From Doc to Dollars: Unpacking the ROI of IDP & Hyperscience

Xabi Ormazabal: Good morning, good afternoon, good day for everyone, wherever you’re joining us from. My name is Xabi Ormazabal, and I lead product marketing here at Hyperscience. I couldn’t be more excited to welcome you today to today’s session, where we’ll be digging into a topic that’s top of mind for just about every enterprise right now, the real measurable ROI of intelligent document processing and of Hyperscience specifically. And if you’ve been anywhere near a boardroom conversation this year, you know why it matters. The ROI of AI has become the question. Every enterprise is being asked to prove what they’re investing in to see that it’s actually paying off, and just trust me isn’t good enough anymore. Today, we’re bringing you an answer built on real data, not projections.

So just to introduce ourselves, I lead product marketing at Hyperscience, as I mentioned. I’ll be joined by Jorge Peña, who is our VP of value engineering, and then our esteemed guests from IDC, Andrew Gens, senior analyst for Document AI and Exchange at IDC, and Megan Szurley, manager of business value strategy practice at IDC. So first up, just in terms of what we’ll cover in the session, we’ll hear from Andrew on the state of IDP today, where the market stands for intelligent document processing, and the key drivers shaping buying the decisions right now. From there, Andrew will stay with us to unpack the ROI equation in IDP, how buyers should be thinking about value, and the factors that matter most when evaluating a solution. Andrew will then pass things to Megan for a closer look at the results of our business value study, the real numbers behind what Hyperscience customers have achieved straight from IDC’s independent research and interviews.

Jorge will bring us home with a look at how we approach value engineering here at Hyperscience, turning that research into a practical playbook you can use. And then we’ll wrap with some questions and answers where you can bring your own questions directly to the group. With that, Andrew, over to you.

Andrew Gens: Thanks, Xabi. So before we begin with the actual data from the study, I think it’s important to consider what is the market context and what are some of the buying drivers that we’re seeing. So to dive a little bit deeper into that, we can see that intelligent document processing is moving very rapidly out of the back office and into the buying plan. A rapidly moving market compounding at close to 20% a year with a lot of factors influencing that, including cost, risk, cycle time, all that tied up with unstructured content.

So to just dive into those five drivers you see on your screen. The first one, content volume. We see that unstructured documents are growing rapidly through volume, much faster than the teams paid to read them can grow. And as those teams grow, that also introduces new costs, leading us into the second driver. So cost per touch is a major one. The manual keying, exception handling, all that time that individual workers spend with the documents trying to get them ready for prime time, that costs money. And so trying to find a solution to streamline that process and in many ways automate it is a big force in order to make those document processes more efficient and smooth.

The next one is cycle time. Cycle time is really important in this hypercompetitive day and age where the competitive aspect of a solution or a process can be dictated by how fast it can be turned around. The buyers expect a turnaround in hours, not days, on a lot of these documents like loans and claims that are handled by these IDP solutions. Next up is compliance and auditability. AI transparency rules as well as industry-specific rules mean that these documents have a lot of specific aspects that need to be vetted and verified and traceable with an explainable record to back that up. And so an IDP solution is a solution kind of built to maintain that record and streamline that whole process, making sure that every step is recorded, every step is traceable, is a big part of making sure that there’s no pain in that process, and you can readily go and check back if there’s any need to do so.

The next one is gen AI readiness. Agents, generative AI models do best when they’re structured or when they’re grounded in structured content from your own business. And so having a source of structured content that is wholly your own that you can reference back to, to train models or to ground answers is a huge boon when it come to being able to get the most value out of your generative AI tools. And so IDP is a very logical on-ramp to that, just unlocking the value from the vast amounts of unstructured data that lives within the classic enterprise. Looking ahead, we can see that there are many steps to the ROI equation.

So jumping right into it, value really comes in multiple forms. One of those is avoiding the touch cost, as referenced previously in the earlier slide. The more you can focus on those straight-through processes, getting document processes automated, the more time you save, the more touches you save. That really amounts to a lot of value that can be pulled up and exposed that really is not able to be expressed when you have so many needs for rework and multiple different personas to look at each document and double-check if everything is as it should be.

Next is accuracy reducing rework. As referenced before, if you have an error rate, that is a cost on your business. Every error means that a document needs to be rechecked, and so having a high accuracy rate really is a boon in that you can just trust that these documents are being pushed forward and structured in a logical way that is actually relevant and accurate to the data being expressed in the unstructured content.

Next value comes in cycle time. Each day saved is a lot of value for the enterprise, and so the faster decisions being able to be achieved through these structured documents, and these automated workflows is really one of the largest areas of value that can be gained, where it may have taken 18 minutes to do one document, it can now be done in six minutes. That time compounds over each document that is passed through, and ultimately, you’re making a lot of savings just through that form of value.

The next is falling cost to run. So as these processes become more known, each workflow becomes more known, both to the vendor providing the IDP solution and also to your experts within your business that are helping to architect these processes. They become more efficient. And that goes hand in hand with the ModelOps, being handled primarily by the vendor, in this case, Hyperscience, where they’re constantly looking to improve their processes, their models to handle these documents more efficiently and faster with a high accuracy rate. And so you can ultimately see a cost per document decreasing over time, both as your organization gains competency and the vendor you’re working with gains expertise and more ability to monitor the consumption of their models that they’re using.

