- 3 months ago
Autopilot Analytics Agent
The Blend Autopilot Analytics Agent is an AI-powered data assistant purpose-built for mortgage lenders. It converts plain-English queries into instant insights, self-building dashboards, and proactive metric alerts to empower originators and executives to optimize pull-through rates and streamline loan operations without waiting on data analyst backlogs.
Blend
See how the Autopilot Analytics Agent eliminates mortgage data bottlenecks. This demo shows lenders how to ask questions in plain English—like tracking lock volumes or pull-through rates—and get instant answers without waiting on analyst queues. Watch interactive dashboards build themselves in seconds, see proactive KPI alerts in action, and discover how role-based permissions deliver tailored regional insights instantly. Accelerate your lending operations with on-demand business intelligence.
The Blend Autopilot Analytics Agent is an AI-powered data assistant purpose-built for mortgage lenders. It converts plain-English queries into instant insights, self-building dashboards, and proactive metric alerts to empower originators and executives to optimize pull-through rates and streamline loan operations without waiting on data analyst backlogs.
Blend
See how the Autopilot Analytics Agent eliminates mortgage data bottlenecks. This demo shows lenders how to ask questions in plain English—like tracking lock volumes or pull-through rates—and get instant answers without waiting on analyst queues. Watch interactive dashboards build themselves in seconds, see proactive KPI alerts in action, and discover how role-based permissions deliver tailored regional insights instantly. Accelerate your lending operations with on-demand business intelligence.
Category
🤖
TechTranscript
00:05Welcome to HousingWire Demo Day On Demand. I'm Allison LaForgia and HousingWire's Demo Day
00:10spotlight the most innovative technology companies in mortgage, giving our audience
00:15a front row seat to real product demos from teams building the technology that moves our
00:20industry forward. In this session, we're featuring the Blend Intelligent Analytics
00:25Agent, which eliminates mortgage and data bottlenecks. This demo shows lenders how to
00:30ask questions in plain English, like tracking lock volumes or pull-through rates, and get instant
00:36answers without waiting on analyst cues. Watch interactive dashboards build themselves in
00:41seconds, see proactive KPI alerts in action, and discover how role-based permissions delivered
00:47tailored regional insights instantly. Accelerate your lending operations with on-demand business
00:53intelligence. Stick around after the demo, and I'll dive into a short conversation where I'll
00:58ask the questions that help you connect the dots between Blend Intelligent Analytics and your business
01:04needs. In this Blend Demo, Mukesh will be taking us through Blend Intelligent Analytics. Mukesh,
01:11the floor is yours. Awesome. Thanks a lot, Allison. All right, so maybe I'll just quickly lay out the
01:17problem that we're trying to solve here. Lenders already have a lot of data, but most business
01:23questions still get stuck in a reporting queue. Imagine a branch manager wants to know why a pull-through
01:29is down or why one channel is underperforming, and the answer can take days or weeks to come back.
01:36Blend Intelligent Analytics changes that workflow by letting business users ask questions in plain English
01:42and get answers immediately right inside Blend. This is not another disconnected analytics tool.
01:50Admins can enable Intelligent Analytics directly in Blend, grant access to the right users, and launch it
01:57from the reporting section. That means that the experience is close to where the loan team already works.
02:03Let me show you how it works. So now on the screen, I'm just showing you a dummy test account
02:08where this is the,
02:09we call it the L-O-T-K, the Loan Officer Toolkit. So this is the platform that an admin
02:15user is going to
02:16use to self-serve Blend Analytics platform to their users. They will just go into the setup screen that
02:22they have, and over there, they'll have Intelligent Analytics available to them. We have some TNCs that
02:28they have to agree to before we will enable the platform to them. Once they accept the terms,
02:33now they can enable Intelligent Analytics. So now the platform is enabled in their environment,
02:39and they can specify the exact users that they want the platform enabled for. So let's say we want to
02:44enable it for Alex, and they invite the user, and that's it. Now that user has access to the new
02:52Intelligent Analytics platform that Blend has launched, and the way they can access the platform
02:56is as simple as they go to the reporting tab, and from there, they can just open, hit this button,
03:01and this will open the platform for them. So we like to start from a familiar operating view.
03:07So here you can see some core performance metrics like submission, conversion, NPS. So the starting
03:14point is not a blank prompt box. It's a business dashboard that gives you immediate orientation,
03:19and from there, you can ask follow-up questions naturally. Let me show you how that works.
03:25So let's say overall business view, let's say I'm a business leader at a lender organization.
