Sapiens Decision is an intuitive modern AI decisioning SaaS solution designed for automating complex decisions. Our visual modeling workbench is based on a proven methodology that covers decision modeling, testing, and governance. Auto-generation of code from models reduces time, cost, and defects. Sapiens Decision is technology agnostic with standards-based integration.
#AIDecisioning #MortgageTech #NoCode #Sapiens
#AIDecisioning #MortgageTech #NoCode #Sapiens
Category
🤖
TechTranscript
00:00Welcome to HousingWire's Demo Day on Demand. I'm Allison LaForgia, Managing Editor of HousingWire's Content Studio, and I'm glad you're joining us.
00:08This is where we spotlight the most innovative technology companies in housing and mortgage, giving you, our audience, a front row seat to real product demos from the teams building the tools that move our industry forward.
00:20In this session, we're featuring Sapiens Decision, a company that's solving industry challenges with smart, scalable, no-code solutions.
00:30Stick around after the demo. I'll be asking Raffi and Justin questions that help you connect the dots between their tech and your business needs.
00:41In this Sapiens demo, Raffi and Justin will be taking us through the AI decisioning platform from Sapiens Decision.
00:48Sapiens Decision is an intuitive, modern AI decisioning SaaS solution designed for automating complex decisions.
00:56Their visual modeling workbench is based on a proven methodology that covers decision modeling, testing, and governance.
01:05Auto-generation of code from models reduces time, cost, and defects.
01:11Sapiens Decision is technology agnostic with standards-based integration.
01:17Raffi, Justin, take us through the demo.
01:19Thanks, Allison. We're thrilled to be here.
01:22So, Justin, let's go to the next slide.
01:25We want to tell the community that Forrester just released their Q2 2025 AI Decisioning Platform Wave report.
01:35And in that report, they talk about enterprises compete on decisions, intelligence, flexibility, and speed.
01:43And they named Sapiens a leader with the Sapiens Decision platform that you're going to see today.
01:48They say that it's a good fit for customers that want to benefit from visionary AI decisioning big thinkers with all of the capabilities to handle the sophisticated use cases of highly regulated industries.
02:01Of course, we're talking about mortgage today, certainly highly regulated.
02:05We look forward to unpacking that today.
02:06So, if we go to the next slide, what you may be asking is, what is a decision?
02:12And when we say decision, we're really talking about those decisions that are a backbone of your mortgage business.
02:19You can think about AUS, clear to close, investor eligibility.
02:24Basically, if you can phrase it as a question and answer it with data, it's a decision.
02:29So, if we go to the next slide, often when we talk with mortgage lenders, we ask them where are your decisions managed today, and they give us three answers, either in system code that you can think of as a loan origination system or a point of sale, in spreadsheets, or in checklists.
02:48The bottom line, it's suboptimal.
02:50Why?
02:51Because with system code, it's slow and expensive change.
02:55You get inflexibility, where you really want flexibility, spreadsheets.
02:59We all run our businesses on Excel, but it's difficult to govern, difficult to audit, difficult to test.
03:04And finally, checklists, all those policies on cubicle walls, it requires a lot of rework and tribal knowledge.
03:12So, if we go to the next slide, the proposition and the solution that we're going to show you today is all about the AI decisioning unlock.
03:20So, what you're going to see Justin do is the bottom half of this slide, which is the AI decisioning cycle, life cycle, if you think about that compared to the legacy cycle, where the analyst is specifying requirements, it's going through a lot of iterations, it's finally handed it off to a vendor or a developer who then codes it, and then you get to engage with it.
03:42What you'll see today with Justin is he goes straight into the no-code configuration engine, he changes it, generates code, and the speed to market and the cost reduction is phenomenal, and you get net new automation that you otherwise can't do.
03:57So, last slide before we jump into the product, you may be asking, so how does this fit into the ecosystem?
04:04And this slide just shows you that you're going to have multiple mortgage use cases in the Sapiens Decision workbench, and those are going to be available as a service to all of your other systems.
04:16So, we're not talking rip and replace, we're talking about working within the pieces that you already have.
04:22So, let's go straight to the demo.
04:24Over to you, Justin.
04:26Thank you, Rafi.
04:28All right, so I am sharing now, Decision Manager.
04:31And before I start, I always like to share a little bit of my background.
04:34So, I came from mortgage.
04:36I was part of a small mortgage company where I actually managed the decision modeling practice for that organization.