The next is workforce reallocation, and this is looking at how are those people who were previously handling those documents going to be leveraged going forward now that you have automated processes. All those folks, you may have a team of 10, 20, 30, or more, you may look to uplevel those folks and move them off of the structuring of content and into higher value workloads. Or if you want to have people that are really process experts to help you refine these IDP workflows going forward or these document workflows going forward, they can maintain that staffing level while allowing you to scale without needing to add extra headcount. And so there’s really multiple ways you can leverage that kind of workforce reallocation and workforce reskilling in order to increase the value that you’re getting from just a small team of folks.

And the final one here is the downstream value from AI provisioning. As we move further and further into a generative AI world and agentic AI world, these systems, as was mentioned previously, do best when they’re grounded in your own knowledge, your own documents, and your own data. And having a system or a solution that’s able to feed that data directly from your own documents into these solutions makes them more grounded, makes them more reliable, and ultimately gets you better value out of those solutions going forward.

Xabi Ormazabal: Yeah, I was going to… I’ll stop there, Andrew. One thing on your previous slide, sorry — on the accuracy-reducing-rework point. I really love that phrase, the cost of a bad record. It’s something we think about a lot in terms of it kind of almost has two downstream impacts. First of all, it’s a badly transcribed or extracted value can actually corrupt a decision of processing a claim or if someone is eligible for a benefit, et cetera. And the second is that that bad data persisting, it kind of limits the ability to trust the data downstream for agentic systems.

So I really like that you’ve painted this broad canvas of the way value emerges, but that’s one of the points that jumps out at me the most.

Andrew Gens: 100%. If it’s one thing for a document to have an error and require rework, it’s another for it to have a cascading effect where it’s impacting decisions negatively and actions negatively down the value chain. That’s a great call out, Xabi. Thank you. And I’ll pass it over to Megan.

Megan Szurley: Great. Thanks, Andrew. So I’m going to walk you through the actual business value that actual Hyperscience customers achieved through the course of doing an IDC business value study. To give you an idea of what to expect and what I’m going to walk you through, I’m going to just start with just background methodology, firmographics of who we spoke to, to give you a good idea of the foundation of the research. From there, we’ll jump right into the actual value drivers of Hyperscience, which include a lot of great metrics and customer quotes that illustrate the value. And I’ll help you understand how we got to those calculations. Next, we’ll move into the actual ROI analysis, which includes a great payback period, as well as some really detailed looks at how we achieve those numbers and the calculations behind it.

So moving forward, in terms of project background and methodology, we at IDC did research to explore the cost, value, and benefits that actual customers using Hyperscience achieved when looking to automate documents and application processing, basically. Through the course of the study, we interviewed six customers, and these customers had to have really, really robust knowledge about their use of Hyperscience across the organization. And they had to be, most importantly, able to quantify out their experience in terms of value and cost. From there, once we got those interviews completed, we built the ROI model, and that’s what I’m going to walk you through today. And we use the standard approach that I’ll give you more details on as we go through these slides.

In terms of who we spoke to, so let’s first talk about the firmographics, because like I said earlier, this is the foundation of the research. The organizations we spoke to were enterprise level, so they were larger in nature with nearly 58,000 employees on average. Average annual revenues were quite large too, at $54.4 billion. And what was interesting about this research is that we had a really global approach to it, and we had good representation across several countries in the study, as well as good industry representation across financial services, insurance, consumer goods, and utilities.

So organizations that really had a lot of complex workflows and process a lot of documents. In terms of the actual environment for Hyperscience, these organizations, and I want to note all the numbers you see here are aggregated across all of the study participants, so these are averages across the participants. They did a large amount of document processing per year to the tune of 3.24 million pages or documents per year. They also processed a lot of pages to the tune of 14.9 million pages per year on average. There was a lot of great findings, including that they had a lot of unique document types and layouts, so it’s showing more complexity at over 1,000 different types of layouts that they were using. And a lot of countries and languages supported with 27 and 14 respectively.

Just again, illustrating the complex nature of the organizations that we spoke to.

Xabi Ormazabal: Megan, I think that’s a really interesting data point. Out of all of these, I think the last one really jumps out at me. I know, for example, talking to some of our customers that do invoice processing, they often have a long tail of many countries, languages, and different currencies supported. Is there anything that jumps out to you about that sort of international feel or flavor that came across in these interviews?

Megan Szurley: I think the complexity really jumped out at me, right? So it showed that Hyperscience, and in the interview specifically, Hyperscience can handle a lot of different types of documents and a lot of different types of languages. And we’re going to see that later on in the slides in the actual findings. But in terms of the actual environment, you can see just the sheer amount that’s being covered by Hyperscience or being automated with Hyperscience shows that these are complex environments that are being supported.

Xabi Ormazabal: That’s great. Right. Thank you.

