The Total Economic Impact of Hyperscience

Jorge Ruiz: Hello everyone, and welcome to this webinar. My name is Jorge Ruiz, and I lead product marketing at Hyperscience. I’m going to be hosting this webinar, and with me, I also have Lauren Hofmann from Forrester who is going to be leading most of the content. I just wanted to do a quick introduction on why we decided to commission this report to Forrester.

Jorge Ruiz: At Hyperscience, we know that unlocking the value trapped in structured, semi-structured, and unstructured documents and making it available to the whole organization can have a direct positive impact on your bottom and top line and also in your customer and employee satisfaction. But sometimes it can be difficult to estimate this impact. When you invest in intelligent automation and digital transformation, you need to know you are really getting your money’s worth. That’s why at the end of last year, we decided to commission Forrester Consulting to conduct this total economic impact study to examine the potential ROI of the Hyperscience platform. Remember that you’ve got a Q&A panel on the right-hand side, so you can submit your questions and at the end of the session, we’ll try to cover as many as possible. With that, Lauren, it’s all yours.

Lauren Hofmann: Thank you, Jorge. Thank you all for joining today. As Jorge mentioned, my name is Lauren Hofmann, and I’m a total economic impact consultant with Forrester. As noted, Hyperscience reached out to Forrester to create a study to help technology investment decision makers understand how to value the impact of their solution. The scope of our conversation today is to first identify what a TEI study is and the Forrester methodology used to create these studies. Next would be to cover the high-level results of the study via an executive summary. Thirdly, understand the journey that a typical Hyperscience customer takes prior to investing in the solution. We’ll spend some time covering the key challenges in their prior state, then walking through the results of the study in more detail. And finally, providing an overall picture of the financial summary.

Lauren Hofmann: TEI stands for total economic impact. It’s the methodology that Forrester utilizes to value what impact means in the context of this report. There are four key components. The first is benefits, second is cost. Third is flexibility, which Forrester defines as the ability to do something more efficient and less costly in the future based on the prior investment in the technology. And finally, risk.

Lauren Hofmann: Let’s take a closer look at how Forrester approached developing the study. As you can see on the graph, we follow a pretty linear process. Starting on the left-hand side under due diligence, this contains both primary and secondary research. The primary research involves Forrester speaking with internal stakeholders at Hyperscience. These individuals sit in a variety of roles including implementation, product, marketing, development, sales, and so on. The goal is to gain a better understanding of how Hyperscience is positioned in the market, what the implementation process is like, and really understand more of the nuance around the solution itself.

Lauren Hofmann: In addition, we perform secondary research by reviewing documentation specific to the Hyperscience solution. We look at product positioning documents, specifications, pricing sheets, collateral. All of the information gathered in this step serves as input for the customer interview guide that we utilize in discussion with an actual Hyperscience customer. So that segues into the second step, the decision maker interview. This is arguably the most important step of the process. It’s a customer interview with an individual that has decision-making authority to be able to invest in the solution. During this discussion with the customer, our aim at Forrester is to be able to source quantifiable information. That could look like absolute numbers, percentage change, small ranges, about the benefits that the customer realized and the costs that they incurred. Our goal is to utilize this information to serve as input for the financial model we create.

Lauren Hofmann: This flows into the next step, which is drafting the financial model framework. The goal of this step is to take all of the actual customer information about the benefits they’ve realized, the costs they’ve incurred, and put those into the financial model. It’s important to note that while I was the consultant that generated the financial model based on my logic, understanding from the customer interview, from the due diligence phase, it’s put through a very rigorous review process with a designated internal model reviewer at Forrester who looks for standardization across the way we model certain benefits. In addition, the Forrester analyst who’s an expert in this space also reviewed this model, the draft of the study, and so on.