03:29I will go right into that, and now this presents the entire view of the business to me. It actually
03:35tells me exactly how to understand it. It's giving me a consolidated view of mortgage as well as home
03:41equity business right in this one dashboard. It tells me the weekly application volume, the trending up
03:47or trending down, where the business is coming from, how many are originating in Blend, and how many
03:54the lender is pushing to Blend, all that information, bunch of stuff. Let's say today I'm
04:00mainly interested to understand where are we losing the borrowers, and maybe I'm not so interested in
04:05equity. Home equities, I'm more interested in the mortgage one, because we are spending a lot of money
04:10on marketing campaigns. So maybe this is the one. In Blend, it comes as public API. So mortgage loans,
04:17public API are the ones that are basically leads that, you know, I'm spending money to acquire
04:23borrowers. And it shows me that this green section that you see here says never opened shell. What it
04:31means is these were borrowers that showed intent in the marketing campaign that, you know, my marketing
04:37team ran. But the borrower never actually engaged and applied for the loan in Blend. So if I look at
04:42the
04:42number, the number is actually significantly high, it's like over a million. So I'm like, okay, how do I
04:48understand what's happening here? So I can literally just click into that and say, I want to drill down
04:53into this public API, you know, data set. And so now the platform starts running on its own, it
05:00understands that I'm more interested in that section. And it actually now, if it needs some more
05:06clarification from me, it's going to ask me all those questions. And then as soon as I give them the
05:10answer, it'll start investigating on my behalf. And in all of this, the thing that we should note is
05:17like, I never had to sort of write any SQL query, I never had to understand the underlying data
05:23structure, the tables and all of that, because all of that is already taken care of for me.
05:27Blend has already trained the AI to understand all of that. And it's just using my natural plain
05:34English language queries, and its own understanding of the data structure to give me the answers that
05:39are relevant for me as a business leader to understand, and then take action. So right
05:44now, I'm only showing the section where, you know, I as a business leader, I'm able to understand
05:49what I'm able to first see what's happening, then I'm able to drill down to understand exactly
05:53what's happening. At this point, this may take a few seconds. So obviously, we'll cut this section
06:01and we'll show them when the result comes. And so now I know the total number of loans where the
06:09borrowers did not engage with it. And I know for a fact that whenever a loan gets created, we assign
06:14it to a loan officer. So I'm like, if my loan officer was assigned to the loan, and the borrower
06:20never opened the loan and planned and never engaged with it, like, I want to find out how many loan
06:26officers were assigned to those loans. And I don't want to focus on loans that were, let's say, more
06:31than 15 days old. I only want to focus on loans that were sort of created in the last 15
06:37days, but
06:37not acted by the borrowers. So now I'm going to ask the follow-up question to the platform.
06:43Can you find out the number of loans where they are less than 15 days old, where it was created,
06:52but not started by the borrower? And there is a loan officer that is already assigned to that loan.
07:03And as you can see, I actually did not type the question. The platform does allow an ability to
07:08literally speak with the data. And in a way that it actually automatically curates my prompt itself,
07:16and then sort of goes and does the digging for me. So as you can see that the platform was
07:24able to
07:25tell me that there are about 15,000, you know, loans where, you know, the mortgage loans have been
07:32assigned to loan officers. So now I know that there are so many loans that have been started, have been
07:39created, but not started that are not too dated. So now it's, it's an obvious next step is if I
07:45know
07:45exactly what those loans are and what the loan officer, who, who, who are the loan officers that
07:50these loans are assigned to, I can just pass all the data to my loan officers and they can literally
07:55pick up the phone and engage with the borrower and help the borrower apply for the loan, right? So
07:59that's, that's how this platform is not just creating visibility into a problem. It's all actually
08:05letting me drill down to understand exactly, you know, what, what's behind the problem and how we can
08:12act upon it to take a meaningful business step. But this, now let me, let me dive into like, okay,
08:21I was
08:22able to do this analysis today by chatting with the platform, but let's say I now want to solve this
08:28problem once and for all. So what am I going to do next? So I, the platform does allow an
08:34ability where I
08:34can create an agent, where an agent is going to monitor this data on my behalf. Uh, and whenever,
08:41uh, you know, it'll run, it can run at a certain frequency and send all that information over to
08:47me and then let me decide what I want to do with it. So let me show you that experience.