Megan Szurley: So moving forward, right out of our customers and all that we spoke to, these are the five key ways that Hyperscience created value for them. First, not surprisingly, workforce productivity and labor reallocation. Staff that was really tasked with processing documents and going through the manual workloads originally really benefited from the automation that Hyperscience provided to them. It really reduced what they called mundane tasks across the board for these organizations or these folks working in there. And most importantly, it helped them scale up with organizational growth so that they didn’t really need to necessarily hire additional staff, but they actually were able to scale their staff to support more processing. Accelerated processing cycle times was also a very big benefit for the organizations that we spoke to.

They really noticed quite specifically what used to take hours or minutes to do now took seconds. And that’s because of the automation, again, and I’m going to sound repetitive here, that Hyperscience provided to the processing workflows. Higher accuracy was a really big benefit, less rework, very few errors. And they really expressed that they really valued that. It wasn’t just across the traditional high structure types of forms or documents, but Hyperscience really handled well for these organizations’ handwritten and semi-structured documents as well. And that was a big value driver. It also went down to really helping with compliance, and they felt like they were really audit-ready. And that’s because they gained traceability that they simply did not have before.

And they got audit trail, so they could really demonstrate with better confidence that they are being compliant and able to respond to regulatory needs. And lastly, kind of echoing what we talked about with the first value driver, scalability without complexity. The organizations that we spoke to really stressed that Hyperscience was easy to work with and apply to their workflows. This really helped them take on a greater volume of documents without necessarily growing headcount as a result. And you can just see financial services customers really echoed that, right? And they really expressed, and I think the quotes just add greater value because they really expressed that Hyperscience can understand handwriting better than actual humans. And that’s just echoing what you’re seeing excuse me, in the actual value drivers.

What does this mean in terms of bottom line? The average annual benefits per organization, again, we aggregate all of the findings across the organizations that we spoke to, was $8.64 million. I normalize that so you, somebody listening to this webinar, can really express this across what they might be able to achieve. So if you take $58,000 per 100,000 pages run with Hyperscience, you’ll be able to see the type of benefits you might be able to achieve based on your expected number of pages that you expect to run through Hyperscience. This average annual benefit was really broken out in three buckets, I would call them.

First being claims and document processing bucket, the bucket of benefits. And this was really a significant area and probably not a terribly surprising area. It echoes back to that first value driver that I spoke about. This really showcases how gaining productivity really helped in terms of value of staff time. Business enablement was an interesting area that we were able to quantify, and it showed that trickle-down effect.

So $3.6 million in additional business value or business enablement came from them being able to take on more work, happier customers, and just be able to process work quicker. That led to revenue down the line. And lastly, IT benefits, and this goes to the simplicity aspect of those value drivers, right? Our IT teams that are tasked with really managing Hyperscience or Hyperscience workloads, models, whatnot, they found it easier and less complex to manage than they did in the past or their past methodology. And as a result, it amounted to about $1.2 million in benefits.

So really large-scale benefits that expresses the value that the organizations achieved. And from here on out, I’m going to walk you through each bucket so you understand what led to these calculations. The first one we want to talk about is our smallest bucket, that orange bucket, because it can’t be ignored. And that’s our IT admin and management efficiency gain.

So these organizations really had a significant gain here in terms of efficiency, 73%. And they really related that back to being able to train Hyperscience quickly. It was less complex, as I said earlier, and they could manage larger volumes of documents as a result without having to be very specialized in machine learning expertise, importantly. So this meant that they needed about 13.5 less full-time employees to manage the workload of Hyperscience than previously they would’ve needed, again, 13.5 more people to handle that volume that they’re handling with Hyperscience. Factoring in our standard assumption at IDC that IT folks make about $100,000 in loaded salaries, basically $1.35 million was attributed to just this efficiency gain alone.

So you can really see how that orange bucket started to build. Additionally, what we found is that Hyperscience also helped these individual teams scale, too. So because these organizations didn’t need that expertise anymore, their current staff could handle managing Hyperscience. And they really found that they avoided hiring 1.4 machine learning specialists as a result.

So that also added into that orange bucket that we just spoke about. And again, just another insurance quote, just really demonstrating that the learning curve is just much smaller with Hyperscience. So giving more flavor behind the numbers that you’re seeing on this slide. The next area, which I think we all expected to see, was how claims and document processing teams really gained productivity. And they gained significant productivity to the tune of 52%, and that’s really because very manual processes are becoming handled by Hyperscience, and it really helped this team scale.

So teams of 117.2 full-time employees could now work at their equivalent productivity level as if they had 60.4 additional full-time employees on staff. So that’s that 52% productivity gain. We valued that at $4.2 million in staff time per year. And you can really see why this is becoming our biggest bucket of benefits that I showed you earlier. Not only that, just further showing the scalability of Hyperscience, these folks could really work and handle as the business grew, and they didn’t need to hire 4.8 additional full-time employees as well.

So that further adds to that bottom-line number, right? And you can see just, again, this insurance customer gave us a really great quote illustrating that there’s peak times for them, and they don’t need temps anymore because their current workforce can just handle that labor.

Xabi Ormazabal: One thing here that stands out to me, Megan, I love this slide, in addition to all the other great analysis, is that it kind of hammers home the point how back office and document processing is not just a cost center, right? It really is like an untapped potential for upside, for driving more profit and revenue. What are your thoughts there in terms of any other tidbits or things that customers mentioned to you throughout the interview process?