Lauren Hofmann: Once Forrester developed the draft model, it’s shared back with Hyperscience and is also shared ultimately with the customer to validate that it reflects their story. One of the most important takeaways from a single customer study, meaning only one customer was interviewed, is that while it’s based on the actual experiences of the customer, it cannot always 100% align with the way the customer might assess the value of the investment. For instance, there could certainly be quantifiable benefit that your own organization would realize that you might not see in the TEI study. I’d encourage you when you read through the study to think of this as a framework. You’re going to be putting in your own numbers, your own assumptions into this framework to help you make a better determination about the value of the Hyperscience solution to your own organization.

Lauren Hofmann: Shifting one step right, the final step is the case study development or release. Once the draft version of the model is accepted, we begin drafting the study itself. Ultimately that draft is shared with the Forrester analyst, with the customer that we interviewed, as well as the client Hyperscience. We do source feedback from the client, Hyperscience. It’s mainly focused on accurate descriptions of the solution. We don’t accept feedback or change requests that obscure the meaning of our study.

Lauren Hofmann: A few disclosures just to reiterate key points. A reminder that this TEI study was commissioned by Hyperscience. Forrester is not making assumptions about the potential ROI that other organizations would realize. It is based on a single customer’s experience. The value of what we’re providing is a framework to help make that valuation. Forrester maintains editorial control over the documents. And lastly, we are not making any sort of endorsement for the technology.

Lauren Hofmann: As we shift and begin to dig into the results of the study, I just wanted to take a moment to reflect on a quote that was shared within the study. As you can see on the screen in front of you, “I’m capturing three times as many fields, but spending 95% less time.” I tease this out because this really captured a pervasive theme throughout the study, which was greater inefficiency through automation.

Lauren Hofmann: Let’s take a look at the three-year impact against Forrester’s modeling. Just let’s just touch on the overall results of the study. Forrester had found based on this interviewed customer’s experience, roughly 126% ROI, benefits of 5.2 million, and a net present value of those benefits which equates to 2.9 million. As we continue to dig more into the results of the study, it’s important to understand what were the characteristics of this organization that was interviewed. They came from the financial services industry, a quite large organization, 50,000 employees. They had a multinational footprint though headquartered in the United States, and they’d been in live production with Hyperscience for five years.

Lauren Hofmann: Building context for the customer’s experience and why they chose to invest in Hyperscience really helps understanding the key challenges they faced in their prior environment. One key challenge they faced was the ability to classify and extract semi-structured and unstructured data. The legacy tool that they had in place could really only reliably process structured data. Even then, it was a very manually intensive process. Another important component of their story was they were looking for an enterprise-wide approach to data classification and extraction. They really wanted to undergo a digital transformation across the enterprise and standardize the way operationally that these things were done. One of the criteria in looking for a potential solution was, can you provide an enterprise-level solution, this degree of standardization that we’re looking for?

Jorge Ruiz: Those challenges are pretty much what we hear across the board. One of the key challenges that we continuously hear is this problem around accuracy. We’ve got customers that have already started in their automation journey. They have already applied some sort of technology to automate document-centric processes, or they might have even created their own system. The problem there that we see is that they often treat accuracy as an afterthought. Many vendors go with this full automation approach that fails to account for human involvement and overlooks the importance of data quality. We need to bear in mind that one single incorrect digit can be the difference between a claim or a loan application being approved or rejected. That means a very unhappy customer. Companies have to either set up a whole team to double check the output that comes from the extraction tool, or literally just sacrifice data quality and deal with the consequences, which are going to include inefficient processes, delays, and very unhappy customers.

Jorge Ruiz: We also hear common challenges like not being able to process handwritten content or low-quality images. It is so common to take mobile captures of documents and send it to our vendors. They have to deal with very low quality, poor images that are completely skewed or have very difficult angles. Also not being able to scale to a hundred millions of documents that normally come during peak times. It’s hard to predict the volumes that we are going to have. Many companies really struggle when they have these seasonal peaks and suddenly they have to hire a ton of temporary staff to deal with those massive surges in volumes. Lastly, many of these legacy solutions that customers have been applying traditionally to deal with their document challenges are not able to handle significant document variation. Just think about process invoices, for example. Every single supplier is likely to use slightly different variants of the same invoice. Many of these tools take one or even two months to set up and train properly.

