Demos: Alloy, Inscribe

Transcription:

Brice Pratt (00:09):

Hey everybody, we have eight minutes, so this will be quick. I'm Brice Pratt. I'm one of the Managers of a Solutions Engineering team at Alloy. So we'll talk a little bit about what Alloy is and then jump into the platform. Probably one of the best ways to describe what Alloy is, is actually to talk about what it isn't. And the one thing that Alloy is not is a data vendor. We are not a data partner. So Alloy is a single point of integration, API based that allows our clients, our customers, to be able to interact with multiple different data services. So there's about 200 we're integrated with right now, and that's going to span across KYC, AML, fraud, third party, first party behavioral device, and the list goes on data services about the business, about accounting as they were just talking about. So we are a single point of integration.

(01:00):

Our clients are able to integrate to us. They send one API, we are able to pull that information and then retrieve that back for them. So the idea is integrate once and then you're able to build upon that policy and iterate over time. So what we're looking at today is actually just building out a policy. So on top of being API base, you can send us from wherever. This is also going to enable anybody that is in charge of building out the policies. So whether it's fraud, AML business or KYC, whatever that might be, you have the ability to then run test, change, update in real time without needing any type of development or sprint that needs to be deployed. So actually what I want to do today is go through and just build a policy. The one thing that we are looking at right now, this is going to be the blueprint overall.

(01:54):

So this is where we can dictate what does it actually take to onboard a business or the business owners or whatever that might be. So this start point is where the data is coming into Alloy. So we are API based. Once an individual or a business has entered in their information, it's sent via API. So that can be from the front end, whether it's a partner like a DAO or it's something that's built in-house. The data is sent to us in this start note is what's going to kick off this overall journey. So what we'll do here first, there's one branch. We can go in here and we can select from prebuilt modules or you can customize from scratch if you're wanting to. But for this, let's just select here. So I'll go in and the first thing that I want to do for an individual is I want to run fraud and CIP.

(02:43):

So there are a couple of different iterations of this. I'll go ahead and select the one that has Lexis and now that first step is going to show up. So we'll actually jump into this into a minute so we can see into more detail. But what this is doing is showing us, okay, once that data is sent to Alloy, this is the first step and then these are going to be the outcomes that are associated with that first step. But we're also talking about small businesses. So what are we going to do for the business? If I go to this start node, then I have the ability to add in a new branch and we'll just call this businesses, and then we can do the same thing that we've just did. So this would be the UBS that are going through the fraud and KYC check.

(03:25):

And then for the business, we're going to select another prebuilt workflow, but this one is going to be business onboarding. So now we add that in. Another step is going to appear. So when the data is being sent to Alloy, it doesn't matter if there are multiple UBS or just a singular one, all of the data is sent together and then we're going to return a decision on that full application. So what we'll look at first is going to be the first step, which is the business onboarding component. So I'll jump into here, and this is where we can decide what's actually running at this particular step. What data services do you want to leverage to actually verify the business? What types of checks do you want to run? This one will be pretty straightforward. The data service is going to show up in green, and then any of the rules associated with it are going to be in gray.

(04:12):

We can come in here and make changes or updates by just going in and selecting the data service. And then these outcomes, those are going to be directly linked to any of these rules that we have created with the ability that we have or that we're giving our clients. They can come in here and they can just point and click utilizing whatever data services that they want to as necessary. Again, push that into production without any type of code changes or front end changes that need to happen. And that is something that can be in real time. It can be very basic changes. If we are wanting to get rid of any of these rules and we have that ability to do so, the outcomes here, we can click on 'em and it's going to show us what would lead to an auto approval or going into a manual review.

(04:55):

And what this is going to translate to in the overall architecture or blueprint for the onboarding of the business, we can see when you jump back, that's the approved or manual review. So I don't want to add anything else. After the business has been verified, we can say, great, as long as we've been able to verify the business and the rules have been met that they're verified, then we are going to say, okay, that's an approved outcome for manual review. You can still assign this as a manual review, so an individual can go in and review them or you can automatically step up to documents, whether that's articles of incorporation, it could also be collecting government issued IDs from the business owners as well. So for this, I'll just leave it manual review for right now and we'll look at the individuals and here the outcomes that are going to be associated with it are denied.

