Introducing Human Centered Automation

Rob Ak: Good morning everyone and thank you for joining us. This is Rob Ak speaking. I manage strategic partnerships within EMEA and it’s my pleasure to welcome you all to today’s webinar, which is all about Hyperscience’s evolution from a document and classification slash extraction tool to a platform. This is a first in a series and we use this as a way to introduce new features and capabilities to our partner community. With me I have our solutions engineer Nick. Nick is a certified wizard when it comes to all things Hyperscience and he’s very kindly offered his services to run a demo for us today.

Rob Ak: If you heard about Hyperscience in 2021 or earlier, you’ll be familiar with our classification and extraction capabilities. We ingest documents, we classify them and extract the output for downstream processing at a target accuracy defined by our customers. Typically, that’s 99% accurate or above. That’s great, but quite often our partners had to involve another tool to perform additional data manipulation or verification checks outside of Hyperscience. That increased the complexity of the task at hand and slowed down the speed of which tasks could be automated. We took this as a challenge and wanted to find a better way to allow humans and machines to collaborate together.

Rob Ak: Enter the Hyperscience platform. The Hyperscience platform enables a new era of human and machine collaboration that hyper-automates business processes to improve organizational agility without sacrificing on accuracy. By combining data, people, and processes into digital assembly lines, we turn complex business processes into simple configurable workflows. We build human in the loop into every automation. Our industry leading ML technology continuously learns and evolves to involve humans only when needed. With Hyperscience, businesses will be more agile and flexible, enabling our partners to deliver better outcomes for their customers faster than the competition.

Rob Ak: What is the Hyperscience platform? It’s basically a combination of what we call blocks, flows, and solutions. The blocks are the functions. The blocks are connected together through flows, and the flows have a specific use case which are the solutions, for example, healthcare claims processing. Let’s take a look at a healthcare claims process within the Hyperscience platform. This make-believe company wants to leverage Hyperscience to process their documents, but there are a handful of things they need to do to ensure documents are being processed correctly and that the data can be leveraged downstream.

Rob Ak: The consultants classify all their documents in a submission and ensure a claims form is present and that they’re being routed through to the correct keying teams. All of the UK customers are being processed by an onshore data keying team and any non-UK customers are being processed by the offshore data keying team. After all the information has been taken from the customer’s claim form and checked against the system of record, Hyperscience will ensure the name address matches so that it is in fact an existing customer of the healthcare company. The next step is to take additional information to ensure the content within the claim form is correct. In this example, making sure the line items add up to total.

Rob Ak: The last step is the decision of the outcome. That means being able to make a decision on this document based on the information that’s been provided. If a claim is below 250 pounds and all the information is correct, then we can pass that downstream for processing. If the information doesn’t check out or if the charge is above 250 pounds, it can be sent to another team for further review. Here’s another example where Hyperscience can run a check with business rules to decide if a document is in good order or not in good order and send it to a different location depending on its state. This can happen before a keying team spends time keying the document. For example, if you have three documents that should be in a submission and one is missing, we can save time by sending that to the customer and flagging it instead of performing supervision.

Rob Ak: Hopefully that highlights some of the changes within Hyperscience and particularly the decisioning around what we can do. I’m gonna hand over to Nick who’ll run through this in a live demonstration.

Nick: Thanks Rob. I’m gonna be giving you a demo of some of the new capabilities in our platform. We are really excited about these new capabilities as they really help transition Hyperscience from being an intelligent document processing platform to more of an automation solution, allowing you to do much more with the data that you are actually extracting, giving you the ability to work with it and helping you achieve your business outcomes all within one platform.

Nick: As we’ve already mentioned, the core of the Hyperscience platform going forward is made up of what we call flows and code blocks. A flow is a workflow that’s made up of a number of activities. These activities we call code blocks. These code blocks are these rectangular or square objects you see on the screen which represent activities that you are able to configure in a way that makes sense for your particular business need. We’ve got all sorts of different code blocks that allow you to do what you need to do, all the way from input blocks that allow you to ingest documents into the system automatically, all the way through to classification blocks and extraction blocks.

