Equidox AI Demonstration

[Pat Needles] Hi everybody my name is Pat Needles.  I am a Vice President here at the Equidox Software   Company. I first wanted to thank each and every  one of you for making yourself available today to   join us for what proves to be a very exciting  and new technology that Equidox has developed   and we're super excited to show you how that  works and the value that is associated with   that. And you can see on this first slide up  top our mission statement, which is “enabling   PDF accessibility through intelligent automated  solutions.” and that's kind of a common theme here   you'll see that throughout today's presentation  and discussion about the genesis of Equidox.   Where we started where we've been and where we're  at today. So we're excited to talk to the folks   especially we've never ever heard of Equidox.  This might be their first time at one of our   webinars – but welcome to you and we're looking  forward to showing you what we have today. (you   can move forward Dan) So what we're going to talk  about today as far as the agenda is concerned is,   who is Equidox where did we come from, where are  we going. As I had mentioned before some super   exciting things that we're going to talk about.  Why make PDFs accessible. I think most people   on this call understand a lot of the reasons why  you might want to do that and that sort of thing   and so we're also going to talk today about our  AI-powered PDF accessibility tool and how that   works and the value of that. And of course we'll  get into the most important thing here today which   is the actual demo of the AI solution. So who is  Equidox and where did we come from? Equidox the   genesis behind it starts in about 2011 when we  were asked by a Canadian federal government to   build a software tool that would make PDFs  accessible for people with disabilities.   And when we started this process and building the  tool our first output was PDF to HTML. Soon after   that we grew to a PDF-accessible output and from  there we have built in a lot of different types of   automation into the tool that we will definitely  be talking about. We have built over say,   over the last three years, some detection tools  into our software application that has made PDF   remediation much simpler and more efficient.  And any of those… any of you folks that have   been that have done remediation that is on  this call today understands that this is a   very painful process. It can be time-consuming  and we've taken a lot of that heavy lifting   off of the table by building in tools into the  software application that allows for fast and   easy remediation. And then secondly what we're  going to be showing you today is our AI-powered   software solution for high-volume templated  at scale recurring documents. (okay Dan) so   you know we talk about the challenges of PDF  remediation and how difficult that can be you   know it's very labor intensive. Folks need  accessibility skills in order to remediate   documents properly. It can be very expensive if  you are outsourcing these types of activities to   third parties. There is the possibility of human  error while remediating documents on challenging  Timelines. (Go ahead, Dan) So the Equidox AI  solution as I had mentioned earlier is a fully   automated PDF solution that really takes the  human element out of actual remediation. And   so what we're going to show you today is how  we import documents into our system, how they   run through our AI engine, and how they come out  completely tagged and completely accessible. There   is no need as I had mentioned for any human  intervention for this too. And we are really   excited to show you how this works here today. So  with that, I'll go ahead and I will turn it over   to you Dan, to do the demonstration of the tool. [Dan Tuleta] Okay great, thanks, Pat. So before   we get into the actual demonstration of Equidox AI  I’m just going to set the table for a little bit   more just in case anyone is new to accessibility.  I'm just going to talk through a little bit of   the process and how it works and also why it's  important to do this so just so everyone's aware.   Equidox in in cooperation with the NFB conducted a  survey of over 250 blind and low vision assistive   technology users. So that would primarily be  people who are using screen reading technology   in order to read digital content on their computer  screen. And we have found, after surveying these   250 blind and low-vision users that at least  two-thirds of PDFs that they interact with on   a regular basis are not accessible. So they  cannot interact with those documents if they   were to use their assistive technology. Adobe,  who owns the PDF format estimates that there are   over three trillion trillion PDFs in circulation  today. So that means there are about two trillion   completely inaccessible documents out there that  are floating around in circulation. So that is   a staggering amount of documents and many people  who aren't blind or have low vision and don't use   assistive technology, are probably interacting  with PDFs on a regular basis and you don't even   really think about it. But those PDFs present  a huge barrier to understanding information for   people who who rely on their assistive technology  to interact with the content. So the main takeaway   from from this slide here many people on this  call have probably heard the the acronym WCAG,   which stands for the Web Content Accessibility  Guidelines maybe you've heard of the section 508   standards. We're not going to get into the  alphabet soup and start breaking down the   differences and the similarities between these  different standards and guidelines but the main   takeaway