Jorge Ruiz: I will be curious as well to see if these challenges resonate with the audience here today. So we’ve got a quick poll. What out of these three challenges do you see as common in your organization when automating document processing? Is it fragmented processes and capabilities across organization? Is it operational inefficiencies due to manual work needed? Or is it limitations in the current tools or solutions that you are using to extract mainly semi-structure and unstructured data?

Jorge Ruiz: This is not really surprising. It aligns pretty well with what we see across many of our customers. The amount of manual work that is still needed as part of many of these critical processes is truly delaying many companies. Now that everyone is trying to go towards digital transformation and improve their operations, that’s seriously hindering their ability to do so. Do you have any comments on this, Lauren? Does this align with what you guys saw during the study?

Lauren Hofmann: Definitely. All three options obviously are challenges that the customer articulated. And certainly this operational inefficiency due to heavy manual lift was definitely a theme. The quote on the slide says, “The third party consulting firm too was impressed with Hyperscience’s performance and capabilities compared to the other tools being evaluated.” So what I’m communicating here is a point that the customer had a pretty rigorous evaluation process that ultimately led them to select Hyperscience. They did hire a third party firm that specialized in this particular field of data extraction and processing. Along the way, the customer considered several leading providers in this space. They did a comparison, a full list of features and functionalities, in an attempt to address the challenges they faced before ultimately settling on Hyperscience. The customer was able to launch a pilot with Hyperscience, and this ultimately solidified the trust in the solution and the direction that they wanted to go in.

Lauren Hofmann: I wanted to speak briefly on the most notable or profound benefit that we found, which certainly ties to the key challenge around the operational inefficiencies. What you’re seeing in front of you is the primary benefit of the study with the $4.1 million present value over a three-year period. It’s this overarching productivity through increased automation capabilities. The customer was able to achieve this degree of productivity out of the box with little configuration. Ultimately this degree of productivity correlates to other benefits around avoidance of hiring new staff members to support operations on an ongoing basis as well as seasonal peaks. Something that is a part of the story that isn’t explicitly modeled financially is that while the productivity is increasing due to the automation capabilities that Hyperscience provides, the volume of data being able to be extracted is also increasing. You’ll see a small quote on the screen just noting they’ve been seeing consistently about 95% machine transcription or automation.

Lauren Hofmann: The next slide is providing an overview of the other benefits that were included from a financial standpoint or cost savings standpoint. The second bar to the right with a value of $902,000 in savings over a three-year period in present value is the cost savings realized through decommissioning their legacy solution. The third bar to the right with a savings of $90,000 over a three-year period in present value is the cost savings experienced through hiring avoidance. And finally, the fourth bar to the right with a value of $73,000 in savings over a three-year period in present value is the reduction in cost associated with labor to train employees on Hyperscience on an ongoing basis.

Jorge Ruiz: We found this very interesting. Obviously, the servicing productivity are probably the most obvious component because it’s fairly easy to measure for most organizations. But what we have also seen from all the customers is that once their employees are not stuck all the time processing documentation during most of the day, they can actually spend more time with customers and sell them other services or even uncover new revenue streams. Another quantified benefit that we have seen particularly in highly regulated industries is the money that they can save avoiding penalties and fines once their automated processes run end-to-end in a matter of hours instead of several days or even weeks. And with very small error rates, essentially that’s allowing them to make most of their system landscape and work all their systems efficiently and avoid some of these penalties and fines that they get when they cannot meet their SLAs or cannot serve the customers in the appropriate way.

