The insurance industry is under pressure to do more with less. Sales teams are overloaded with manual tasks, support teams face growing operational complexity, and customers expect faster, more personalized service. AI promises transformation, but where does it truly create measurable value?
In this practical, live demo session, Decerto moved beyond AI buzzwords to show how intelligent automation can be embedded directly into insurance workflows. Drawing on real implementation experience, the session demonstrated how AI supports sales teams, streamlines operations, and improves efficiency without disrupting existing systems.
Watch the recording to discover:
- AI-powered sales support – accelerating response times and reducing manual effort
- Virtual assistants in daily operations – improving communication and task management
- AI-assisted straight-through processing (STP) – delivering faster, fully automated policy issuance
- New AI-driven growth opportunities – unlocking cross-sell potential and strengthening retention
Rather than replacing experts, AI should amplify their capabilities. See how Decerto’s Agent Portal integrates intelligent automation into core insurance processes, helping teams work smarter and drive sustainable growth.
Missed the live session? Watch the recording or read the full transcript below.
[Speaker: Jim Erickson]
Hi, everyone. Hello and welcome to our session today, Smarter Insurance with AI, including an Agent Portal live demo, which is the focus of this event today. Happy to be here. This event is sponsored by Decerto and hosted by Digital Insurance. My name is Jim Erickson. I'll be your moderator today. I am a former editorial director. I'm a market research analyst. I cover a lot of issues specific to insurance also, and so I always look forward to these sessions to catch up on the latest and greatest in our industry. Obviously, everything in my work today has to touch on AI somewhere. And while that is certainly inevitable and unavoidable and important, and it is important, it also has to be put in the context of our working processes, our same objectives, and desired outcomes and challenges we face in the insurance industry, including things like agent satisfaction, churn, and the usual metrics by which we judge our performance and improve our processes in search of that perfect customer experience. So I have a couple of good guests here today, who are going to be demoing this solution that Decerto has. We have Małgorzata Niemiec, who is the head of delivery and the AI lab at Decerto, and Maciej Wir-Konas, who is the head of the agent portal. I've been discussing this event with these folks for a couple of days, and I'm really interested to follow up on this. And for those of you who are not fully aware, Decerto is a leading technology provider and partner. Been around for more than twenty years, and they're delivering high-end IT solutions in insurance. They have comprehensive services from systems architecture and software development to integration and long-term maintenance, in addition to engineering expertise. Decerto's flagship products are Higson, a business rules engine, and the Claims AI system for the agent portal, which is what we're going to be focusing on today. So let's get right to it, and I'll start by handing it over to Maciej to get us going today. Maciej, you have the floor. Go ahead, please.
[Speaker: Maciej Wir-Konas]
Okay. Thank you. Thank you very much, Jim. As you mentioned, nowadays, most webinars would show you a slide about AI, that's for sure. But today, we are going to do something different. We are not going to talk about AI first. We are going to talk about Sarah. Sarah is an insurance agent. She sells policies. She supports her clients. She's the person clients call when something goes wrong. She's good at her job, and like most people who are good at their job, she spends a significant part of her day doing things that have nothing to do with why she became good at it. So today, we are going to follow Sarah through one typical Friday, and somewhere along the way, you will see AI, but that's not really the point. Before I hand over to Małgorzata, a quick note for those of you who are watching with a specific challenge in mind. At the end of today's session, we are going to offer a free proof of concept, your documents, your process, and your data. Just two weeks from first conversation to a working pilot. No commitment required. Just please keep that in mind. So, Małgorzata, I will now pass the discussion over to you because before we get to Sarah's Friday, I think it's worth understanding how we built what we are about to show.
[Speaker: Małgorzata Niemiec]
Thank you, Maciej. So to start, I would like to say a few words about the difference between talking about AI and talking about the real problems. I spent fifteen years on the insurer side before joining Decerto. Director of product, pricing, and underwriting at Allianz, before that, at Warta in the Talanx group. I built teams. I built models. I sat in rooms where we talk about data, predictions, and optimization. And what I learned is that the technology is rarely the hard part. The real challenge is understanding the problems we actually need to solve.