Megan Szurley: Well, I think we’re going to get there, too, because we do see, and you saw that last bucket that we haven’t talked about yet, right? This drove business enablement. So the scalability of this team led them to take on more work, right? And more exciting work, too, right? So they could put more judgment into their work rather than just doing a lot of manual data entry-type tasks.

So that really had that effect that it worked all the way down the business because they could scale, they could take on more, they were working more intelligently, right? And it really impacted the business overall as well, which was a great finding.

Xabi Ormazabal: Yeah. So great to hear.

Megan Szurley: So moving forward, we’re going to start talking about KPIs or key performance indicators. We saw that both our IT team and our processors, right, they’re becoming more effective in how they work. And that’s largely because documents are more accurate, right? 89%. They’re also having much higher levels of automation for their documents and claims as a result of using Hyperscience in comparison to their previous methodologies to the tune of 214%.

So large impact here about how the automation is driving the gains that we just saw. But not only is just having better data helping individuals work better, it’s going down to the compliance and audit side of things, too. Better, stronger, more accurate data really helped our team, our audit and compliance folks, really improve how they worked. And you can see a lot of stats here, like 37% improvement in audit readiness, 43% increase in data governance and retention adherence, 36% reduction in compliance-related errors. All of these things help, right? Further document that they are being more compliant in how they work. They felt like they had 20% lower risk of fines and penalties as a result. They had more evidence, 60% more evidence, and 67% more improvement in just being able to see the transparency behind the model decision.

So overall, these organizations, not only did the data automation really help the folks working with it, but it really also became a compliance kind of lever as well for the organization. So that’s how we got to that other value driver that I spoke about earlier. This part I think you’re going to be excited about, and I touched on it with our productivity. This is really getting down to how did those productivity gains that we saw improve business, right? Overall business in terms of revenue. And we’ve got some really exciting stats right out of the customers’ mouths that we spoke to in terms of claims are getting processed 61% quicker. There’s 39% fewer processing errors, 50% faster claim and document completion. This led to higher customer satisfaction to the tune of 44%, as well as higher career satisfaction, 24%.

So not only was it an internal factor, but it also was an external factor. And when you have higher customer satisfaction, that starts leading to the bottom line of additional revenue, right? So very specifically, we asked the customers that we spoke to, “If you had to attribute any additional revenue back to your use of Hyperscience, how much would you attribute?” And they attributed $4.32 million in additional net revenue. And that was, again, really largely driven by the stats you see on this page, right? They had less customer churn, they had less error rates, less delays, happier customers. They could take on more work because, again, thinking back to our productivity, these teams could take on more work because they could scale with business growth.

That led to additional net revenue for the organizations that we spoke to. What did you say? I can’t hear you, Xabi.

Xabi Ormazabal: I was on mute. I said I love those intangibles, what we think of intangibles, that you guys were able to capture here in terms of happier employees and happier customers, which is really a great thing to see come through as well.

Megan Szurley: Absolutely. And then just one last slide to walk you through, and this is really the summary, right? This is showcasing that overall factoring in what we just walked through. So IT efficiency, processing productivity, better accuracy, improved business enablement. It factored into us calculating a 615% three-year ROI with a 7.3-month payback. How did we get there?

So first, we applied a 12% discount factor against both benefits and investment. When you apply that just to factor in kind of time value of money, right? It came out to 20.2, nearly $20.3 million in benefits per year or over three years, rather, and a discounted investment of $2.8 million over three years. That represented and amounted to a net present value of $17.4 million, and of course, that strong 615% ROI that I explained earlier. And our financial services customer really summed it up well and I think echoes the sentiment that we just saw through this presentation. We could have done automating without Hyperscience, but it would’ve taken twice as long and needed twice as much effort. Thank you. Jorge, over to you.

Jorge Peña: Very well. Well, all I can say is results like this make me feel like a proud parent. I spend most of my time worrying about how do we take wonderful technology, competent technology, and the efforts of all our engineers and turn them into something that customers see as a successful investment. For those of you who have not met me, my name is Jorge Peña. I lead value engineering at Hyperscience, and what that actually means is that I spend my time considering what is required for our technology to be an effective instrument in your ability to get work done.

So for what you’ve seen from IDC, what makes us very proud is the fact that these are results that customer actually obtained from using our technology, and we use that to motivate and to inform our ability to create better and better products. My focus, and my focus today, is to look at the opportunities for you to look into the future, to see how you, by considering the use of Hyperscience, would obtain similar results.

But I want to do upfront is, because most of you are new to this, but there’s a significant part of the audience that has also been with us for a number of years, and I want you to stick around for the duration of my conversation, is this applies to you, too. Whether you’re beginning the journey or you’re somewhere in the middle of the journey, there’s always an opportunity for us to work together to articulate, define how is it possible for you to obtain results to the ones we’ve just seen.

So we have a way of thinking about how to construct this journey, and let me start with that. The first element of our point of view is that everything starts with a set of goals. These goals get articulated in the form of, I want to do something faster, I want to do it better, I want to do it by consuming less capacity, less resources, maybe less and simpler technology, with the objective of deriving the type of benefits you just saw today. Because all of these are elements that get transformed into what is called work to be done. And work to be done illustrated here in the form of a multitude of endeavors. I want to be able to process more of something. And if I process more of something, as you have seen, that can lead to, if I do more, I derive more revenue.