from this is that these are the standards  that are being used to measure the accessibility   of websites and documents and digital content  and the number of lawsuits that we're seeing   every single year is on a steady rise so the  lawsuits are coming in fast and furious around   websites and documents and digital accessibility  as we move to a further and further digital world   world. So organizations have to be aware that they  are not exempt from making sure that their content   is accessible and you have a responsibility to  make sure that you're providing digital content in   a way that can be used by everyone. So with that  said, as Pat kind of alluded to a few slides back,   the the challenge with PDF accessibility for  many organizations is just the sheer volume of   documents that they produce and distribute. So  the the number of pages can make it extremely   difficult to manage on a consistent basis.  Outsourcing this work is extremely expensive.   You also lose control of the deadlines making  sure that that outsourced company is getting   those documents back to you in an efficient  manner and meeting your deadlines. You can lose…   you can have concerns over the security and the  privacy of these documents if you have anything   that needs to remain confidential and also you're  putting the faith in these outsource firms that   they are going to be delivering high-quality work  which they may not be. And also remediating these   documents internally for these organizations has  a lot of the same challenges but mainly you have   to dedicate very large numbers of staff to be able  to manage the volume of documents and it becomes a   big killer of time for these organizations and the  employees that are the ones that are having to do   the actual remediation work. So what we have done  is we have introduced Equidox AI which can fully   automate the accessibility portion of making  these documents accessible using artificial   intelligence, a combination of machine learning  and computer vision. So we are essentially   applying training to these artificially  intelligent models to teach it what these   documents look like, the different structures  that are making them up and then it can fully   automate the process to ensure that the documents  are accessible and being delivered to the customer   without any issues of using it with their screen  reading technology. So Equidox AI, it all starts   with human programming. So I just want to make  it clear that Equidox AI is not “auto-tagging.”   There are many people who will throw around the  term “auto-tag,” meaning that it's automatically   creating tags for the PDF and the tags are what  are being used by screen readers and assistive   technology to interact with the content. Equidox,  although it is applying tags in an automated way,   it is not auto-tagging. Auto-tagging requires  basically hard coding to teach an auto-tagger that   this type of font should be a certain type of tag.  It is very inaccurate it's very unreliable it will   not produce compliant or usable documents. So what  we do with Equidox is we start with human work   being done up front. We train the AI models based  on the structures of that specific PDF template   that whatever that template may look like, it  could be a bank statement, it could be a utility   bill, it could be an explanation of benefits on  an insurance form… So we have a combination of   our data scientists who are doing the actual  AI training, we have our software developers   who are implementing this technology, and we of  course have accessibility experts on staff who can   assist in that training of the AI models because  they know exactly how to tag a PDF document. So just to go through a little bit of the process…  So Equidox AI is implemented postprocessing.   And what I mean by that is you do not have to  completely uninstall and recreate a process for   building your PDF documents. Many organizations  already have a very well-established process in   place to create and deliver these documents to  their customers. We are are placing ourselves   into that workflow seamlessly without having to  completely rebuild that process from the ground   up which could be a very time-consuming and costly  endeavor to do so. The first step with Equidox AI   is, you would provide sample documents to us  in order for us to analyze the documents. And   then we would train the AI models based on how  we analyze those documents. So we teach the AI   models all of the different elements that are that  are accounted for in your template. We apply that   training and then we would integrate the workflow  so we would figure out how do we want to deliver   these documents to the customer. Is this something  where they are downloading a PDF from one of your   portals? Maybe it's their bank statement, for  example, or is this something where you would   like to manage it internally, where perhaps you  have a bulk import of documents every single month   that need to be made accessible before they go  and get posted online? Once that is that decision   is made as to how we're going to integrate  the workflow, the process would begin where   Equidox would apply the tag structure to these  documents. The documents are then exported from   the system and then they are delivered through  whatever mechanism we've determined. Whether   it's directly to a specific customer of yours, or  if it's to some sort of data repository so that   these documents can be uploaded to your internet  or website or wherever it needs to go. So just a   a quick rundown of how we analyze PDF documents  we don't need to get too technical with this but   basically these color dots that we're seeing on  the screen here