Over the past few years at Decerto, we've spoken with agents, BPOs, underwriters who are overwhelmed by manual work. Up to eighty percent of their time is spent on repetitive tasks that have nothing to do with why those people got into this field. We've met operation teams struggling with growing process complexity. We all know customers expect fast, personalized service and have no patience for delays. In insurance, speed is conversion. Quotes that take two hours are lost to competitors responding in five minutes. The MIT studies show that your chances of qualifying a lead drop twenty-one times when you wait just thirteen minutes instead of five, not to mention NPS statistics.
But let's come back to AI.
One of our clients, a top ten European insurer by premium volume, came to us with a very simple request. And this request was, "We want AI." And our answer was also very simple. "Please tell us where it hurts first, and we will use AI to solve exactly that." When we built the AI lab at Decerto, we didn't start with a list of AI capabilities. We started with conversations. We sat down with sales, product teams, agents, claims handlers, underwriters, and we asked one simple question:
"What part of your day would you happily give to someone else?"
These conversations, not industry trends, not analyst reports, are what shaped everything you're about to see.
I must say that during these conversations, the answers were remarkably consistent.
Across different companies, different markets, different lines of business, people were spending enormous amounts of time on things that felt like administration: data entry, document checking, searching for answers that already existed somewhere.
And the consequence of all that time wasn't just inefficiency, it was distance. Distance from clients, distance from the decisions that actually matter. So we built from there, not from the technology out, but from the problem in. And today, Maciej and I are going to show you what it looks like in practice in our Agent Portal, our platform for insurance agents. We've selected six scenarios, six moments in a real working day, and the question we are going to answer is not, "What can AI do?" It is, "What does this actually change for Sarah?"
So Maciej, let's go to Friday morning.
[Speaker: Maciej Wir-Konas]
Okay, so let's concentrate. It's Friday morning, nine o'clock. Sarah has a coffee on her desk and a new message in her inbox. A potential client, let's call him Michael, wants to insure his car. He has already done his research. He knows what he wants. He sends Sarah a photo of his vehicle registration document and asks for a quote. Without Agent Portal, this is what happens next. Sarah opens the car registration document photo on her phone, opens the policy form on her screen, and starts typing. VIN number, seventeen characters, case sensitive, year of production, make, model, and so on. Also, owner name, address, and the other data. If she is lucky, it will take five minutes, but if Michael has handwriting like mine, it takes longer. And while she's typing, Michael is waiting, and the sale, of course, is waiting. So let me show you what this looks like inside Agent Portal.
Okay. So, Sarah opens a new offer. The system immediately gives her the option to scan a document, and she uploads Michael's car registration document photo, taken on a phone, slightly at an angle, just typical real world conditions. And this is what happens. The system reads the registration document and populates the form with all the data needed. We have to add just driving record data, number of accidents or tickets. Of course, we can additionally get the data automatically from different external sources. Then Sarah confirms, reviews, and moves on. From photo to completed form, just about five seconds.
[Speaker: Jim Erickson]
You may be muted, Małgosia.
[Speaker: Maciej Wir-Konas]
Ah, okay. Sorry.
[Speaker: Małgorzata Niemiec]
Okay. Yeah, I'm sorry.
[Speaker: Jim Erickson]
Oh, there you are. Keep going, please.
[Speaker: Małgorzata Niemiec]
Okay. Sorry. Okay, so if I may, I want to tell you where this came from because I think it matters. One of our clients, a large insurer working with an extensive agent network, told us that the agents were spending close to forty percent of their working day on data entry. Not advising clients, not solving problems, just typing.
When we heard that, our first thought was, that is forty percent of the relationship that never happens. That is forty percent of the time Sarah could be spending understanding what Michael actually needs, finding opportunities to cross-sell, to up-sell, which is so important for insurers.
And our reaction was, "Let's build an AI-assisted OCR tool."