04:42So, I've seen firsthand the benefits we can gain by documenting your decision models in production, and some of what I'll be showing you today are things that my team did to help smooth out the origination process.
04:54So, this is Decision Manager.
04:56It is the authoring tool.
04:57I should say that your underwriters, your closers, they're not working within this tool.
05:01They're working in the LOS, they're working in their normal applications, and those applications are making a call to the logic that is deployed based on the models that we build here in the authoring tool.
05:11I like to start in the repository because, as Rafi described, typically your logic is spread out.
05:16It's in the LOS, it's in Excel, it's pinned to cubicle walls, and what we're advocating for is a better way.
05:22We can store your logic in a repository.
05:23Here is an example that I've set up specifically for mortgage as part of our mortgage accelerator, and really what we're trying to showcase is we know there exists logic throughout the whole life cycle of mortgage.
05:35Here are some ideas to kind of get you started on where that logic could be.
05:38For our use case today, we're going to take a look at a portion of an AUS model, a flow, if you will, that is evaluating a DSCR loan.
05:50So, loans that don't qualify based on borrower income, but rather based on the rental income of the property.
05:55So, Justin, we're in the non-agency world at this point.
05:59In the non-agency world.
06:00And it's been interesting because there isn't a DU or LP for these type of loans.
06:06So, we're advocating that because that whole process is based on rules, we can manage rules.
06:12We can manage rules better than the LOS.
06:13What happens if we build an AUS managing the rules in decision?
06:18Cool.
06:19So, when I look at this, each of these boxes, and we'll take a look at one, is a decision model.
06:23They're all here to support the question.
06:25In this case, what are the eligible products, my DSCR levels, that my client's eligible for?
06:31But to answer that question, I have a series of sequenced questions I need to answer first.
06:36What is the qualifying credit score based on all the borrowers and all their credit scores?
06:40What's the ratio code or the DSCR ratio based on the income and the payments?
06:45Do I pass the general knockout questions?
06:48Which products am I eligible for based on LTV, based on first-time investor guidelines?
06:53Do I pass any state overlay logic?
06:55What am I eligible for based on reserves?
06:58And given all of that data, we're able to answer this complicated question of, therefore, what should I offer my client?
07:06And if I drill into a model, this is a simple example of a decision model.
07:10They're all built to answer a business question.
07:12This question is that general knockout.
07:15Is my loan eligible based on the general threshold guidelines?
07:18I've decomposed that into three supporting questions, which is part of our methodology.
07:23Take the bigger question, break it up, and you'll see as I click on the highlighted or the underlying conditions, that's telling me it's derived within the model.
07:32So my occupancy eligibility is a standalone question that's based on the occupancy type.
07:38This is about as simple as what we call a rule family gets.
07:41You can consider it a decision table.
07:44I've got one condition, one conclusion.
07:47Investment properties are eligible since these are DSCR loans.
07:51Everything else is ineligible.
07:52And then we can produce very sophisticated messaging.
07:55So here I'm expanding our testing panel.
07:59So if I were to pass in a owner-occupied loan, I would see the rule row that I hit, and I would see the relevant messaging.
08:08Now, that messaging can be tailored for, say, the borrower.
08:11So it's helpful, but not as detailed as it could be.
08:15And maybe we save the more detailed message for that underwriter.
08:18And as a client, what I found is this ends up being a very great training tool because we're able to tell the underwriter exactly why this loan is not qualified and the specific rule and rationale.
08:31All right.
08:32And then just one more example.
08:33If that's one of the easiest rule families, this is about as complicated as they would ever get.
08:39I'm evaluating my DSCR ratio and my credit score against my loan amount to determine is my loan amount eligible, does it fall within the respective guidelines?
08:50That's great, Justin.
08:51Justin, so just to recap then for the viewer who may be less familiar, what you're showing here is a design time environment where you have set up a decision flow that is providing the eligibility, yes or no, or maybe, depending on the output, based on what you feed the model that you will deploy and expose as a service.
09:15And what you're able to do here is really set that up based on the policy, and you're not having to write it in a complicated document and throw it over the wall to a vendor or a developer.
09:27You're just setting that up right here.
09:29You're able to test it, put in the messages you want.
09:32And so I just wanted to set that as you move into the next phase.
09:36Thank you, Rafi.
09:37And as you mentioned, testing is one of my favorite bits.