If I do it faster, my customers are happier. If I do it with less capacity, then my profit margins go up. So the intent is, by us working together, we get to define how the work gets done now in context of what you can see here, which is very process-centric because you have a methodology to do this work. Now we just want to help you do it better, faster, and with less capacity.

So let’s take a simple example like a mortgage application. What is the work to be done here? And let’s consider the context in which that work needs to be done. You are going to have dedicated resources to do the work. There are several categories of resources that do this work. There are those that are the knowledge worker, let’s say the underwriter or the claims agent. These are individuals that have years of experience, that know exactly what to do. What they want is give me the right information, give me the right data, give me the correct data. Give it to me at the right moment in time for me to do my part of the work, which is make a good decision. Make a decision that is favorable to the enterprise I work for, and is favorable for those who are my customers.

For that, we use technology. There’s not a single case, any customer that we work with, where there’s only one stack of technology. It’s a multitude of technology. It means that the people have to interact with this technology and be efficient and proficient with that technology. Our contribution to that is that we take care of the human side of the work, which is a document shows up, documents were created by humans for human consumption. They were never designed for machine to take care of them.

And now humans are burdened with the fact that there’s a bunch of other technology that requires them to manage these documents, understand what they are, make sure that they are the right ones, the required ones, extract the information out of those documents and put them in the right place for the rest of the systems to follow through and be able to do the right work, adjudicate properly, faster, obviously, and with less capacity.

So let’s continue that conversation. So now that we’re doing the work, one measurement for doing the work with humans, machines, any endeavor, is that you are concerned with the throughput of the work, meaning am I getting done what am I supposed to getting done in the amount of time I’ve been given to do it, and correctly, so that when I finish the work, that work is useful.

So let’s take examples about getting work done. What is entailed here when it comes to the participation of Hyperscience and then the frame of reference of Hyperscience, which is I have documents. These documents have a purpose. I have to understand if the document is correct. Is it valid? Do I have it? And then more difficult then is now let me extract the information out of the document in order for me to follow the process. This is a, most cases, a non-value-add work. It’s work that I need to do because I have no other way of doing it, but it’s not essential for me to do it, hence it’s a burden. Burden translates into friction. Friction is what creates the difficulties and the deficiencies in any given process.

So let me explain what we mean by that. So let’s say I’m assessing a loan eligibility, and let’s assume that this is a loan for a more difficult case, buying a home. You know that all of us have experienced that journey. It’s a difficult journey from the sense that, one, it’s stressful from the sense of the participant. I want to get the loan. I want to get a good loan. I want to get a good rate, and I want to get it- Now, because if I don’t get it now, I may lose the opportunity to buy that home, especially in competitive markets like today.

So the idea is someone is going to ask me to submit a bunch of information. I have to submit proof of income. That comes in a multitude of forms. I have to submit identity to prove that I am who I am. Then, from the side of other participants, they have to provide the assessment of the property, the value of the property, any difficulties with the property in terms of, for the adjudicator, to make sure that they do the work properly and assign a loan as it’s supposed to be.

So guess what happens? A ton of stuff goes wrong. When things go wrong, that becomes rework. Rework is the negative effect of a process that encounters difficulties. Difficulties are generated, in this case, in our frame of reference, by the existence and the requirements of these documents, the nature of these documents. And as we saw before, these documents contain a tremendous amount of their own personalities, meaning they come in different forms. The data is all over the page, so there’s no standardized way of, for example, for a bank statement. It could come in different language. It could come with artifacts that are handwritten.

So what happens? I’m the person deciding to give you a loan or not, and I want to give you the right loan. Every time I encounter a difficulty with a document of that type, and I have no mechanism to do it any better, that constitutes a high probability of rework that then becomes additional work for me to go do. So what happens then? Let’s say that I have missing information.

Let’s say that I translate income improperly or that my dates are not consistent. You’re giving me a reference for income that is not within the period that my process requires. That means that the process gets delayed. That means that customer satisfaction goes down. That means that I am not getting done the volume of work that I was supposed to get done that day, which means that my queue of work to be done begins to increase. My stress increases. The general sense here is that as all that is going on, and as I try to resolve some cases, I get through, I do everything correctly. That work becomes the work that actually gets done.

That’s the only time that I actually derive a benefit. So when you saw the conversations prior, when we say we improve productivity, productivity is measured not on how much effort I put in the day. It’s measured against what useful work did I actually get done in that day. That means how many loans did I approve? How many loans did I deny, which is not a good measure for anyone.

But how many good loans did I approve? That becomes my additive level of risk. If I stress the system, I have no mechanism other than hurry up, try to do my best work, and hope I do not make mistakes. So that’s when outcomes manifest themselves, either positive or negative. And what we’re trying to do here is introduce the idea that the outcomes that you desire are the good ones. That I do good work, I do it faster, and I do it with fewer resources, or allowing those resources to do more of that work, reduce my queue, increase my customer satisfaction, and so on.