The technology we built is a response to that, not the other way around.
[Speaker: Maciej Wir-Konas]
Okay, so let's go further. It's ten fifteen. Sarah is still working with Michael's application, and he has sent over one more document, a scan of his previous insurance policy. Let me choose it.
Okay. Just the standard procedure. Verifying a policy scan manually means checking the document. Is this document genuine? Does the policy number match? Are the signatures present and consistent? A thorough manual check takes time, so sometimes things get missed. Let me show you what the Agent Portal just did. So the system analyzed the document structure.
It identifies the key elements inside, so the policy number, contact information of the insured client, signatures, and cross-references them against the application data. In this case, everything checks out. We have a green light. So the document appears authentic. Sarah gets a confirmation and moves on. Now let me show you the second scenario, because this is where things get interesting.
Okay. We have the same process, just finishing an offer, binding a policy, but a different document, just vehicle photos. Okay?
And so this time, the system is analyzing the vehicle photo, and the system flags a warning. So specific vehicle information on the policy scan doesn't align with the application data. Sarah sees the warning immediately. She doesn't have to find it, it's surfaced for her. The decision, of course, is still hers. So just investigate, go back to Michael and escalate. But she knows in seconds what to do. And then there is the vehicle itself. So we will try to add another photo.
Okay.
Yeah, so now the system verifies, so does the vehicle match the declared make and model or color? Oh, okay, we have a photo upload failure, so I will try again. Yeah, this can happen of course.
But you have a significant signal here, just to try this another time. Okay. So, we will try again with maybe a different photo. And we have here some examples. Okay. It's like a typical situation during, you know, live demos, so-
[Speaker: Jim Erickson]
Live demo. It's the live demo.
[Speaker: Maciej Wir-Konas]
Live demo, yeah. Okay. So, AI processed it. So now you can see that we check does the vehicle match the declared make, model, and color. It's correct. But we have the information, is there any existing damage? So there is a flag, damage detected for damages. You could see a photo. So in this case, for Sarah, this is valuable information. She can document it, she can have a conversation with Michael, she can make sure the policy accurately reflects the actual condition of the vehicle. No surprises later for either of them.
[Speaker: Małgorzata Niemiec]
Yes. And I want to share something about this feature specifically because the story behind it is a good example of how we work. Our client came to us because the operations team was spending an overwhelming amount of time on verification. It cost both time and money. The process worked like that. The agent submitted documents, and when the operations team flagged an issue, they would escalate to the agent, who would then go back to the client. It was so slow and rarely a pleasant experience for anyone involved. Our solution changed that entirely. Discrepancies are caught immediately. The agent and client can resolve any issues on the spot. In many cases, no operations team is in the loop, no back and forth.
And our client didn't come to us saying, "We need AI." They came to us saying, "We need to fix a process that's costing us too much."
What we believe at Decerto is that the business case comes first, not AI. AI is only as valuable as the problem it solves. What is also important to underline is that we can effortlessly review, process, and extract data from virtually any type of document. Scans, handwritten notes, digital files or images. With today's tools, format simply isn't a barrier anymore.
[Speaker: Jim Erickson]
It's really interesting. I like both the features for the VIN registration automation, and this photo upload, and it has a kind of exception management built into it to flag that an improper vehicle photo uploaded has been spotted. This is really the kind of low-hanging fruit that I think a lot of people would love to see automated. A question for you, Małgosia, when the system flags a conversation, what does it do? Does it stop the process? Does it prevent Sarah from executing her process with the customer?
[Speaker: Małgorzata Niemiec]
No, no, no. And this is a deliberate design choice. The system surfaces information. It doesn't make the decision. Sarah sees the flag, she has full context, and she decides what to do next.
[Speaker: Jim Erickson]
Yeah.
[Speaker: Małgorzata Niemiec]
We believe very strongly that AI should make the human more confident, not invisible, especially at the point of sale where the relationship with the client is being formed. At the same time, I need to add that we also work with insurers who take a different approach. They prefer the agents not to participate in some processes. In those implementations, agents simply submit the documents, and the checks and validation are carried out by AI, and if there is a problem, a dedicated team solves it in the next phases of the process. So it is the insurer that decides how to implement it.
