doing quite often is with academic partners. Because academic partners, they usually develop a lot of knowledge in a particular field in our case, in a particular field of AI. At the moment, typically our academic partners, they cannot access our production data. So they need to work on public datasets and they’re not always very similar to our own datasets. So then there is quite some overhead to take their developed model and redo the experiment on our own data. Maybe even take few steps back because the performance was not the same since it’s slightly different data, right?

Adrienne Heinrich:

Then we have to redo several steps of the model development. So it’s not a very effective approach. That’s why for us it is really so important to work towards solutions that make collaborative innovation easier so that we can leap frog together with valuable new products for our customers. Yeah. That was it from my end Chris.

Chris Barnett:

Thank you, David, thank you Adrienne. That’s a great overview. I really appreciate that. So next we’re going to go to Samir and he’s got a short presentation on some specifics about collaboration tools in the world of AI, and then we’ll have a panel discussion. As I mentioned, everybody should feel free to post questions as we go.

Samir Mohan:

Great. Thanks Adrienne and David, for that description, I think at length of not only the problem, but maybe the reasons why it needs to be solved, why we have a really great opportunity here to enable AI and do it in a way that’s regulatory compliant, that’s data governance compliant. So just brief background about myself, Samir Mohan Partnership Engineer at TripleBlind. I’ve done primarily big data platform, architecture and engineering, AI platform architecture and engineering at large financial services companies, professionals consulting companies and startups. And I have a background in cloud and hybrid cloud implementations in a number of different languages.

Samir Mohan:

I think what I see common across that myriad of company size and industry verticals is that AI is an opportunity, but also a challenge in the compliance and the regulatory requirements in order to get it right, in order to get the right answers in order to have models that are effective, performant and like I said, compliance. So, I’d like to introduce TripleBlind to the esteemed attendees of this webinar. TripleBlind is essentially a privacy suite, a set of tools and techniques that have been wrapped under the covers of a very expressive API and user interface that allows for analytics and AI and machine learning tasks in a number of different industry verticals, healthcare, financial services.

Samir Mohan:

Primarily in order to enable data scientists and analysts … excuse me, to collaborate, to collaborate across organizations, to collaborate across geographic boundaries in a mathematically and cryptographically provable secure and private way. So, you’re going to hear some fairly fancy terms here and some different understandings of how privacy and data compliance are approached. And we hope that that is of interest. So, as David and Adrienne described and who better to describe the problem than customers, the stewards of AI of an esteemed financial services organization like UnionBank?

Samir Mohan:

Who better to describe that problem, but maybe graphically, this could give you a concept of what that looks to the sheer number of steps and checks and processes that are required in order to even do simple analytics, joining two datasets in order to derive analysis of an organization that may even be your own subsidiary, that may even be a company inside of your multinational umbrella and you simply want to gather that dataset, bring it in-house and perform some model training for example. It’s actually quite difficult sometimes depending on the regulatory burden that applies in either jurisdiction that you may be subject to.

Samir Mohan:

So, it’s a lot of steps and frankly data decryption, data transmission, data replication, complex non-deterministic process, having to trust your third parties is a challenge. It has great perils in addition to the great opportunities. So, what TripleBlind does is simplify that process through math and cryptography. We take a different approach than your typical maybe tokenization or synthetic data, or homomorphic encryption or some of these other privacy techniques and make that workflow of determining assets, determining datasets that you want to work with, determining models that you want to leverage and boiling those down to their bites and transacting without needing to have some understanding of that particular object.

Samir Mohan:

So instead of having to understand the specific elements of a dataset that need to be private, why not just encrypt the entire thing and perform your operations on that encrypted data? Beyond that, we have some privacy and intellectual property protections that allow organizations like UnionBank to build models, to license models, but to have the security and privacy and intellectual property guarantees that allow for that fast collaboration, that fast commercialization, but to do so in a way that doesn’t require the trust and the many checks and balances, the contracts and indemnification clauses that you would otherwise need when operating in a more conventional sense.