Lauren Hofmann: This next slide is focusing on the unquantifiable benefits, the qualitative benefits that came up in conversation. One of the most profound unquantifiable benefits that came up was the customer support. From my understanding from the customer, Hyperscience has a very proactive consultative approach to customer support and really does center the needs and feedback of the customers that are utilizing their solution. The second I’ll focus on for a qualitative benefit is this increase in the volume of data extracted. When I did speak with the customer, they had shared that they were on track to do 220 million fields, and that was before all of the volume was on the tool. So we understand that this is much more than what they were able to extract in a prior environment due to limitations around the prior system’s inability to really ingest and process a lot of the unstructured and semi-structured data.

Lauren Hofmann: Another benefit we discussed in the study was the reduction in the number of support tickets. This isn’t unusual to see when you’re decommissioning a legacy solution and introducing a new system that does have more up-to-date user capabilities, navigation, and improved UI. A lot of that inherently does reduce the number of support tickets. I do wanna emphasize a greater consistency in data accuracy. An independent group performed an audit on a sample of the customer’s documents and reported accuracy rates are consistently above 99.9%. And finally, an optimization of organizational resources. This ties into this hiring avoidance. Given the productivity that the customer had experienced, there isn’t necessarily a need to hire new people for the same work. In fact, many of these individuals are being repurposed towards higher value activities for the business.

Jorge Ruiz: I really want to emphasize the importance of data accuracy. We hear that from some customers over and over again. They really have this need to maintain or increase their data accuracy levels. That’s why we consider it paramount for us. That’s why we have this approach where customers actually set their desired target accuracy rates in the system. And then our proprietary machine learning models automate against that. That’s what is allowing this customer to reach that 99.9% accuracy in both dealing with handwriting documents and printed documents. That’s one of the key differentiators or reasons why customers choose Hyperscience.

Jorge Ruiz: In terms of some of these unquantifiable benefits, one that I find particularly interesting is around employee morale. Right now that everyone is talking about this great resignation, it’s very important to be able to retain talent in the organization. Obviously when employees have more fulfilling working days rather than spending hours and hours on tedious data entry, when we reduce their overall stress levels, they’re more likely to stay with us.

Lauren Hofmann: This next slide is a quote directly from the customer, just emphasizing again the importance of quality, especially when automating and executing. And I would add a further layer around the nature of their work in financial services in a heavily regulated industry where there is a lot more at stake from a compliance, legal, and governance standpoint. Lastly, we’ll look here again at the financial summary of the study: once again, 126% ROI and $2.9 million in benefits in net present value.

Jorge Ruiz: Thank you, Lauren. That was great. There’s a question asking about what leading providers were evaluated and compared, and what were the main requirements? So maybe you want to clarify a bit what you guys did.

Lauren Hofmann: I can certainly address the question around what were the criteria used for evaluation. The TEI studies are not intended to be a competitive analysis. So we don’t release the names of any competitors. But I will tell you that the solution requirements that they had expressed during this investigation phase were: one, being able to achieve higher automation rates related to image classification and data extraction. Second was around the ability to process semi-structured and unstructured data as that was not a capability that they had in their prior environment. A third was maintaining data accuracy rates throughout the automation process. And then fourth was the ability to scale these image classification and data extraction capabilities across the enterprise in a diverse array of contexts, different business units, and cross-functional areas.

Jorge Ruiz: There’s also a couple of questions around what we mean by accuracy and whether it was extraction accuracy or key value mapping inaccuracy. We are referring to extraction accuracy. We normally do it at a field level. So if there’s a hundred fields in a document that need to be extracted, if the machine does 99 correctly, that’s a 99% accuracy. And that other field will normally trigger a user interface very easy to use so that an employee can add that input. When our machine learning model doesn’t feel confident enough to meet that accuracy target, it just raises its hand and says, “Hey, human employee, can you please help me with this particular field?” We’ve got what we call a human-centered approach to automation where we help our organizations to work and collaborate very easily with machine learning so that they can obtain the highest levels of automation at the highest level of accuracy.