[Speaker: Maciej Wir-Konas]
Okay, so maybe before we move to the second half of Sarah's day, just a quick note. If you are already thinking, maybe this looks like our problem, we would encourage you to drop a message in the chat, so we will make sure the follow-up includes the details on how to arrange a scoping conversation. It costs nothing and takes about 30 minutes, so-
[Speaker: Jim Erickson]
That's great, Maciej. And also, I want to tell people too, I'm taking notes. We're going to leave some time at the end for Q&A, so please use this opportunity to talk to Małgosia and Maciej about... I think this is fascinating, and I could talk all afternoon about it. And the last thing I'll say before we jump back in is we're going to have a recording of this session available, and we'll send a link to it to you as soon as it's available. So, Maciej, sorry to interrupt. Why don't you carry on, please. Thank you.
[Speaker: Maciej Wir-Konas]
Okay, okay. Thanks, Jim. Yeah, so let's go to another situation. So now it's 11:30. Sarah is on a call with another client. Let's call her Anna. Anna is renewing her home insurance, and she has a very specific question about a clause in the policy terms. It's just something about coverage for damage caused by a third party on the property. It's a legitimate question, so the answer exists. It's somewhere in a 180-page GTC document. But Sarah, of course, can't recite 180 pages from memory, and putting Anna, her client, on hold while she searches is not the experience either of them wants. So let me show you the AI assistant in the Agent Portal. While Sarah is still on the call, she types the question into the AI assistant, this question about a third party on the insured property, just in plain language, as you can see the question, exactly how she would ask a colleague. So she gets a response.
The system searches across the relevant policy documents and returns an answer. We can go deeper into the discussion with the AI summary. So it's not a summary of where to look, just an actual answer. And Sarah has this answer in seconds. She relays it to Anna. The conversation continues. No hold music, no typical "I will call you back."
[Speaker: Małgorzata Niemiec]
Yes, and this feature came from a very simple observation. Agents are expected to be experts across multiple products, multiple product versions, multiple sets of terms and conditions, and it's an enormous amount to carry. And what we noticed was that a lot of agent errors, wrong information given to the client, delays, unnecessary callbacks, they weren't happening because agents didn't care. They were happening because the knowledge was buried somewhere.
We didn't want to build only a chatbot. We wanted to build something that turns company documents into an instantly accessible knowledge base. That is why we added a virtual assistant to our agent portal. Our assistant operates on a RAG-powered knowledge base, and the agent remains the expert.
The virtual assistant just makes sure the knowledge is always within reach.
[Speaker: Jim Erickson]
Małgosia, are there specific document formats or styles that the documents have to be in to be consumed?
[Speaker: Małgorzata Niemiec]
It works with documents you already have, policy terms, procedure manuals, product guides. There is no need to restructure or reformat anything, and it integrates with the existing system environment. And that is actually a theme you will notice across everything we've shown today. These features are designed to fit into what already exists, not to replace it.
[Speaker: Jim Erickson]
Great.
[Speaker: Maciej Wir-Konas]
Okay. So yeah, we can go further, I think.
[Speaker: Jim Erickson]
Sure.
[Speaker: Maciej Wir-Konas]
So, yeah, it's at 2:00 in the afternoon, so Sarah's phone rings. It's one of her long-standing clients. Let's call him Robert. He was in a collision this morning. Everyone is fine, but the car is damaged, and Robert is already asking, "What do I have to do? How long will this take, and what happens next?" So Sarah needs to register the claim.
So Robert sends her the documents, so we will now upload them.
Okay. So, there is a-
[Speaker: Jim Erickson]
Could that be a photo or a scan, or... So they could shoot a picture, or they could scan it on a printer or something, or?
[Speaker: Maciej Wir-Konas]
Yeah, like all of the options. Yeah.