09:40But if the data in the LOS has my loan amount, my loan purpose, my LTV, the occupancy type, so on and so forth, when I run this test, this is what I would get back from a production standpoint.
09:51There's my qualifying credit score.
09:53I took my DSCR ratio, called it a silver because it falls in the middle of the guidelines.
09:58And then I evaluate my eligibility criteria and my good, better, best of my product list to come back with this is the eligible product in this scenario.
10:07Okay.
10:09All right.
10:10So now let's pretend we're the analyst.
10:12And I've deployed this logic, but I want to monitor how are my loans actually resolving through that logic.
10:19So I'm going to leave Decision Manager for a moment, and I'm going to jump over to our Decision Analytics dashboard.
10:24So once that logic is in production as code, every time it is being interacted with, the responses plus all the data are being logged and available for me to run analytics on.
10:36So I can come over here to my Decision Conclusions and say, let's take a look at that flow, my DSCR eligibility flow.
10:43And specifically, I want to see how it is resolving my final step there, what are the products that I'm telling my clients are eligible for.
10:51And in this case, I can see, all right, I've got a distribution of eligible products, but unfortunately, I've got a big uptick in ineligible loans.
11:00So my next question is, what is driving that situation?
11:05Well, I have available to me all the other decisions.
11:07So maybe I start at the top and say, is it my first-time investor product eligibility guidelines?
11:14No, that's pretty flat.
11:15How about the general knockout questions?
11:18Okay, here's something that is actually creeping up.
11:20Let's dig more into that.
11:22At this point, I might come over to my Dan AI, which allows me to write natural language queries.
11:28We've got a handful of them structured already, but this is specifically an example of what I did as a client, where I may say, okay, let's take a look at the general threshold logic.
11:39And let's say maybe in the last 60 days.
11:42And show me all of the messaging that I've been receiving as it relates to that particular decision.
11:47And what I see is all around loan amounts.
11:50Maybe I want to display the category, since we saw there were multiple categories in that particular decision model.
11:59So this is cool.
12:00And just like you showed us in the prior module, where you're a non-technical analyst setting up the logic, you go through a governance lifecycle, deployed it.
12:08It's now running in production, and you're able to query those results, again, as a non-technical person.
12:14You can just write it in your Dan AI feature and get back the results that you're interested in.
12:20Exactly.
12:20And now I know that what's driving the ineligible logic or ineligible conclusions has to deal with my knockout questions that I call general eligibility.
12:30And for the most part, I see the loan amount is driving that.
12:34So maybe I want to increase my eligible loan amounts.
12:38And then the finding messages, which are more tailored to the actual loans coming in, as opposed to the informational messages, which are more generic.
12:44Now I get examples of the type of messaging that's coming back from the individual loans.
12:48All right, so now I'm going to tell my product owners, let's take a look at these guidelines, data analytics team, see what you can come up with.
12:57But in the meantime, I'm going to jump over here to the transactions.
13:01And what this is, this is all of the actual production loans that have gone through the process and all of the transactions logged here.
13:09And what I want to do is make this my A-B testing.
13:12I want to take these production transactions and use them to see what they would look like if I make a change to my model.
13:18So I want to send this to the decision manager as new test cases.
13:23But first, what I think I'll do is I'm going to filter out.
13:26Really, all I want to do is take a look at the test cases or what is the production data where the values are ineligible for my general threshold eligibility.
13:37So I'll filter my tests just to those.
13:40And now I'm going to send that set of tests back to decision.
13:44And maybe we'll call this my demo test group.
13:48We'll send it back to my DSCR eligibility flow.
13:52Version 1.1 of the model, which was what I'm working in, the approved version as opposed to potentially a draft.
13:58And maybe I'll just send over 50 of the tests, including the actual answers that I got.
14:05So what's happening now is decision analytics, what we call Dan, is collecting that data, formatting it so I can import it into decision manager,
14:14and doing it all for me and sending it back to that particular asset, which is the DSCR eligibility flow,
14:20and making that group of tests available for download.
14:24I don't need that at the moment.
14:26I can come back to my flow, back to my test groups.
14:30Here is that test group.
14:32You know what I'm going to do?
14:32I'm going to mark that as a persistent set of tests because I want to use that in any versions going forward.
14:38I can open up that test group.
14:40And I can execute, in this case, all 50 of those tests.
14:46Now, I could have exported all of them.