(05:45):

There's a step up required manual review and then an auto approval. So this is an out of the box configuration, but this is all going to be customizable. So let's take a look at the checks that are happening for the owners, and this is going to be a little bit more and a little bit heavier because there's going to be fraud checks associated with this. So in green, these are the data services still, and then the rules are going to be these gray nodes that are associated with them, and you'll notice that it's going to be built into a waterfall approach. So also leveraging Alloy, you have the ability to run only the data services that are necessary. So for a cost mitigation standpoint, check the fraud of the business owners first, as long as that is. Okay, then let's move on to the KYC and then ultimately the KYB if that's something that you want to do.

(06:31):

So the way that this has been built out is checking things like a denial list first and then also looking at the velocity, but then we can move on to more major fraud hits. So synthetic ID is going to show up here. We can look at SSN warnings for deceased, and all of those attributes that are going to show up or that we can create, those are all going to be editable here within the dashboard itself. Now, anytime that we're making changes or updates to them, you also will have the ability to test. And so that's really going to be the key thing is there is the UI where you can make these updates, you can test them, and then you also have the ability to push them to production without needing any type of change. So where I've seen this personally a lot is in the case of a fraud attack, if a client of ours is, let's say, not stepping up to document verification and they are under a fraud attack, then they can easily switch over, add in document verification, and then slowly reopen the funnel from there.

(07:27):

But they're able to then act, act or react in this case in real time to be able to stop those types of attacks. So scrolling over the last two data services that are showing up here are going to fulfill a ML as well as the KYC, and then we have the rules or the tags associated with them. And the idea here is when you're sending in all of this data, we're going to create one application. And so I'll just scroll out here so we can see, but we want to look at the entity holistically. So the last few I'll look at is going to be all of the entity's information in one page. So this will show you all of the checks that they've gone through historically. It will show you documents that have been uploaded, notes that have been added, or cases that might've been created. Cool. I'll scroll down. And that is right at time. Thank you everybody. Appreciate you for coming out today.

Brent Kasper (08:20):

Raise your hand if you've ever seen a fraudulent big statement before. Anyone. Let's go on up on the screen. Can anyone spot the fraud? It's okay if you can't. Most people can't. We've done this test with hundreds of fraud investigators around the world and they can't spot the fraud. Even when we tell them that the document that they're looking at is a hundred percent guaranteed to be fraudulent. The reason that they can't, it's not their fault. 90% of all fraud signals these days are undetectable to the human eye. It's only getting worse as technology continues to scale. So what does that actually mean for you? It means that you're making decisions about your customers, sometimes good decisions, but based off bad and untrusted data. Here at Inscribe, we're here to stop that. I'm Brent Kasper, VP of Sales and Success.

Conor Burke (09:07):

And I'm Conor Burke, Co-founder and CTO at Inscribe. And today we're going to show you how you can reduce risk and increase revenue with power of AI and using Inscribe.

Brent Kasper (09:16):

Let's take Jane Deer here for example. We want to make a decision about Jane. We want to maybe underwriter as a customer first. You have to get to know Jane. You can do that a couple different ways. If you're a business, you have plaid data, which we can integrate with or open banking data. You can gather information via documents via a collect portal, like a branded collect portal like this via web app via API, or even through Alloy, the integration partner that just spoke here a few minutes ago. Lots of ways to gather information about Jane, but what you also want to do is give her real time feedback on the correctness of the information that you're actually getting. Is that a recent bank statement, A recent pay stub? The worst thing that can happen is Jane submits their information, waits three days and then comes back and says, Hey, send us a better bank statement.

(09:57):

But in this case, Jane's done a great job. So we're going to go ahead and accept her documents and we're going to go ahead and submit them into Inscribe. And a couple of things are going to happen in real time. We're going to start getting information around the authenticity of those documents. We're going to see things like, is it the forensic trust score? So we're going to look and say, is the metadata the right metadata? Has it been manipulated by things like Photoshop? We're going to run through our network based detectors where we looked at millions of fraudulent documents to determine if this is the right version of a Wells Fargo statement that we would expect to go see. And when we do this, we're going to present to you a trust score, a score from zero to 100 that allows you to verify the authenticity of the information you just received about Jane.