So that’s where Hyperscience comes in. We’re a machine that addresses those particular elements of difficulty, those particular elements of friction. What we try to do, the intent of the technology, what makes me happy about the results that we observed is that the technology actually works. And when it works, it leads to good results. There’s evidence that when we architect it properly, ingest it properly into your current way of working, it allows you to improve the outcomes. And the evidence is clear. It does do it.

So we want, obviously, an improvement in outcomes measured in terms of increased productivity, more work getting done by the same number of people or the same amount of work done by fewer people, or the quality of that work increases without increasing the cost of doing it well. And one of the most valuable ones is that the people doing the work derive a high level of satisfaction from doing good work, doing the right work, helping people get what they need, especially in cases where you do benefit adjudication. And the other one is that it’s good for the business. And also very important is that you begin to reduce all the work that you really do not want to do because you did not do it well the first time.

So let me show you a few examples. The way we do this is that we need to define a baseline. A baseline that says, I need to know where you are today so that I can project, forecast, simulate where you could be in the future if you’re a new customer. If you are a current customer and you have results like this, what we’re interested in seeing is can we optimize? If you are obtaining better, more positive outcomes, how do we help you optimize? How do we help you, for example, work on a different workload, on a different revenue stream that has the same type of burden associated with the fact that you have to deal with documents?

So a simple metric like how much do you get done in a day? Is that the right number? For example, someone could say, “Well, it’s great that I can do 40 of these a day.” Well, is that what you want? Is that what drives your business? Could you be doing 50? What if you did 60? How much does it cost you to do it? That’s an important factor because the economics of it are if I do more and I do them at the same cost, can I scale that cost? Can I afford that cost?

And then obviously, how long does it take me to do it? So if the answer is I could do it in half the time and double the cost, that may not be the right solution for you. And then obviously, there’s a daily processing cost, which is the aggregate of all these things. So now imagine a better future, and one where not only do I do more, I do that more for less cost. I reduce the amount of time it takes me to do it, so now I’m satisfying a multitude of dimensions in terms of goodness. I’ve increased my volume. I increase my volume, I increase my revenue. If I increase my revenue, and at the same time I reduce my cost, I increase my profit. And if I decrease the time that it takes me to do it, my customers are happier. My customer satisfaction goes up.

If I reduce the burden on the individuals doing the work, that means that they have more time to do better work. And if overall I reduce the cost of doing everything, then everybody wins. And that is the objective of value engineering. Right?

Xabi Ormazabal: So one thing there, Jorge, if you can go back to the previous slide just for a second. I really liked what you’re showing here because it’s easy to digest on two levels. Obviously, there’s a dollars and cents impact, but I think we can all respond or understand getting more things done in the same amount of time or spending less time per processing of a particular unit.

So I really loved kind of the message here around AI kind of being a force multiplier for your customers or your efforts. What are your takeaways here in terms of how does this resonate when you’re working with customers around that sort of doing more with less or with the same or doing it in less time?

Jorge Peña: The best part of this conversation is that by having defined a current state baseline, we have very clear indicators of the needle moving up or down and to what degree the needle moves. So customers respond very positively to the fact that we’re not simply stating degrees of magnitude. I’m getting faster. I’m getting better. I think I’m doing better work. What we prefer is objective observations of the deltas. If it’s better, how much better is it? In this case, if before it was 40, and now I’m 41, is that good enough, right?

Xabi Ormazabal: Yeah.

Jorge Peña: So yes, the idea of value engineering, when we’re proposing we do collaboratively, is that we establish the most important part, a frame of reference. And secondary is the way we get to the second stage, which is the after. It has a degree of magnitude because now we know what to shoot for, and if we get there, then we know that we did it correctly. If we do not know a baseline, then any number that we pose is just that, it’s a number without a frame of reference.

Xabi Ormazabal: Yeah. Makes sense. So I hope that answers that question.

Jorge Peña: Yeah, definitely. Now, how do we get there? How do we do it? What is our preferred approach? What do we recommend you do? The reason why I’m here is I would love to engage you in doing this because one, it’s a very useful exercise, and second, it’s a fun exercise. And more important, it is a very rewarding exercise when you get to the end of the journey and you observe what you’ve done and you say, “Look, we actually did something that was useful. We did it well. We did it in less time, and we can continue to optimize it.” So here’s our proposition for how Hyperscience proposes you engage with us, those who are new to us and those who are current customers. There’s never a perfect time to do this. There’s always a time to do it, and the best time to do it is to do it now.

So the idea is we start with the first steps, which is we need to frame the context of what we are going to look at, the work that needs to be done, the processes involved in doing that work, the people involved in doing that work, the current metrics of that process, and obviously the current existence of systems or IT implications for us to have a frame of reference for where Hyperscience would be injected.

So it’s both a business analysis, it’s a process analysis as a definition of objective KPIs and a technical frame of reference for us to be certain that not only can we activate the work, but we have an opportunity for IT to embed it into their technology stack. Then once we have that defined, then we go into the actual fun part, which is how do we construct it? How do we architect? How do we help you understand all the possibilities, all the scenarios? The scenario is as simple as are we going to process one type of document? Are we going to process 100 types of documents? Are we going to process a region, a multitude of regions? Are we going to process one type of work stream or a multitude of work streams? All of that gets done in the following stages.

