Kyle Hoback: Let me give you a quick overview of what Ilana can do and what’s included. So she’s an insurance underwriter, like Raj said, kind of a digital form of one, but insurance underwriter nonetheless. And her skill right now is really focused on new application submissions. So this is around commercial, general liability, property, and umbrella policies. She’s monitoring inboxes ingesting, classifying emails, extracting key fields, and really trying to understand what’s coming through the documentation and then working with underlying systems, existing systems, to be able to update them so that submissions get created.
So as far as documents and data, this includes things like ACORD 121, 126, 131, and 140, loss run reports, statement of values (SOVs), and broker carrier application forms. There’s a nice workflow here that kind of steps through the different pieces here. But kind of tied to that, I’ll focus on the integrations here on the lower right first. So the integrations allow all the systems to get updated that are part of the process today: Guidewire, Bold Penguin, Duck Creek, Lexis Nexis Risk Solutions, PEGA, and others. So if there’s bespoke systems or ones that aren’t included here, at its core Ilana is really an integration software as well. So being able to work with these tools at as well as additional ones is really important for her to make sure that she’s able to update all of these things.
And of course, because it’s digital, in the upper right here you can see the analytics that are available, which we’ll get deeper into, being able to track these submissions from when they’re received all the way through until they’re submitted, what’s happening, what’s included within them, where there might be issues and how they’re processing, where they’re really generating a lot of value, where people need to get involved, that sort of thing. So, we’ll touch on that a little bit. Raj, any comments here before I jump into the demo?
Rajesh Datta: No, I think there’s a lot of things to touch on, but let’s do it in the demo. We’ll talk a bit about the forms. We’ll talk a bit about the analytics. We’ll explain the integrations, what we prepare for and what we can also integrate with that we don’t know. Let’s go through the demo. I’ll talk a bit more there.
Kyle Hoback: While I pull this up, audience, want to make sure you’re engaged and getting what you want out of this webinar. So whatever questions you have, please submit them here in the background. I’m going through my screen right now, so we’re not going to be able to capture them live, but after we’re done with the demo here, we’re going to go through as many of the questions as we can. So please submit them as you think of them and we’ll try and cover as many as we can after. If we can’t get to them, we’ll find some way to get in touch with you or get a response to you.
Let’s get it started. So, let’s start with the data and start where the data kind of comes into the process. This is an email that might come in that Ilana’s going to be monitoring this inbox, and this might be one that pops in. So you can see considerable data and details within the email itself, in the body, things like the name of the insured, the mailing address, effective dates, risk profiles, lots of different data points that are coming in. So, not in a structured format. It’s an email. Person sending this created their own format. Maybe they send it this way every time, but everyone else that sends submissions probably varies up how they’re going to do it. So unstructured format, some definition to it, but quite a bit of data captured in the body of the email. But then you also have the attachments. So here there’s an SOV, here’s an ACORD 125. So quite a bit of data captured within this email. And this is the type of thing that Ilana is going to sit on and try and work with. This is one example. I’m going to jump to another one, but Raj, do you have any thoughts here?
Rajesh Datta: Just one quick point. You’re seeing an example here. The amount of inputs and the types of data that we can get as a part of the intake process, it varies from your semi-structured to structured to unstructured, however you want to categorize it. These are some examples. We’re able to handle various inputs as required to start the intake for the underwriting process. So you can go on from there, but I just want to touch on that.
Kyle Hoback: Sure. Yeah. And then here’s another example. So very different format of email. Quite a bit more documentation is attached. Various ACORD forms, the 131, 127, 140, 125, and 126, if I go through them quickly, as well as a loss run report, as well as some additional details in an Excel file here. So, some details in the body of the email, but obviously significant data that’s captured here in attached documentation. So, these are the types of things. A lot of emails might be somewhere in between, but typically what’s happening here is when these emails are coming in, then the underwriter’s going into to Duck Creek or other systems and really keying on this information in, copy pasting. Sometimes PDFs are tough to copy from, maybe there’s a window or dual screen action where they’re kind of bouncing back and forth or just toggling windows. So, significant manual process is typically required just to get all of the information into the systems of record and systems of note.
So this is where Ilana’s going to vastly simplify what’s happening. Ideally, she’s able to pick up all of the data, all of the documentation, understand exactly what’s in every piece of it, and just update the underlying systems. In other cases, she might need to escalate to one of her real world teams. And that’s what I’ll show you next, is kind of how that plays out, kind of what the audit trail is in all cases, but also how to escalate and work with other teams. Any other comments here, Raj, before I jump there?
Rajesh Datta: No, no. Sounds good.