[Speaker: Jim Erickson]
Yeah.
[Speaker: Maciej Wir-Konas]
So scans and separate files, separate-
[Speaker: Jim Erickson]
Yeah.
[Speaker: Maciej Wir-Konas]
different formats. So he sent her a police report, a repair estimate, and the claim form. We can add them in parallel and start analyzing with AI.
Okay.
Okay. So manually, this takes time, and while Sarah is processing, Robert is waiting and calling back. So now, after Sarah opens a new claim, she uploaded three documents from Robert simultaneously, as you could see, and the system processed them in parallel. So what we have here is the summary. So incident date, extracted. Location, extracted. Claim type, identified.
Also, there is a small description of the situation. So vehicle information from the police report, also cross-referenced and populated. Sarah sees all the extracted data laid out for review, so she can check it.
Yeah, with some details, and everything looks correct. She can add the claim to the policy, and she can see it in the Agent Portal just in one place, easy to navigate. So the claim is now registered. Robert's file is updated, and Sarah can call him back with a status in minutes, as you could see, not hours.
[Speaker: Małgorzata Niemiec]
And to put that in numbers, with one of our insurance clients, average claim registration time dropped from 18 minutes to under three minutes. That's not a vanity metric. These numbers translate directly into cost savings, NPS, renewals, and the kind of client relationships that actually last.
[Speaker: Jim Erickson]
This is great. And Maciej, your point about all this being visible in one place is such a big deal for some of the folks that I work with who have to navigate applications to surface these things, and they're not processed in parallel, as the... You know, this looks pretty utilitarian. But what about, what if a document's unreadable, or it's a blurry photo or something? What happens with handwriting, that kind of stuff?
[Speaker: Maciej Wir-Konas]
Yeah, so it's a good question. But the system is designed to handle real-world document quality, including handwritten content, photos taken in low light, scanned documents that aren't perfectly aligned. Yeah, to be honest, we also tested pages that had been flooded with coffee, to ensure the system remains reliable even under less than ideal conditions. And where confidence is lower, the system flags specific fields for Sarah's, for the end user's attention, rather than silently guessing. So she always knows exactly what was extracted cleanly and what needs a second look. So as you could see during my presentation, transparency is built into every step.
Yeah, okay. So let's go further. So it's 3:30. Sarah has had a full day, and there is one more thing in her queue, a new business submission from a client. The client has sent an email with insured details, a submission document attached. So without automation, this is a multi-step process. The email is read by Sarah. Data is extracted manually. The files go next to underwriting. Then the underwriter assesses it against criteria. A decision is made, and at the end, the client is notified. So on a good day, it's like 24 hours, sometimes even more. So let me show you what straight-through processing looks like in Agent Portal. So here you can see that the client's email arrives. Agent Portal picks it up. The system reads the email and the attachments. The solution, of course, can be integrated with the most popular email providers such as Outlook. So extracting the insured name, driving record data, line of business.
Yeah, vehicle information and the relevant financial values. So everything is extracted and mapped automatically. No manual entry. And now this is where it gets interesting. So the extracted data is passed to Higson, our dedicated business rules engine, available also as a standalone product. The rules it evaluates against are yours entirely, written by your underwriters, maintained by your business teams, no code required. So Higson makes sure they are applied consistently to every single submission, not just the ones that land on the right desk on the right day. Does this risk fall within appetite? Does it meet the conditions for automatic binding? So let's try.
And you can see that some of the elements did not pass the review that was required. So the automated underwriting has been made, and here you can just react. So you can, of course, override that decision, and go further, if you decide that those elements are okay to bind the policy. Or you can stop here, ask your client and change the offer, change the scope of insurance.
So we can try with another submission. Okay, yeah, the status has changed.
Okay.
Now let's...
Okay.
Yeah, so here, the policy was bound automatically, so all the conditions are met within automated underwriting, and you, as an insurance agent, don't have to do anything. So there is a policy bound automatically, and you can go further with post-sales activities.
Okay.