Now, what we build in that is an artifact, is a model, is a simulation environment where you can play what-if scenarios to your heart’s content until you’re satisfied that you’ve landed on something that, one, is reasonable within the timeframe and affordability. Second, achievable, that we’re not over-predicting what is possible. And third, that you can defend, that you can justify, that you feel comfortable taking to decision-makers and saying, “We’re basing our decision to engage Hyperscience, implement Hyperscience, because we have high confidence that if we do it in this manner, in the way we have jointly architected, we will get there.” The construction of the business case, think of it as a necessary artifact. This is where I would say the rubber meets the road.

We’ve done all the definition of what the problem looks like, its dimension and possible solutions. We have architected what we believe to be the constructs to get us there. And now comes the thing is, well, how much is it worth? How much is it going to cost me? How long it’s going to take for me to see and observe those benefits? What do I need to do internally from a change management perspective to adapt and properly utilize the technology? It all comes in here, and this is where the financials either make sense or do not.

And then obviously, once you’ve decided to do it and we follow the path that we have engineered, then what you observe today is it’s real. The machine does do the work that it’s supposed to do. If we do all the steps prior to here, it’s guaranteed that those results will show up. They will manifest themselves because the technology is that competent, and by architecting it this way, by engineering it this way, we make sure that a good, competent piece of technology is also a useful business instrument. The last aspect of all of this is what? This is grounded on evidence. The exercise of value engineering is to be able to articulate on the basis of evidence the path for you to obtain equivalent results.

So evidence means we know that it works. So we can promise you that if we jointly engineer this for value, which is an interesting perspective, engineering it for a desired outcome, not engineering it for the sake of the machine proving that it can do things at 99%, is to convert that 99% accuracy into a useful outcome. Then the most important part is does it actually work?

So part of the engagement is we get to test if the machine actually works on your documents. We know that it works on the documents we’ve shown you so far, but your documents may have a different nature. The volume of your documents certainly impacts the desired outcomes. The variability of those documents certainly impacts the machine’s ability to do it well, consistently well.

And then obviously, if we do all of that, the next one is do the numbers make sense? Do the finances, does the economics make sense? Because you could continue to do this as an alternative to using technology and be better off if you were to use technology. So the point here is not to assume based on industry numbers or entitlements that are derived from assessments, is to actually validate it with your documents against your processes, against your desired KPIs.

And then the last element, as we have discussed, the most important, the one that keeps me interested in doing this and the one we derive the most, I would say, the most satisfaction and the most pride is the fact that it actually works, and for you to observe that it actually works so that you can then continue to optimize what you do on a daily basis.

And then back to you, Xabi.

Xabi Ormazabal: Thanks so much, Jorge. That was a great overview of our value engineering capabilities and that effort led by Jorge here at Hyperscience. What I wanted to leave you all with today is a couple of great resources. So we have two QR codes on this slide. You could scan them and actually download that wonderful IDC ROI report that Megan and Andrew discussed with a 615% ROI over three years, and a lot of details in how those insights were generated.

And then the second QR code on the right is to be able to schedule a value engineering workshop with Jorge to really help you prioritize that starting point for your business case, get that document-level roadmap, and quantify your savings and your capacity estimates. Also, not just on the slide, you have it in your resource centers while here on On24. You should be able to see both of these QR codes. We also have a companion piece of build versus buy white paper that we published last year that helps a lot of organizations work through the logic around does it make sense to buy a platform or to actually build it myself?

So a lot of interesting considerations in that piece of research as well. So, before we close out, we have a lot of questions that came in through here. I just want to kind of go through some of these that might be of interest to the audience overall. So the first one, I’d say we’ll start with Andrew, and maybe Jorge, you can chime in as well. One question about the IDC ROI study. The results are based on six organizations. How confident should other companies be that they would see similar outcomes?

Andrew Gens: Great question. Yeah. Well, I think if you have a high volume of unstructured or semi-structured documents, and you’re willing to put in the work as a customer consuming this IDP solution, I think you can have a high degree of confidence that you can see similar outcomes. Really, there’s a lot of onus on you, the customer, to put in some effort on the front end, and also once the solution is implemented, in order to make the most value out of your solution adoption. And so that involves things like Jorge was mentioning, understanding your workflows ahead of time, really tapping into those process experts within your own business, those people that understand these workflows best, and understanding where they think improvement is needed. What are your most high important docs where you can get the most bang for your buck very quickly?

And then, also leveraging process experts with your vendors, such as Jorge and the value engineering team at Hyperscience, where they can come with their industry experience, their solution experience, and inform you best on where they think the value can most easily be derived most quickly and where they can expand from there. And one last thing I think is also to be mindful of the fact that the value is not just contained to these individual workflows. Ideally, if you’re implementing IDP correctly, if you’re cutting down those silos between organizations within your company, you can ultimately leverage these documents, the data that you get structured in order to create a more centralized source of knowledge for your organization as a whole, that won’t only benefit that individual workflow or those individual workflows that were optimized, but will also feed into generative AI processes throughout the enterprise and make all the efforts across divisions and across use cases within your business more efficient by leveraging vetted and easily accessible information.