Samir Mohan:

TripleBlind approaches this data, privacy and compliance in a different way. We essentially say, leave the data where it is behind the organization’s firewalls, and then encrypt that source data as a one way encryption and allow for a counterparty who has intellectual property, a model say or another dataset to be able to transact on that encrypted form of the data, without any loss of fidelity, without any changes to the underlying data and how it’s distributed, so that the models can be trained with absolute accuracy and allow for those models or those analytics outputs to then be reused further down in your pipeline of leveraging for commercialization or retraining continuous learning, things of that nature.

Samir Mohan:

So it’s a really sophisticated process that occurs under the hood. But what you see on the surface is a very expressive set of development tools that allow a data scientist to incorporate into their existing flows or to build new ones in a very intuitive way. So practically speaking, what might that look like? What are the use cases here and Adrienne and David touched on this as well, but to dig a bit deeper, we have the concept of vendor management and financial services or in healthcare contexts and government use cases. You have a great number of suppliers and providers in your business workflows.

Samir Mohan:

The concept of not needing to trust your counterparties with the model that you’ve been building over a great amount of time with your proprietary data or not having to trust your counterparty with your private dataset that contains your customer’s financial transactions. But instead allowing that same level of collaboration, that same level of insights within organizations between organizations, across geographic boundaries while asserting cryptographically provable protections and privacy, is I think a strong use case for a solution like TripleBlind.

Samir Mohan:

We also have credit card fraud detection use cases where again, in the context of say federated learning, where you have private datasets, a private model that is desired to be created. The concept of encrypting once, never decrypting the bytes of data are representing the financial transactions of three different banks who have zero or very weak trust between each other, yet still able to create a machine learning or an AI model that can then be leveraged for like I said, commercialization licensing, maybe potentially back to those originating banks within your say multinational organization or between organizations is also a strong use case.

Samir Mohan:

Finally, alternative data. Adrienne talked about the concept of bringing heterogeneous datasets, bringing datasets from different areas of your business, from different suppliers into your particular analytics or AI use case where you want to enrich your datasets, you want to build really robust AI models that leverage many more hyper parameters than you might have otherwise been traditionally training with. That’s the power I think of leveraging a solution that does not require what we call semantic understanding of the data. So the data scientist is free to play with those data assets as if they were SQL data stores on their location.

Samir Mohan:

Those datasets can actually be completely remote, completely governed by different data governance policies and compliance rules. So this concept of bringing that enriched data in for analysis is very strong. I’ll just end with the concept that TripleBlind is quite different. I think that’s why UnionBank and TripleBlind have partnered up to run a proof concept and explore options there. We want to help UnionBank leverage our privacy suite for their very sophisticated analytics and machine learning and AI use cases. I think UnionBank selected TripleBlind in part because we approach things in a different way than the techniques they may have been examining prior.

Samir Mohan:

So these things like GDPR compliance are areas of focus for us. And we hope to discuss that more with you in the Q&A section. So I’ll [crosstalk 00:26:09] that to Chris.

Chris Barnett:

Great Samir, thank you. I appreciate that. So two great presentations. So I’ve got several questions coming in from the audience. So I thought I would start the panel session off with those if that’s okay. So first of all, and I guess this is probably for you Adrienne, but maybe Samir [inaudible 00:26:29] coming also. It says, for the current stage of cooperation between UnionBank and TripleBlind, what are you hoping to accomplish? What are your goals? Just for right now.

Adrienne Heinrich:

Yeah. Good question. It’s really on topic. Yeah, for us when we are developing machine learning models without the TripleBlind solution, right? So just the usual way of working, we would need multiple iterations to improve the model performance. And when the performance isn’t still unsatisfactory, we would try to understand when it underperforms and adapt it such that it has a higher chance to improve its performance. Now in the POC that we are having with TripleBlind, we wonder whether this blindness towards our data, for the third party would actually hinder our partner to effectively improve the performance of the machine learning models.