09:02So, uh, I'm just about to start create an agent. I can, uh, you know, I can, you know, there
09:09are
09:09different ways in which I can create an agent. Let me show you the visual way of doing it and
09:14the
09:14visual way of doing it. Uh, the platform allows me, you know, different steps where I can sort of
09:19literally drag and drop, or I can just tell AI exactly what I wanted to do. But even before I
09:25tell AI something based on how I've used the platform, it actually automatically suggests
09:29the kind of agents that can create, it can create for me and that can do the job, uh, you
09:34know, so I
09:35don't have to come to the platform every single time, but I'm just going to paste the, the same
09:39context that I showed just now where, you know, uh, there were a lot of, uh, marketing campaign
09:45leads that were generated where borrower did share an interest that they were interested,
09:49but did not really follow through. So I'm just going to create an agent, which, which will
09:54run every Monday morning at 10 AM. And it will basically identify all the loans where the loans
09:59were created, but not started and send all of them over to me over email. Uh, and as soon
10:05as I do that, um, again, just through a prompt, the agent got built right away. Um, the schedule
10:22is set. It knows exactly how to fetch the data. Um, you know, how to summarize it and what artifact
10:28to build. So if you go in, like, it'll say weekly public API, unstarted loan report, different
10:33sections, and then it'll email it to me. So I'm just going to configure my email here.
10:39Sorry. Um, yeah, got done and I'm going to just publish it. And the moment I publish it,
10:44this agent is now created. It's already gone. It's going to run every single, uh, every Monday
10:49and send, send that information over to me. So this is just one use case where we were able to
10:54understand what's not working well, uh, in my business, uh, diagnose it a little bit, and then,
11:01uh, have automation in place. That's going to ensure that whenever the business is sort of going
11:07south, that automation in place, that's going to notify me. And from there, I can then decide
11:11what I want to do with all, all those loans. Um, let me, um, let me now jump into, uh,
11:17some other
11:18experiences that this platform offers. So I already showed, uh, a dashboard. So, you know, if there are
11:24dashboards that are already built, how a business leader could engage with it, but the platform offers a
11:29lot more, I can literally just go and ask some questions to the platform and it'll actually
11:33provide answers to me. So let me show you that experience.
11:41A, when I land on the platform, it already knows a bunch of things about me and the data it
11:46has,
11:46and it tells me exactly, are you interested in any of these? So I can literally just click on them
11:51and it'll, it'll let me start, uh, you know, it'll start, uh, finding data for me, but I can also
11:56just,
11:57just ask whatever I wanted to do. So I'm saying, what's my start to submit conversion rate this
12:02month and do this analysis only for mortgage loans. And now it'll go and start looking for
12:08that data that I just asked for and come back to me with the results. And again, in the result,
12:14it's not just giving me, uh, a result set. It also sort of gives me a little bit of detail
12:19as well
12:19about, you know, what it found. Um, and not just a SQL query that I'm, I'll run somewhere else.
12:28All right. So as you can see the number, uh, the result from the platform, the platform is
12:32actually able to give me the exact conversion number. I did not ask, uh, what the prior conversion
12:37was, but it already gave me that answer. So it's telling me what's the month till date. It's not
12:42saying it's the whole of July and how many started. I didn't, didn't ask that question. So the platform
12:47also sort of understands the broader intent than the exact question that you asked, because it sort of
12:52knows that if they give this exact answer, what's the next follow-up that the user is going to ask
12:57for. So, and actually it also broke down like which channel is performing how, although I didn't ask
13:04for that thing. So it tries to serve me with as much data as it thinks is necessary without me
13:10necessarily asking for it. So I was actually going to say like, can you also compare this metric with
13:15the previous month, but it has already done that for me. So it's saying, okay, this month, these are the
13:20three different channels. Whereas if we compare with the previous month, um, you know, how it fares. Um,
13:30yeah. So, so basically in the conversational part, again, like it does not lose context. If I just ask any
13:35additional questions, like, can you just break it down by loan purpose? Because, you know, maybe, uh, there
13:40are some loans that are for mortgage, some for refinance and stuff like that. And if I just ask follow
13:44-up, it
13:45has context and maintains that context when it gives me the result. Um, so maybe, uh, from here,
13:53maybe we just, um, uh, go to the next section, which is basically, uh, we built significant trust
14:00and control in the platform. So if you ask the platform a question that it probably does not
14:05know enough about, or probably does not have the necessary data access, or it has not been trained
14:11on, for example, if I just say, how was my business yesterday? And this is a very ambiguous
14:17question. The platform knows exactly how to answer that question as well. It'll, it'll basically ask
14:21clarifying question. Like, what do you mean by business? Are you asking for all the loans that
14:25were started or submitted or funded? It actually has those bells and whistles in place. That's one.