Kyle Hoback: All right. So this is WorkFusion’s WorkSpace software. This is where you can see two of those submissions that have come through. And as the underwriter now, instead of going into email, going into Duck Creek and opening up the different systems, I can go in here and everything’s contained and everything’s captured. So by that, I mean, if I click into Amazing Realty here, you can actually see that I have kind of a summary of information we’ll kind of step through, but also the email, the ACORD form, the SOV, have all been captured and initially worked by Ilana so that as the SME here, I can come in and just kind of finish the job. But this includes kind of two areas here from the summary. You can see some enriched values I think Raj is going to want to talk through a little bit. But also kind of a summary what’s happening. So here, like the insured name was picked up, and not only was it picked up, but it was analyzed that it was found and matched across all documents. This is a good sign that the documentation’s probably pretty clean because it’s identifying and matching these fields across the various documentation. So, Raj, I think you were talking about enriched data a little bit.
Rajesh Datta: Yeah. I mean, just as a part of submissions, we know there can be gaps in some of the data. There may be some discrepancies as well. We have the ability and we know carriers are leveraging data for validation and enrichment needs. In this case, we partner with the provider, such a provider like Bold Penguin. That’s why you see Bold Penguin there. But in the other instances, we can integrate with any other external provider. And the examples here are getting data from internal systems, getting data from a provider like Penguin, getting data from external systems, other premium sources that may be leveraged in the current underwriting process. We can provide or we can accommodate those types of integrations.
Kyle Hoback: Perfect. So, the data’s not even limited to what’s in the emails themselves, but it’s very much focused on the emails themselves. So let’s get into that a little bit. If I click the email here, you can see that email I had shown before, but it’s marked up and kind of in a different format. So it’s in this format so I can interact with it if I need to. In this case, doesn’t look like I need to because it captured all the details, things like, what date it was sent, which is maybe a structured basic field to capture, but things like the name of the insured and the mailing address, far less structure and consistency there, yet it was able to pick it up because the machine learning is able to work across the different formats and layouts of the data and still know what name of insurance companies kind of look like and pick it out amongst the rest of the text. Same with addresses and same with the details at the bottom, like who’s submitting it, what their phone numbers are, these sorts of things. So it captured data in the email itself. And then I can show you the other documentation here as well. ACORD forms, standard forms, relatively structured, although, even though it’s a structured form, it’s not structured data in maybe a technical sense just because it’s still in a PDF, has to be worked with.
But then also, sometimes there’s variability in how these are submitted, particularly just around sometimes they’re fuzzy, sometimes they’re rotated, sometimes they’re poor scans. Sometimes there’re maybe faxes that come in, different ways that the data could be received that kind of make it tougher to just plan on a predictive format. So here again, the machine learning is useful to be able to work across the variety of the submission, even if the structure of the form is pretty consistent or very consistent, still able to kind of work through some of the variability that can come in to make it challenging to get the data out.
But it still did get the data out, and you can see a handful of fields here that have been collected, kind of pulling them from various spots of the form and structuring it off to the right hand side.
Rajesh Datta: Great. One quick thing to add here, Kyle, that, I mean, again, in this view, we can say that if there’s a discrepancy or some sort of gap, that’s where the continual learning takes place, right? From a model standpoint. So although we’ve trained the model to handle these inputs, understand the semi-structured data that may come out of a form such as a 120, ACORD 125, or any other ACORD form depending on line of business, but if there are discrepancies or things lacking for any input, the model will learn and the next time that’s run through, will address it.
Kyle Hoback: Okay, perfect. So then let finish here with the SOV. Again, quite a bit of variability. This one’s not going to be the standard form, so understanding the location information, the address information, as well as things like the insured name that’s been consistent across each of these. Just really understanding the data and what needs to be collected so that you can really get that new submission through as quick as possible.
So this is one example. Maybe the other one went straight through. Maybe there’s others here that need some adjustment as well. But the idea is you’re not only given a way to work with the documentation that wasn’t fully automated, but you also have an audit trail for what’s being collected. So once this is done, this is packaged up, able to go back in and see where all the data points came from, especially in those straight through cases where it’s just nice to know what happened, even if you don’t don’t need to see it, because the automations will care about all of it.
Then, just to wrap up the demo here, I’ll switch to analytics. And here, I mentioned this before, considerable KPI data just to understand what’s happening and how Ilana’s performing, as well as how she’s working with her teammates, the real world teammates that she has. So automation rates, daily submissions, knockouts, other detailed details around the NAIC code breakdown, average time for an underwriter submission, different ways to adjust and review the submissions. So, a handful of analytics here just to understand what’s happening and making sure that not only that Ilana is performing and not only that the underwriters are getting the benefit from her, but just really understanding how the overall value is being accrued across the automation, across the organization.
Rajesh Datta: And also, Kyle, as with everything from an automation standpoint and in the analytics platform within Control Tower, this is a configurable capability specific to the needs of the customer.