Adrienne Heinrich:

So that’s what we don’t know, what we would like to understand better. And if it does not hinder our partner in the development, that it’s okay to be blind towards the data, and that it’s really sufficient to work on our data and to improve model performance. How did then our partner deal with this obstacle to overcome it easily? Because it’s not our usual intuitive way of working, but I’d really also like to hear some view on this.

Samir Mohan:

Yeah, absolutely. I think the experience that I’ve had prior really informed why I think TripleBlind is so powerful. When I look at the processes that data scientists typically use. In the very ideal world, they’ll always have access to the raw data, but the amount of compliance and trust and data governance, red tape as it were that a data scientist has to go through, or that the company has to endure in order to get access to really high quality data is as big of a challenge at least. I think an approach that addresses that concept of confidence and approaching a zero trust approach to data scientists being able to work on very sensitive data potentially or very identifiable and very personal data, I think is really where TripleBlind comes in as a strong differentiator to the other solutions out there.

Samir Mohan:

We’re very cognizant of the concept of data scientists needing to look at the data on occasion, depending on the processes, depending on the amount of pre-processing they need to do, depending on the nuances of the data. But we believe that a very intentional approach to defaulting, opting for privacy and allowing elements of data to be revealed at certain levels of visibility, whether it be through a synthetic data view in order to understand the data composition or identify outliers, or identify correlations between the variables, versus unmasking certain fields that allow for a data scientist to understand more deeply what the raw data looks versus a full privacy where you run your machine learning jobs because you’ve understood the composition of the data and you’re able to then do your iterative training or do your iterative pre-processing without having to see quite all of the private details that you had previously.

Samir Mohan:

I think that’s where we have some strength. So it’s a default of private with the option of some level of disability and really we want that onus, we want that ownership of those rights to be on the data owner. So that’s how we give that level of control back.

Chris Barnett:

That’s great. Okay. Thank you. Thanks both of you. So, that’s a great segue to the next question here. One of our audience members says, sometimes setting up a new tool can be hard and can be challenging. Can you give us a sense of what’s required to get started and get going? So Samir, maybe you want to comment on that and then if Adrienne has any reaction from her team using the tool maybe.

Samir Mohan:

Sure. Another one of my favorite things about TripleBlind for example is that because we are API native, because we are Docker and Kubernetes native approaching our solution from a very easily integrated, very modern approach to software development and product development, things are very weekly coupled. So a Docker instance is set up with four or five commands, takes maybe an hour or two to set it up by a single DevOps resource and a single developer working with our Python SDK can get familiar in probably three days or less. It’s a very intuitive process for a data scientist or a data analyst, because it looks a lot like the same code that they typically write. So it’s very quick to get started.

Adrienne Heinrich:

Yeah. For us, we did have some starting issues just … yeah, indeed because it is like a different tool. We have to get used to it, we have to understand how to get started, but what’s good is that there is always support whenever we have questions. That’s what it is about, if it takes a bit longer at the start to get it started, when you’re used to it and you know the tool then that is forgotten later, it’s just a first stepping stone. So, what is really important that we get there and that we get to do the analysis that we are currently doing.

Chris Barnett:

That’s great. Thank you. So the next question here, pivots away from the technical side back more to some of the use cases. So one of the use cases that was mentioned was credit card fraud detection I believe, talking about credit cards. So this person’s asking about how do cross-border transactions or cross-border activity come into play there? Either of you just jump in if you’d to comment on that.

Samir Mohan:

Sure. So, from experience with another client that we’ve been working with, we have an Asia Pacific opportunity. It’s a conglomerate I would call it, a organization with a lot of subsidiaries and a lot of joint ventures in the Asia Pacific region. And they have datasets at these different subsidiaries that they want to be able to bring in to build a federated model, a model where the dataset stayed private because of the geographic and data residency constraints that are placed upon each of those different datasets at each of those different subsidiaries. So, what they wanted to explore with TripleBlind, they had looked at another opportunity and that company folded and they looked at TripleBlind as a really great option was the concept of not moving any of those datasets in these different subsidiary companies who have their own data governance like I said.