[Speaker: Małgorzata Niemiec]
That's really cool. We actually had a question about binding too, so I think you were spot on with that observation. Okay, should we proceed from there, Maciej, to the next-
[Speaker: Maciej Wir-Konas]
Yeah, yeah.
[Speaker: Jim Erickson]
Question for Małgosia. There are, obviously, a lot of vendors working in this space, and most of them are built on the same sort of large language models. You're the engineer here, so what makes the economics of your approach different or more interesting to prospective customers?
[Speaker: Małgorzata Niemiec]
Okay. It's a fair observation, and an important one.
Most AI solutions in this space are built around flagship large language models. They are applied uniformly across all tasks. That works in a demo. In production, at scale, it creates a cost structure that is hard to justify and often hard to explain to a CFO.
And our approach is different. We look at optimization not only through the lens of outcomes, but also through the lens of costs. That means we do not automatically reach for the most powerful model for every task. Instead, we break each task into its components and match the model to the complexity of that specific step.
Take data extraction from a structured document, a well-defined repetitive task. A lighter, faster model handles it reliably at a fraction of the cost. There is simply no reason to use something more expensive. But when the task involves real complexity, unusual document structure, incomplete data, borderline cases, that's where the more advanced models earn their place. The result of applying different models to different tasks is the same output quality, at significantly lower operational costs. It is a smarter architecture, not just smarter models.
And when you're running this across thousands of submissions a month, that distinction in cost is really, really material, believe me. And in fact, one of the core responsibilities of our AI lab is exactly this, continuously testing models against real tasks, evaluating both the accuracy and the cost profile across different use cases. So when we recommend an approach, it is not theoretical, it is based on hands-on testing.
[Speaker: Jim Erickson]
Testing, before Maciej resumes, testing is so important here. Fifteen years ago, we went through the big data conundrum. We went through, before that, the unified database model. We're only going to need one database in the end, and all these things were based on workloads and, you know, what we began to call horses for courses when columnar databases came along, was obviously very much purpose-built. So just having the GPS on your dashboard could solve so many problems with an internal database, which seemed remarkably comprehensive, and is. So I really appreciate that it's not a one-size-fits-all engineering approach. I think that's a must-have for anyone in this space-
[Speaker: Małgorzata Niemiec]
Yeah.
[Speaker: Jim Erickson]
crowded space of vendors. So Maciej, go ahead, please.
[Speaker: Maciej Wir-Konas]
Okay. So, before I close with Sarah's Friday, I want to address something directly, because if you are evaluating solutions in this space, you are talking to more than one vendor, and you should be. So the question isn't whether AI can read a document or automate a submission. It can. That's for sure, and most vendors will show you a demo that works. But the question is what happens in month six, when the edge cases arrive, when your regulatory environment shifts, and when you need, for example, to change a rule without raising an IT ticket? And what differentiates Decerto is three things. So first, we have vast insurance experience. We have spent twenty years working with some of the largest insurance companies in the world, and our specialists combine deep industry knowledge with modern technologies, ensuring products of the highest quality. The second point, our implementation model starts small and proves value before it asks for a budget. Every client starts with one module, one team, one pain point, with a working pilot in two weeks. And the third point,
I can say that we don't create dependency on our systems. So Higson's rules are owned by your business team, and Agent Portal sits on top of your existing systems via API, so you are not locked in.
But let's come back to Sarah. So Sarah made it home before 5:00. That sounds small, but think about what that actually represents. Every hour Sarah spent this week on data entry, on document checking, on searching for answers, that's an hour she didn't spend talking to Michael, to Anna or Robert, and building trust, understanding needs, doing the thing that actually makes someone a great insurance agent. So we didn't set out to build just an AI product. We set out to solve a set of problems that real people, real Sarahs, described to us in real conversations, and the technology came second. The problem came first.
[Speaker: Małgorzata Niemiec]
Yes. And if I may, I want to come back to something that was said at the very beginning about AI being everywhere right now. That is true, and a lot of what's out there is really technically impressive. But if it doesn't solve a real problem, it isn't a solution. It is just a demo.