14:49And I can take a look, and it makes sense from an aggregate perspective.
14:53All 50 of those tests are ineligible because that's what I filtered my results on.
14:59So I'm going to duplicate that tab and just set it aside here for a moment because I want to do a comparison in a moment.
15:06So now, good news.
15:10Our product team has come back, and they said, here's how we want to change the guidelines.
15:14They're making changes to loan amounts.
15:16They're making changes to FICO scores.
15:18But it's all here in this big text document.
15:21So as the analyst, I could certainly read through this and build it.
15:24But what if I could just have the AI do that for me?
15:27So I'm going to copy that, and I'm going to make a change to this model.
15:32I'm going to put it in a task.
15:34The task is what the container is that sends the model through the governance process.
15:40Because of how I decompose my logic, I know that it is the general threshold eligibility model I want to modify.
15:47It's saying, do you want to link it to the flow?
15:49I do, actually.
15:52And now I'm going to invoke a chat session where I can paste in my text requirements and ask it to build what will ultimately be the replacement for this particular rule family.
16:03Oh, that's cool.
16:05So you brought in production data, did a simulation, got the updated requirements, and now you've essentially used a co-pilot to create that decision model structure out of that natural language for you.
16:19Exactly.
16:20I can open it up, take a look.
16:24I'm pretty happy.
16:24I've got my eligibility criteria and my ineligibility criteria, all my conditions.
16:29It did use different what we call fact types.
16:32Those are the conditions that go into the model because it's pulling them from the requirements.
16:36But we've got the ability to sync our glossary.
16:38So I'm going to quickly replace the conditions it's used with the ones that I want it to use.
16:46So rather than the credit score, I want to use the qualifying credit score.
16:53Rather than the DSCR maximum loan amount eligibility, I use one that's got an indicator on it.
16:59Not a problem.
17:02Loan amount's slightly different.
17:03And what's nice here is we design a glossary for reuse, and I am reusing the elements from that glossary using the type ahead to find the values quickly.
17:18All right.
17:19So now I've taken their values and said these are the ones I'd rather use.
17:22So what you're showing there is you're not polluting your data structure with additional terms just because you – but it's all giving you what's already there.
17:31Exactly.
17:33At this point, I can remove this particular rule family, drag over the new one, which will adopt the new updated terms as it comes over, close the chat window, and now I see that everything is connected.
17:50I can run my validations.
17:52Again, unique to the decision product.
17:54We want to make sure that your logic, once in production, has integrity, is complete, has no overlaps or gaps in the logic.
18:01That's cool, too, because you think about you're using Gen AI there in the copilot, but now you've got the human-in-the-loop validation to make sure that you're not injecting anything hallucinatory.
18:11Right.
18:11I don't want to deploy a strictly AI-built model.
18:14I want to make sure that it still adheres to our better practices and your actual guidelines.
18:20Perfect.
18:20I can go back to the flow, revalidate my flow.
18:26I made those test cases persistent, so I'm going to go back to my test group.
18:30Here, again, is my set of demo tests.
18:34Since I've made a change to the model, it always takes a little extra second when I execute the test because it's compiling the logic.
18:41Hopefully here only just a minute.
18:42So this is emulating what it'll be like in production because the engine here at the test time is the same as the runtime.
18:50Exactly.
18:51We like to say that the model is the code, so I'm testing the model.
18:55Effectively, I'm testing what is ultimately the production code.
18:58There we are.
19:02And then once it's done, I can visualize the test group.
19:06And, again, I already see that I've got far more eligible.
19:09But if I needed a reminder, I could go back and compare my results to the former results, full A-B testing.
19:17So I've taken my use case, monitored it in production, used some analytics to figure out what I should change, taken that textual requirements through the help of our co-pilot AI, made the change, and then compared it to what the actual logic would look like had this been the logic I deployed to production.
19:34Amazing.
19:35So that slide that I showed, the AI decisioning unlock, you just demoed that in a couple of minutes, a pretty substantial change relating to I want to increase eligibility of this product for my borrowers.
19:49And you did all that analysis, used an AI co-pilot, you governed it, you retested it, and you can see the difference right away.
19:57You're ready to deploy it and go.
19:59Exactly.
20:00Nice.
20:02Okay.
20:03That's a wrap, right?
20:05I think so.
20:06Okay.
20:06Well done, Justin.
20:07You're a rock star.
20:09What an interesting technology platform.