Samir Mohan:

And instead being able to collaborate with that data in a private way. So, that’s I think where you have customer information from these different businesses that are semi autonomous and a Center of Excellence at this particular multinational that wants to be able to enjoy the benefits of that scale, of that breadth of their company, but not be subject to the siloing and bureaucracy and red tape and processes and manual checks and balances that are required in order in a conventional sense, ship that data across those firewalls and into the Center of Excellence. So instead they want to use TripleBlind to do it in a more modern way I would say.

Adrienne Heinrich:

For this one, maybe not just on credit card fraud but in general on financial crimes, I would just to mention that it can be very valuable to collaborate across your own organizational borders with others abroad or within your country, because usually when things happen like money laundering or scams, phishing, the money gets distributed. There’s this scatter gather pattern and it gets scattered outside your organization really fast so that you lose track of who the criminals are, where your money is going. So therefore it’s actually very valuable to collaborate together to be able to do network analysis across different institutions. I think there, there would be actually really a lot of profit for that.

Adrienne Heinrich:

Because there everyone is actually on the same side, you’re not competing with each other. You all want to catch them and make life harder for the fraudsters.

Chris Barnett:

Okay. That’s great. Two great answers there. Thank you. So the next question coming in from the audience, I think Adrienne this one will be for you and they are asking about UnionBank’s future plans, obviously answer only what you can, but the question is a good one. It says, are there plans for UnionBank to commercialize data/algorithms and introduce new revenue streams?

Adrienne Heinrich:

Definitely to introduce new revenue streams. So we are actively looking at what will let us leap frog? What are the new game changing innovations that we can be involved in and that we can drive? So, definitely thinking also about are there new revenue streams? Not necessarily directly from commercializing data. I would word it this way. We want to work with data, we think data is extremely valuable and we want to bring data together. Like what I said, to bridge industries because we really see the power of doing that. Typically, AI works quite well on a very well defined problem in a silo, but where we want to get to, is that we can just throw all the data that we have so that you really get this 360 degree view of a certain use case or a customer.

Adrienne Heinrich:

That will really give you this big jump forward, but we cannot do that if we don’t start. We as the world, we really have to make lots of steps still in there to get there, to fail, to learn and to try again. And for UnionBank, it’s not strange to us to think that way because we’re a part of a conglomerate, it’s called Aboitiz and this conglomerate it’s really 100 years old. It’s a very big and old conglomerate in the Philippines. There are many industries, so there are lots of business units covering power, land and real estate, infrastructure, obviously the financials like bank and insurances and also food. So we have quite a widespread industry based there and we are also looking to act more actively, collaborate together. Yeah. And that means really also data sharing and bringing our data together.

Chris Barnett:

Okay.

Samir Mohan:

Okay. Actually [crosstalk 00:39:57].

Chris Barnett:

Samir, go ahead.

Samir Mohan:

Yeah. I think that’s so very interesting. The concept of maybe augmenting credit underwriting, is that something that you’ve considered maybe utility customer at that subsidiary is making on time payments and all of a sudden they ask for this massive credit line on a product, a mortgage or something that, and you may be able to get some insights out of seeing the credit worthiness, the on-time payments from that utility payments from that utility customer.

Adrienne Heinrich:

Yeah. Very true. That alternative data is something that we are also investigating, especially for the Philippines there are lots of unbanked people in the Philippines. I think it’s maybe 50% of the population. So naturally also lots of people who would want to get a loan, they don’t have the necessary data or papers, history to actually check all the rules. And the good thing is with AI, you can actually use this alternative data to score the people and then do a risk scoring and still give them the chance to get a loan with which they could not get with traditional methods.

Adrienne Heinrich:

So that’s how we try to drive financial inclusion, as well with such type of data that you mentioned.

Samir Mohan:

What a great point, right? This is not to get too teary eyed, but this is an altruistic benefit of AI. Inclusion, you look at the healthcare outcomes that can be improved by augmented clinical diagnosis or drug development, and I think AI is really, it’s a force for good when we use it for.