What we showed you today maps to six specific moments in a working day. You might recognize all of them. You might recognize just one. You may also have similar challenges, just in a different part of the organization, whether in client services, claims handling, or somewhere else entirely. The patterns repeat, only the context changes.
And maybe one more thing from my side. We would like to recommend evolution, not revolution. You do not need to transform your entire operation to start seeing the value of AI. Every client we work with starts somewhere different, and then they build from there, at their own pace, in the direction that makes sense for them, not for us.
So pick the problem that costs you the most right now, start there, and if you would like to find out what starting there actually looks like for your organization, we would like to offer you a free proof of concept based on your data, your process, your documents, and we will deliver a working pilot in two weeks.
Just reply to the follow-up email we are sending out tomorrow, and we will get back to you within twenty-four hours to schedule a thirty-minute scoping call. There will be no sales pitch, no commitments, just a focused conversation about one specific problem.
[Speaker: Jim Erickson]
Okay. Well, that's... So you will be getting a follow-up email to solicit your thoughts on this and get that proposition for a pilot. I think it's all very compelling, and I appreciate having an engineer as well as an experience expert with the portal on this call to sort of understand what's going on behind the scenes. What Malgosia was saying also about this being API driven, that it sits on top of existing systems. I believe Decerto works with core systems and pretty much all the aspects of insurance, back-end as well as front-end. But I think the point that different carriers start at different points and have different pain points is precisely the right approach. I remember when the initial CRM rollouts were being made in the organizations that I worked in, and we were promised it was going to change the way we work, and people did not want to change the way they worked. They had the relationships and their processes and the same objectives that they'd had before the technology arrived. Very important that it be transparent. And you were never working in a unified place. And I think different solution providers have addressed this in different ways. But
I also appreciate that this was a live demo. You were watching someone experiencing, working through it, not just a recording of a perfectly operating solution, so that was really good. I do want to remind you that we'll send a recording also of this session to you as soon as we have it available, and we do have time for a few questions. So I'm going to start out with, okay, well, you said two weeks to a pilot. How about this? How long does implementation actually take? That usually it depends, followed by a very long number. What's your response to that?
[Speaker: Małgorzata Niemiec]
I appreciate the honesty of this question, and the honest answer is it really does depend. But let me give you something more useful than that.
[Speaker: Jim Erickson]
Thank you.
[Speaker: Małgorzata Niemiec]
The features you saw today are modular. So if you want to start with document scanning in the sales flow, that is a contained scope. It doesn't require you to rewire your policy administration system or your claims platform. It integrates via API with what you already have.
So we've had clients move from a first conversation to a working pilot in one week, not because we rushed, but because we scoped tightly, focused on one specific pain point, and built from there. And I think the clients who struggle with timelines are usually the ones who try to do everything at once, and our recommendation is always the same. Start with the problem that hurts most, prove the value, and then expand.
[Speaker: Jim Erickson]
Okay.
[Speaker: Małgorzata Niemiec]
So it is my honest answer. It depends.
[Speaker: Jim Erickson]
No, I think that's a fair answer, too, you know. And it's not, no situation is identical, as you pointed out, but I think the versatility, the variability of this tool seems pretty flexible too, as well as transparent. Here's the next question. What about data security, GDPR, a lot of PII and financial information being passed back and forth. Who wants to handle that?
[Speaker: Maciej Wir-Konas]
Okay. Maybe, Małgosia, I will answer. So,
I think that all of us will agree that data protection is absolutely fundamental, not only to our platform, and it is embedded directly into our system architecture. So rather than treat it as an add-on, we do care about data security, and all information is handled in accordance with GDPR and applicable data protection regulations. The solution also can be deployed in your own cloud environment or on premise, depending on your organization's requirements. We don't retain client data beyond the processing transaction. It's very important. And we would also recommend a dedicated conversation with your security and compliance team. The specifics of your environment matter. So data protection is built into how we architect these solutions from the beginning, from day one.