Nick: However, what we now have are additional blocks that allow you to do things like checking that you’ve got all of the required documents. If you don’t have all the required documents, then fire off some type of action like sending an email to somebody or potentially an SMS notifying them that you are missing a particular document that’s required as part of your document pack. We also have other blocks giving you the ability to break up the routing based on a particular layout or document type. You might want to route a particular document for manual supervision to one team and then have another document sent for manual supervision to another team. And then we’ve got other kinds of blocks allowing you to do database lookups or API lookups for things like data enrichment, potentially doing things like validations looking up into either internal systems or external systems, allowing you to do validations on the documents, even comparisons between documents.

Nick: In this example, we’ve got a policy number that’s being validated against an external source and a name and an address validation being done against two types of documents, even the ability to do calculations where we are summing up total line items and checking that they correlate to the final total in the document. For our demo today, I’ve got a slightly simpler one where we are using an email input block and we are gonna send it through for processing where it’s gonna go through standard classification and extraction. But what we are also gonna be doing in the background are some document checks where we are checking that the right check boxes have been selected. If the right check boxes have not been selected, we’re gonna flag that in the user interface, send it through for further processing again, and then we’re gonna be doing some further checks in the background where we are gonna be checking that documents that we expect are present. If not, we’re gonna flag that in the output JSON. We are gonna check that the ID numbers match between the application documents as well as the proof of ID. We are gonna check that a signature is present and we are gonna be checking that the training costs are not breaching the thresholds that we’ve put in place in our organization.

Nick: At the moment I’ve only got two submissions in the platform. I am gonna send an email which is then gonna be picked up by our email listener and then sent through into the Hyperscience platform. I’ve got two documents: a training application form and a proof of ID. Let’s send that into the right email box and then take a look at that email coming into this mailbox where I’ve got an email listener that’s listening in. This email should arrive in the next couple of seconds. Here it is and the email listener should pick this up very shortly. Very shortly we should see this transitioning to a read email, which you can see. That means that going into Hyperscience now if we refresh the UI, we should move from two to three submissions, which is the one that you see here with the status of processing. That’s just one of the examples of the mechanisms that we have of getting documents into the system in an automated fashion.

Nick: This is gonna take a minute or two to actually run where the machine is going to automatically classify the documents based on what it needs to look for and then begin the extraction process. Let’s go and take a look at the documents and the particular use case that those documents support. In this example, we’ve got this first document, which is a training document. Somebody is requesting training, they’ve gone and filled out this document with their general particulars and then they are filling out the rest as well: training center details, their manager signing, the cost of the relevant line items for the course. What we are doing is then taking some of this data and revalidating it. The first validation that we’re doing is the ID number against the proof of ID, checking that that corresponds. In this example in the top right, we’ve asked the employee to only select one option, but they’ve mistakenly selected two, to be doing some checks against that. And then what we are also gonna be doing is checking that we’ve got the signature present because an application form more often than not is not valid until there’s some type of proof of signature.

Nick: Let’s go and see what the Hyperscience platform has done. We can see that it’s already classified those documents and it’s done so correctly. We can see we’ve got a training application form and we’ve got an ID card. The status has changed to one where there is a supervision task that is available to us. A supervision task becomes available when the machine is not confident enough about something. When you look at the Hyperscience platform, we lead with accuracy and we let automation be a byproduct of that. You’re able to configure the accuracy that you’re looking for when it comes to classification and extraction and the machine is always gonna be working against that accuracy level. Where the machine is not confident enough, it’s gonna raise its hand and ask for assistance rather than be wrong and extract data incorrectly.

Nick: Here we see the machine requesting some assistance. All you would need to do is go and perform the relevant task and the machine is gonna take you directly to the field that it needs assistance with. In this example, the machine is looking for a course code number and you can see the machine is not really sure because of this arrow over here, we’ve got some squiggles in here, some mistakes. And all we need to do as a data keyer is just provide the right data out of that entire document. It’s the only field the machine needs assistance with. So I’m gonna send that through for processing.

Nick: As I mentioned in this particular workflow, there’s gonna be some work going on in the background where the machine is gonna do some checks against that document and if the document fails a particular check, it’s gonna raise the verification failure to our attention in the user interface and we can see that another supervision task has become available here. What we can see over here on the right hand side is that it’s warning us that multiple check boxes have been selected. For this particular training application document, we were only supposed to select one of these options: either initial training application, resubmission, correction or cancellation. It doesn’t make sense having both checked. So the system is flagged as of multiple checks.