Jorge Ruiz: Mik is asking, do we need a separate workforce or an employee to work with Hyperscience? Normally when organizations already have a team of keyers, data keyers in the organization, they can perfectly do this same job, but in a much more efficient way. Normally the amount of training needed is insignificant, very quick. Business users can even create themselves layouts to work with some of those forms and documents or quickly train those models themselves with just a few hundred samples to be up to speed in just a few days.

Jorge Ruiz: There’s quite a few questions around handwriting. How was the accuracy percentage impacted by handwritten data elements within the documents? We provide the best machine learning performance, particularly for handwriting. That’s something that we hear from our customers over and over again. It is definitely one of our strengths. Our customers deal with a ton of documents that have printed data in them, but many of them have handwriting or even a mix of handwriting and printed data. Just think of a form that a bank or a supplier is sending you and you just fill it out and need to take a picture and submit by email. Those are very common use cases. We can perform really well and maintain all those accuracy levels that we are talking about with both handwriting and printed.

Jorge Ruiz: There’s a question here for you, Lauren. It says, what was the cost of capital used to calculate the PV?

Lauren Hofmann: Unfortunately, we’re not able to disclose. The customer did not disclose that. Even though these interviews that we conduct are confidential and anonymized, there are limitations. Furthermore, unfortunately, anything related to the underlying calculation that isn’t expressly written in the tables are things unfortunately we are not able to really provide detail around as it is kind of Forrester’s proprietary modeling there.

Jorge Ruiz: There is also a question around: Is this a study based on an on-prem deployment or SaaS? I can take this one. In this case, this was an on-prem deployment. This is a customer that has been with us for a number of years, and that’s initially how most of our customers deployed our platform. But Hyperscience can be installed on-prem or in a private cloud so that data can be processed within the customer’s private network. But more and more customers are now taking advantage of our full SaaS offering for even faster time to value and limitless scalability. Obviously they don’t have to wait for IT infrastructure in that case, and they don’t have to worry about configuration and upgrades. That’s definitely one of the most popular options these days.

Jorge Ruiz: We’ve got some questions here around languages. Does it support Asian and RTL languages? We natively support all main Latin languages, plus Arabic and Korean. But we are also in the process of adding other Asian languages, and we can support over 150 different languages with integrations with providers. If you have specific questions about a specific language, reach out to our team and they can guide you on whether we can support you or not.

Jorge Ruiz: This one here is quite direct. They’re asking if there are many vendors offering document processing automation, and what are the key differentiators from Hyperscience? I think I’ve already mentioned the way we treat or approach to prioritize data extraction accuracy as one of our key differences. I would also say the fact that we are built to ensure that humans and the machine work collaboratively through this very intuitive and very user-friendly interface, using then this human input to fine-tune the underlying machine learning models. This approach obviously makes deployments less risky, easier to execute and to scale than other legacy approaches to automation. I would also mention the high level of configurability. This intuitive user interface is built essentially for business users to easily set up layouts, models, and supervision tasks in under 30 minutes, without adding complexity or dedicated developer resources.

Jorge Ruiz: I think we have covered most of the questions here today. If I haven’t covered all your questions, please just reach out to us by email or through our website. There’s one question here that I’m not sure exactly what it means in terms of the study that you guys did, Lauren: what levels of outliers/bad decisions were acceptable? I don’t know if that refers more to the way we measure accuracy or to the study itself.

Lauren Hofmann: Unfortunately, I wasn’t sure if that’s related to data accuracy. If so, the customer didn’t disclose any further context around that.

Jorge Ruiz: We’d like to thank you, Lauren, for the work you guys did and for your presence in this webinar. Again, I would recommend everyone to download the report today and read through all that content and reach out to us if they want to continue the conversation. Thank you for assisting this webinar and goodbye.