[Speaker: Jim Erickson]
Okay. All right. Sounds good. Another question here. The straight-through processing demo was interesting. We operate in a specialty lines market where every risk is different. Is any of this relevant for us?
[Speaker: Małgorzata Niemiec]
Maybe I will just take this question. So this is a really important question, and the answer is yes. For straightforward lines of business, yes, but also for complex ones, AI has its value. In the latter, the complexity is often in the documents. A submission from a broker can be dozens of pages across multiple attachments. Before an underwriter can even begin to assess the risk, someone has to read all of that and extract the relevant information. And that's not underwriting, that's administration, and AI can easily handle this.
So you can make underwriters faster and better informed, and that applies everywhere.
[Speaker: Jim Erickson]
Okay. I got two more quick ones. One, here's a live question from the audience. How does this differ from intelligent OCR and traditional workflow automation tools? It's a little broad, but, you know, how would you describe it versus OCR and traditional workflow automation?
[Speaker: Małgorzata Niemiec]
I will take this question, Maciej, if you don't mind. So traditional OCR just reads text from a document. It doesn't understand what to do with it. And traditional workflow automation follows fixed rules. If a document looks different than expected, that process breaks. Our AI goes further. It understands the context, and, thanks to that, the OCR, the data extraction, is much, much better.
[Speaker: Jim Erickson]
Okay. Okay. Now last one, and then we'll let you folks get on with your day. I really enjoyed this presentation, by the way. I recommend you maybe pass the playback to some of your colleagues if they're interested in, if you know someone who's interested in this kind of thing.
So last question here. All right. Well, Decerto, like I said, you have experience in sort of the whole front and back office, but how does this connect to any given, you know, core system? You showed the agent experience, but what about the SI part of it, the systems integration, how does it connect?
[Speaker: Maciej Wir-Konas]
Yes, so-
[Speaker: Jim Erickson]
The API, right?
[Speaker: Maciej Wir-Konas]
Yeah. So our solutions are designed to connect to existing systems, rather than sit alongside them as something separate. So policy administration systems, claims platforms, underwriting workbenches. We have integration experience across a wide range of environments, including legacy systems. So the data extracted by AI flows into your existing data structures. So for example, Sarah's claim registration doesn't create a parallel record somewhere else. It fits entirely into your claims system, and the underwriting decision fits your policy platform. So this is something we do with a clear purpose. We are not asking you to replace your core infrastructure. It's very important for us. We are adding an intelligent layer on top of what you already have. So that's also why individual modules can be adopted independently, because they plug into what is there, rather than requiring you to commit to a full platform replacement, like a big process.
[Speaker: Jim Erickson]
Yeah. That's a very important commitment to make to a lot of people on the other side, on your client base. Nobody's going to rip and replace for AI's sake alone. I know that. And fortunately, I don't think we have to. I think there are just some really fascinating developments on the automation side. And, again, I would also think back to, I appreciate Malgosia's discussion on the large language models, and that it's not one-size-fits-all. It's horses for courses, you know. There's really a utilitarian angle and a modular approach, you know, and not a full bent makeover, which this is not. I think it's a good place to test ideas and to examine how to improve pain points. And you're going to get a follow-up email, I guess. The folks at Decerto are offering this connection and a pilot. So that's something I think you might be interested to follow up with. So, we're going to let you get on with your day today. I want to thank Malgosia Niemiec and Maciej Wir-Konas from Decerto for their excellent presentation today. I thought it was very transparent. I appreciate the live demo. I want to thank the audience for staying with us and asking good questions. We didn't get to all of them today, but we'll be happy to follow up with you on any that remain. Decerto and Digital Insurance, thank you for joining us today. My name is Jim Erickson, and with that, I will say so long, everybody, and have a great week.
[Speaker: Małgorzata Niemiec]
Thank you.
[Speaker: Maciej Wir-Konas]
Thank you very much.
Get in touch with us to discuss how we can implement it in your company.
Contact us