Nick: Let’s assume that I am a knowledge worker, I’ve got the relevant rights to reach out to the user and make a correction. After speaking to the employee, I’ve deemed that this was supposed to have been an initial application. So I’m just making the change and then submitting this down for further processing. This is now gonna go through for further processing. Now as part of the custom code blocks as part of the workflow, we’re gonna do some additional checks on this document and then those will be flagged accordingly in the JSON output, which we will then send to our downstream systems.

Nick: It’s gonna be checking and doing validations against the ID, it’s gonna check that the signature is present, it’s gonna check that the training cost isn’t gonna be breaching our threshold. Once it’s done that it’s going to return a status of complete. Once all of that data has been verified and checked and all of the data has been extracted, we are getting some automation statistics returned back to us. We can see that the machine by itself was able to identify a hundred percent of all of the fields and it achieved 97% automation in terms of data extraction and transcription.

Nick: Here is the application form and you’d be able to click on all of the various fields and see the data that gets extracted and you’d be able to see that it’s all extracted very accurately. Here’s a good example where machine learning, which is what Hyperscience is built on, trumps OCR anytime. If you take a look at the ones, the I’s and the L’s that look very similar, this is where OCR would fail. Hyperscience is able to extract this accurately because we associate every field with a data type. In this example, it’s an address. This is what gives the machine context, especially when it comes to handwriting. The machine then knows how to accurately extract this information.

Nick: Here’s one where we made a mistake and the Hyperscience platform is trained to ignore mistakes. It’s trained to understand human intent and it ignores the mistake here and goes and transcribes the rest of the data accurately. Here’s a good example where the employee wrote outside of the box slightly, the machine expanding its vision allowing it to go and get the rest of the data and then transcribing that very accurately here as well. The second document is just a proof of ID where the machine was able to automatically identify all of the fields and then extract the data accurately there as well.

Nick: Just to show you what’s happening in the background, all of that information gets extracted with a JSON format and would get passed down into your downstream systems. And here is proof of some of these validations that actually happened as part of those custom code blocks that I mentioned that you can build into a workflow now in the new Hyperscience platform. We can see that we’ve got a KYC pass over here with the ID number matching between the application form and proof of ID. The application form has been signed. However, we’ve got a problem here with the threshold for the training has been exceeded. The total of the training is 4,100, the total budget allocated is 4,000. Now based off this threshold being exceeded, you would be able to automate some type of notification back to the end user if required, like sending them an email or potentially sending an email to your call center asking them to reach out to the employee. Potentially this could even be formally tracked in a ticket management system as well.

Nick: Let’s take a look at one of the other submissions that I had loaded previously. In this example, we only have one document, which is the training application form. We are missing the ID that gets flagged in the output here as well. You can see that we have a warning proof of ID is not present. Once again, just automating that piece of checking whether you’ve got all of the documents present or not. And once again, if you don’t, being able to then automate some type of action in terms of a notification. Lastly, just for today’s demo, we’ve got this other submission that I’d loaded previously. In this instance we’ve got both of the documents, but if I go into the training application form, we can see that this document has not been signed and that gets picked up by the system as part of the validation. And you can see a warning over here, the application form has not been signed, and once again you can then automate some type of action to be taken.

Rob Ak: Thank you very much, Nick. Let’s do a quick recap of everything that we’ve discussed. It’s kind of four key areas that we’ve touched upon here. So number one, it’s flows. Now these represent a business process effectively assembled to create end-to-end solutions for specific use cases or industries such as health insurance, claims processing, mortgage origination, or account opening processes. The next keyword is around code blocks. So code blocks execute as part of a flow and enable you to assemble the processes that best fit for your document processing needs with ease and speed. Next one is connectivity. Whether you need a flexible method to ingest documents, validate the information presented on a document or enrich the output of your documents by automating the addition of valuable data from external systems so that humans don’t have to, Hyperscience provides you with a powerful platform that fits right into your complex ecosystem. And then the last one is supervision. Hyperscience builds human in the loop into every automation. What sets us apart is our ability to know when we’re likely to be right as well as when we’re likely to be wrong, bringing in data entry teams to review and resolve tasks only when required.