Jorge Peña: If I can add to that is every process, it doesn’t matter if it’s a natural process, a man-made process, is always capable of being optimized. Do we have confidence in these results? Yes. We always have confidence in the fact that if the technology does what it’s supposed to do in the way that it’s supposed to do it, it will improve a process. In order to attain the results, you have to first know how much better you want to be.

So the definition of goodness is a relative thing, is how did I do it before, and how much better do I need to do it to say that I obtained the results that I wanted? Think of it as anything else, improving your ability to run faster. So you put the work in, there’s a methodology for you to run faster. There’s certain things you need to do in order to become a faster runner, and you get to find that if you follow all of that, there’s a high degree of probability that you will be able to run faster.

So we have the same mindset when it comes to this type of technology, is that we know processes burdened by documents, and the fact that the nature of these documents is tremendously variable and introduces a significant amount of burden, that removing all of that, by default, will have a almost guaranteed improvement.

Xabi Ormazabal: Makes sense. Another question that came in here, this is a great one for you, Andrew, to give your vision. Where do you see IDP heading over the next few years, and what’s driving organizations to invest now?

Andrew Gens: Yeah. I think there are a lot of improvements we can anticipate in the next several years. As these models become better, as the ability to select the right model or the right AI tool or machine learning tool for the right workflow improves, I think we’re going to see an increase in accuracy, which has been happening over the last several years and will continue to be the case, and also see a reduction in cost. These sorts of things also involve an increase in specificity to your own business.

So the ability to incorporate a custom ontology, for example, leveraging the information from your own business to refine a model that’s being used for intelligent document processing to your own scenario, whether that be industry, a use case, et cetera. Making that information more pertinent and more applicable, increasing that accuracy, and then through that ModelOps perspective, decreasing the cost per page as we better figure out how to leverage these models effectively over long periods of time with high volumes of documents.

Jorge Peña: That’s great. Yeah, and I’ll chime in on this one a little bit as well. One thing I’m seeing a lot from our customers, for example, at Hyperscience, is how AI for intelligent document processing is definitely moving away from kind of a monolithic conception of you have models, you throw documents at them, you get outputs. But into kind of a richly textured approach where you have a number of different types of models and techniques working in concert.

So we talk about the concept of inference layering and how you bring in GPU models and CPU models, how you might have things on-prem, how you might have things using frontier models. So something that we’re seeing a lot is in even optimizing for GPU performance for some of these models. So we obviously didn’t get into a lot of technical details on this webinar today, but that’s an exciting frontier for us that we’re seeing a lot of interest from customers. Another-

Andrew Gens: If I could actually butt back in very quickly, Xabi.

Xabi Ormazabal: Go for it.

Andrew Gens: In terms of another element of that question, what’s driving organizations to invest now? I think there’s another element that I touched on briefly in the first question that’s worth repeating, which is that as generative AI and agentic AI continue to permeate throughout our businesses, it’s very important to lay the foundation for solid, reasonable, grounded AI that’s going to produce high-quality, high-accuracy results, whether that be for an IDP workflow or generative AI or agentic AI usage throughout the business. And a key element of that is going to be making sure you have the right data and the right data in the right places. And so getting in with an IDP solution now to begin unlocking the value that’s hidden behind these unstructured, semi-structured documents, is a big part of those early steps in really making your AI initiatives more effective in the long term.

So you don’t have to play catch up by trying to unlock that data on the fly once you’ve already adopted several solutions and you’re trying to figure out how to make them more effective in your business.

Xabi Ormazabal: Yeah, great point. Maybe we have time for one more. There’s another great question that came through here, for Andrew, and then great to have Jorge chime in as well. In your opinion, what separates the organizations that get outsized ROI from intelligent document processing from the ones that see more modest returns?

Andrew Gens: Yeah. Another great question. I think some of that goes back to the answer I gave to the first question. Again, preparedness and an understanding of your own processes plays a very large role. If you’re trying to tackle some of the hardest processes first, that may not even be the ones that will present you with the most value. You may find yourself falling short. I think it’s understanding which processes you want to target, have an effective plan, an efficient plan of how you want to spread these IDP workflows throughout your business, and then having an understanding of how you can break down silos between business units in your organization to actually leverage the results of an IDP solution across your enterprise and not just have that information or the benefit be locked in one area of your company.

Jorge Peña: Yes. And to add to that, a complement to that is the ones that derive the most amount of benefit and value of the technology are the ones that are focused on the work that needs to be done and how the work gets done better, and not a focus of how do I get processing one single type of document better than I did before. It’s not about the documents. It’s about the work required because you have documents. It’s a contribution of documents to how you do your work. And if you include the complete stack of documents inside a process, because what you worry about is the effectiveness of the process, you derived significantly more benefits, whether it’s by productivity, error reduction, speed, cost, everything comes with it.

Xabi Ormazabal: That’s great. Thank you so much for that, Andrew and Jorge. I’d just like to thank all of our presenters, especially Andrew and Meagan, for taking the time to join us today for this presentation and showcase these amazing outcomes from this IDC ROI study. Also, Jorge, your expertise in value engineering and a lot of the different conversations you’ve been driving with customers and prospects. Like I mentioned, lots of resources for everyone in ON24, the whitepaper itself, book a consultation with Jorge to go deeper into value engineering and what it means for you and your organization with IDP. Thanks all for your time today. Have a great rest of your day.