Chris Barnett:

Fantastic. This is great. So we had two audience questions left and I’m going to say, I think the next one you actually probably answered. If you want to throw out anything additional, that’s great. But the next question was, what about datasets outside of financial services? Maybe you just covered that, or maybe there’s a little more commentary on non-financial datasets in collaboration.

Adrienne Heinrich:

Yeah, indeed. Mainly what I was saying, maybe I can just stress that I really believe in future we will see much more of those industry bridging use cases and open innovation and partnerships are a very important aspect in driving … and enterprises innovation, maturity. So more and more collaborations will take place with the goal to bring those datasets together and to leapfrog with the gain value. Yeah. And it would really allow us to understand our customer better and then to provide better value. It is a clear win-win for us.

Chris Barnett:

Great. All right. So last question is we’re coming up on time here and I do want to try and squeeze this in, although it’s a big question. Again, either of you folks feel free to jump in. So what about experiences with other privacy techniques? We’ve been talking a lot about TripleBlind, but what about other techniques, tokenization, there’s synthetic data, many people have heard of there’s differential privacy. What are your experiences there? What comments could you make on those?

Adrienne Heinrich:

Yeah. If I go first, so for a machine learning model to work well, it needs to be trained on a set of data that is actually representative to the real world case. So all the situations, patterns, behaviors that you would see in the real world somehow need to be covered by your dataset. Meaning if you want to work with for example, synthetic data, if certain behaviors or patterns are not covered, the machine learning model will more likely fail. So if you know your real dataset, well you can create quite a good synthetic dataset. In my experience that also worked, but you need quite some knowledge already. What I also experienced is that we didn’t have so much knowledge.

Adrienne Heinrich:

We created a synthetic dataset and it didn’t really work well, and then you need to find out what is not covered, where do you have the gap? So, it is not always an ideal situation, but also in this direction, maybe more research needs to happen to be able to create better synthetic datasets.

Samir Mohan:

Yeah.

Chris Barnett:

That’s great. Samir final remarks on that.

Samir Mohan:

Sure. I would just say from my end especially as an engineer I’ve spent a lot of time dealing with test data, test environment management and maintenance, promoting assets from test environments to production, and then trying to reverse engineer what the problem was that caused a deviation between outcomes and test environments versus production. It’s not very fun, it’s something that I think is incumbent on data engineers and data scientists as they progress through their careers. But I really find that the concept of having data which is cryptographically and mathematically provably transformed, one way encrypted and not able to be reconstructed as either the individual members within the dataset or the algorithm that was derived.

Samir Mohan:

I think that level of safety and privacy in your assets allows people to feel more comfortable about building your AI model or building your analytics routines and deploying them straight to production with 100% fidelity of your underlying data. So the model looks the same as if you had trained it in the raw. In fact, the performance is absolutely fantastic. The scalability is absolutely fantastic. Our patented approaches are extremely performant, and I think that’s one of my favorite value propositions for an approach like TripleBlind is that you get rid of so much of the chaff, the excess, the inputs that are just inputs for your final output, your test data.

Samir Mohan:

Like that replication, that maintenance, why not cut some of that out, and instead use an approach that just operates on your real production data, but in a mathematically provably safe and private way. I think that is from an engineer perspective my favorite value proposition from TripleBlind. So, thanks for that question Chris.

Chris Barnett:

That’s great. I love that energy and the enthusiasm from both you guys Adrienne and Samir. Thank you so much for your time and for the good conversation. We’re just going to go ahead and wrap up now and thank our audience for coming. Thank our panelists. We will send out to everybody who’s registered a quick email with a link to the recording and specifics on how you can follow up if you’d to learn more. So thanks everybody and have a great day. Appreciate your time. Take care.

Adrienne Heinrich:

Thanks. Thanks for having us. Thanks for the exchange.

Chris Barnett:

Thank you.