Conor Burke (10:37):

And the trust score is just one of the ways Inscribe uses AI machine learning to catch fraud. Specifically, we use logistic progression using tree variables. Firstly, how familiar we are to document that you've submitted. Secondly, how severe the fraud results are. And thirdly, are there any anomalies or false positives likely in the result?

Brent Kasper (10:55):

So what does that actually mean? What that means is you can actually make programmatic decisions about your business using Inscribe. You're an SMB company, you've got maybe a high volume of loans, low value amounts. You might be able to set a trust score of 75 and say, we're going to automate anything above that because you want to move people through the process quite quickly. Maybe you're on the opposite side and you've got larger loans, a little bit lower volume, and you want to take a little bit more of a manual review process, either workflow supported with Inscribe because you have the context to look at where that information is actually coming from. So in the situation with Jane, maybe you see something that once warrants further review and you're going to go ahead and pull up that Wells Fargo statement and double click into what's actually going on here.

Conor Burke (11:38):

As you can see here, Inscribe is surface a number of issues, including have we seen a previous version of a document? If the document was created using a suspicious piece of software. If the document Wells Fargo document in particular differs in some way to what we'd expect and many more issues in particular with this document, we can actually recover the previous version. So in this case, you can see that the document we showed you earlier on actually had a previous version, which shows that, for example, the name and address was altered and many more details. And this, for example, is something that a human eye just cannot easily detect both Inscribe and the power of AI. You can surface this immediately and through a trust score based decisioning.

Brent Kasper (12:22):

Yeah, see these details matter because we know that time and time again, fraudsters use the same information over and over and over. So before we reject this, we're going to go ahead and add Jane to our block list so that she can never defraud us again. How do we know that people like Jane have tried over and over and over? We've had customers do a backtest of all their credit losses that they had last year, and we've seen in times over 10 different instances of the same person defrauding a company for well over six figures. They didn't find that out until they looked at Inscribed. And this is companies that have manual review processes, best in class systems to work through this. But the fact of the matter is the context and the pattern matching at scale is virtually impossible for humans to solve for.

(13:02):

And so they're at a loss to stop this from actually happening. And so now as you go ahead and reject Jane, you can actually now quickly move ahead and start spending more time with trustworthy customers. What do you actually do when you get a trustworthy customer? Well, once you trust that the information you've got is verified, you actually want to go ahead and make an underwriting decision. How much did you actually give this person? And then scenario, you see a bunch of approved docs, 84, a lot of green, and what you start doing is you want to parse out all that information, all that banking information, and start making credit decisions. Things like average daily balance, what's your expected income number of days of insufficient funds, gambling losses, suspicious deposits? All of this information should go in and play into your ability to use cashflow to actually make an underwriting decision. And we call that Credit Insights. This is the web app version of looking at this, but this is also all available via the raw data via API. So you can ingest this into your own proprietary model, so you can look at all the categorization, all the transactions, whether you got that from Plaid, it, max, or from these documents, it's all available to you. So you have one singular platform to ingest all of that, and it's all done via AI and in real time.

Conor Burke (14:10):

And what's even more powerful about using AI for credit insights and fraud detection is just how quickly it can adapt to new threats or new dynamics within our user base. At Inscribe, we've trained our models over the last half decade on millions of data points across our customer base, but we've also built a team of great, we're like the brightest machine learning engineers and data scientists to be able to build these risk insights for you so you can use this off the shelf and integrate 'em into your business. At the same time, we really pride ourselves on building secure and usable AI so you can be rest assured that your customer's data is safe, but you're also taking advantage of the latest technology to grow revenue and reduce risk.

Brent Kasper (14:51):

So to break that down, as Connor just mentioned, you're going to be able to stop fraud as it's occurring in real time, spend more time approving customers, and creating more equitable system by parsing out all that information and having one singular set of credit insights to look at. Long story short, you'll be able to reduce some of your losses, increase your revenue, and grow your customer base.

Conor Burke (15:10):

Thanks so much for your time today. If you have any questions, please visit Inscribe AI or catch us after the event.

Brent Kasper (15:16):

Cheers.