14:31Second is we've built enough controls in place. So if lender A is asking a question, they will
14:37always only see data that is relevant for them. They'll never be able to see any
14:41other lenders data, because we want to make sure that, you know, that's the most important thing,
14:45at least in the mortgage industry that our lenders are able to see only see and access only their data
14:50and not other lenders data. Although our platform does allow sort of an internal benchmarking as well
14:57as external benchmarking as well. So we have built, uh, you know, uh, stuff in the platform where,
15:02where a lender can say, all right, can you tell me what is my conversion and how do I compare
15:06it with the rest of the lenders of blend? And so we have those, uh, functionalities built in as well.
15:11Similarly, if a lender says, uh, can you do analysis of my branch X and performance of my
15:17branch X and compare them with performance of branch Y, or how does my loan officer,
15:23you know, uh, you know, Alex, Sam compare, perform compared to, you know, someone else, uh, in terms of the,
15:29the number of loans that they originate each month or the NPS of loan officer X versus NPS of loan
15:36officer Y. So we built all those, all those things in the platform, um, you know, uh, with, uh, we're
15:42trying to be very, very careful in terms of, uh, like I said, putting the necessary guardrails in
15:47place so that the platform never hallucinates and never, and always shows data of lender A to lender A
15:54and not to anyone else. That's the end of it. Let's jump into some questions. First, leadership
16:04teams frequently need ad hoc reports on things like pull through rates or cycle times, but a common
16:10problem is that business intelligence teams face massive backlogs. How does the intelligent analytics
16:16agent change that workflow for a non-technical manager? Yeah, that's a great question. And that's
16:22the exact problem that, uh, we kept running into as well, uh, you know, over the years, uh, where,
16:29uh, you know, we had to maintain teams of people who had deep understanding of the data and the data,
16:34data structure, uh, and, and also business in order to answer all these nuanced questions that kept coming
16:41from all our lender community. And so with the new platform that we have, uh, adopted, uh, what we've done
16:47is we've trained the platform enough that it understands a, the entire data underlying data and
16:52data structures. And it also understands a lot about our business. And so with that, we don't
16:58necessarily need technical people who are going to write SQL queries to find answers to a business
17:04question, to a nuanced business question. I should say the platform is there in the hands of the user
17:08and they are able to ask very deep nuanced question and the platform is responding back to them. There are
17:14times where there are some, some questions that the user may ask, which the platform has not yet been
17:19trained on and the platform keeps asking clarifying question before it'll actually respond back to
17:24the user. And, uh, it also passes the platform also passes that, uh, context back to our admin
17:31user saying, Hey, these are the kinds of questions where I'm not trained sufficiently on. So if you train
17:36me sufficiently, then I will make sure that I actually respond back, uh, quickly back to my business
17:41user. So that's how we believe, uh, we, we feel very confident that we will serve our lender community
17:49in a way where no matter how nuanced question they have, they won't have to go to a technical user
17:54to
17:54ask a business, to write a SQL query, to get answers to their questions. And most analytics platforms
18:01require you to log in and look for problems after they've already happened. How does the system help you
18:06stay ahead of operational issues like a sudden drop and lock volume? Yeah, that's a, that's a great
18:12question. And basically this platform allows us to set up agents. So if, if a lender knows that
18:18these are the kinds of things that they are always sort of, uh, you know, uh, careful about, they don't
18:24want the pull through rate to go down. They don't want the submission rates to go down. They don't want
18:28the funding rates to go down. They can set up agents that will basically constantly monitor those KPIs for
18:34the lenders. And when of those KPIs, the values of those KPIs go above or below a certain threshold, it
18:40can
18:40alert the, uh, alert the relevant users. It can alert via email or a Slack integration or a Teams
18:46integration. So people don't have to come to the platforms asking, hey, how are things? Because
18:51essentially the agent will go to them saying something is good or something is not good or, you know, something
18:56is
18:57doing really well. And keeping financial data secure is non-negotiable. What data safeguards
19:04are running in the background when the agent processes borrower files? That's a great question
19:09as well. So essentially what this platform does is the actual data is not directly exposed
19:15to the agent, um, at all. Agent basically understands the underlying data structure and the,
19:21the question that business user is asking. It then trans it based on its understanding of the data
19:27structure and the context of the business problem and the ask that a user has. It then constructs
19:33a SQL query and the SQL query is then run on the warehouse to pull data from the warehouse and
19:38then
19:38pass it back to the user. So we don't expose the actual data directly to the agent. We basically
19:44leverage agent to construct the query and fetch the data for the user.
19:51So I will now find this complaint where the patient will reinstall it to the user.
19:53where the patient delivers a lot of information,
19:53at least in relationship with anacey andFi.
19:54So I should ask that how to carry that because