20:12Thank you so much for taking us through AI decisioning from Sapien's decision and what you're bringing to the table and how the no-code technology can help better manage the cost per loan, reduce cycle times, and improve quality.
20:28Let's dig a little bit deeper into some questions.
20:31That sounds great.
20:31Thanks, Allison.
20:32What are the top use cases for your software in the mortgage industry?
20:38Oh, I love that question.
20:40Let's make it real, right?
20:41So I'm going to throw that over to Justin.
20:43He was on the floor using this technology in an IMB.
20:47So, Justin, what comes to mind?
20:48There's so many, so it is hard to choose.
20:50But what's, like, top of mind for you from a use case perspective?
20:54So certainly low-hanging fruit is agency overlays, jumbo or non-conforming guidelines.
21:00Anything that's in the matrix, you're having to figure it out every day.
21:02It's repeated.
21:04It's versioned.
21:05You need a way to manage and test that logic.
21:07But where you really start to see success is where the logic lives that's not normally managed in the LOS or in the PPE.
21:15It could be QC audits.
21:17It could be condo classification codes.
21:21It could be evaluating, is this loan ready to be disclosed?
21:24Have I made sure it has the appropriate amount of mortgage insurance?
21:28These type of questions can be costly, and it's not typically something that your LOS is built to monitor for.
21:34That's great.
21:36Let's talk a little bit more.
21:38You just mentioned an LOS.
21:40So do you build software like a custom LOS or integrate with existing systems?
21:47Yeah, that's an important consideration.
21:49Thanks for that question.
21:51So definitely we integrate with existing systems.
21:54So we are a purpose-built AI decisioning platform and focused on what we do well, which is managing decisions.
22:02And also we have designed it to plug and play and live as a good partner within the ecosystem.
22:09So we're not looking to rip and replace LOSs and POSs and PPEs.
22:13We're looking to put decision logic that is suboptimally managed in those places.
22:20And as Justin just described, net new automation for logic and decisions that aren't handled.
22:27And then just provide those as a service.
22:29So it's just another third-party call from those existing systems.
22:34Now, Rafi, Justin, this is pretty cool.
22:38I mean, in real time, you showed the capability to make some pretty significant changes.
22:43But I have to ask, who are the users of the system?
22:48Is it do I have to have a whole IT team?
22:51Is it loan officers, business analysts?
22:54Yeah, no, that's a really important consideration.
22:57How do you make it real?
22:58How do you make it operational?
23:00And I'll throw this one to Justin as well because he's lived it in terms of like who the user is and also to disambiguate who the user isn't.
23:09So, Justin, maybe you can field that question.
23:11Sure. So from my experience, mine was a team of three, and two of them came from the floor.
23:18They were team leads in a particular area, and in that area, there was a challenge.
23:22So they built the models to solve the challenge for what was their former team.
23:26And they just stayed on my team and started to do that everywhere throughout the organization.
23:29And I'll just mention, Justin, it's not that you three were net new.
23:34You guys were already there working.
23:36So it was just like here's a tool to do your work better.
23:39I think that's an important consideration for people thinking about this.
23:44Right. Yeah.
23:45And then it's also interesting that a lot of times IT necessarily isn't interested in managing changes to these types of rules.
23:53So the folks that were on my team were better versed in what the business was.
23:57Now we've got the ability to document those rules, test those rules, validate those rules, and IT manage putting them into production.
24:03And that's awesome.
24:05And on the question of loan officers, they're not interacting with the design time where you were demoing today.
24:12They're in their standard workflow.
24:14It's the systems that they're interacting with already are interacting with what you put out there.
24:18So no change to their process per se.
24:21And I have to just one more question before we wrap up.
24:26I have to ask again.
24:27And this is no code.
24:29So there's no point after what you showed me where there's some back end where I'm going to have to code to implement this because it looks pretty user friendly.
24:38A hundred percent.
24:39So what Justin showed you can then be sent for review, and the person can test and review that as they want, and you can set that up to wherever you want.
24:49And then you just deploy.
24:51And that process where we have to wait a second for it to generate the code to test, it does the same thing to the runtime engine.
24:58So there's zero translation, true no code.
25:01Well, Rafi, Justin, thank you for joining me today.
25:05To our audience, for more information about AI decisioning from Sapiens Decision, click the link below.
25:12Thanks, Allison.
25:13It was a pleasure.
25:14Thank you both.
Comments