these color dots just represent   the AI's analysis of different variations  of documents. So they are grouped together   in different areas by the AI based on different  anchors that it's seeing. So for example these   documents all contain pie charts so that is what  would group them together in a specific cluster,   and it would trigger the AI to look for certain  variables and certain elements on this PDF that   are consistent with the training that we have  provided. Here's another example where you can   see that a number of documents, in this case,  are sorted into different segments based on the   number of columns. So you have certain pages that  might have two-column layouts. Other pages might   have three or four or a single layer or single  column layout per page. So it's always based   on the template and the training that we provide  from that initial analysis of the documents that   you've given to us so that we can teach the model  exactly how to tag the various elements. One final   thing to call out is just the ability to even tag  tables. So if you think of like an easy example to   think of would be like your credit card bill it's  a very static document could be the same document   template can be provided to hundreds of thousands  maybe even millions of customers for a credit card   company. And that document is being produced on  a monthly basis. Certain customers of that credit   card company might have a single charge for the  month on that credit card and their statement   might be a single page with a single line item for  a single charge in that table. Another customer   might have 500 charges on that credit card for the  month and that statement, although it comes from   the same template, will be very different. It  might be 10 pages long and have page after page   of tables. So we are able to actually find these  variances and we can account for them to make sure   that everything is being tagged accurately. So  it's not just a static layout on every page there   can be variables across these templates and we're  able to train the AI models to identify these   variables and account for them accordingly. So  the the training of the AI model, just so everyone   is aware, we do define the different elements. So  if you've never tagged a PDF document before, PDF   documents will contain primarily text but there  has to be a structure applied to the the text on   the page. For example, there are headings that are  used for navigation purposes. Tables have specific   layouts that need to be tagged in a certain way  so that assistive technology can interact with   them. There could be lists. The reading order  is a very important aspect as well. If a page   has three columns a screen reader needs to be  told to read down the left-hand column first,   then the middle column, then the right-hand  column. If it's not trained on how to do that,   the screen reader could easily just read clear  across the top line of all three columns rendering   the entire page as completely useless. So these  are the types of training that we that we put   into the the models to teach it for all of these  specific variables for the specific template. And   then of course we have to test the model. So  this is not just a one-time we teach it what   this document looks like and we just hope for the  best. We of course test these models and continue   to train them throughout. And so our goal is  always to both pass the automated checkers as   well as making sure that these documents are  fully usable for the end user who's going to   be interacting with them. So we do apply both  manual checks as well using screen readers to   sort of replicate how a blind user might be  interacting with that document. So these are   very very important aspects of PDF accessibility  not just auto-tagging and hoping that it passes an   automated checker. It really only matters if the  end user that's receiving the document is able to   interact with it and that's what really matters  most to us. And also, I alluded to this earlier,   but there are different methods of deploying  Equidox AI technology. So we do have we do   have an interface that we have built which I will  be demonstrating here in just a moment. So we've   built an interface that a customer would be able  to run this technology themselves if they prefer.   We can also embed this technology through through  APIs. So like I said before, we can make sure   that this is fitting into your existing workflow  and you're able to apply the tags and deliver an   accessible document in a fully seamless transition  where no human is required at all. And then we   can also operate this as a managed service. So we  can run the entire process and technology for you   and deliver those documents to you, depending on  the use case, depending on security requirements,   anything. Every customer may have some different  requirements and we're here to work with you and   figure out the best method of delivering these  documents in an accessible way. Okay so what I'm   going to do now is actually transition out of  this slide deck and we will get into the actual   demonstration of Equidox AI. So if I just go  to this other tab here I have in my browser,   this is the interface that we have built just  to kind of demonstrate the technology to make it   easier to understand what's actually happening.  So right now I have a completely blank slate I   have nothing really to work with as we can see so  what I'm going to do first is I'm going to go to   the Upload Documents tab. When I arrive at this  tab here I can then just open up the folders on   my hard drive and in this case I'll just drag a  .zip file that contains some bank statements that   we have scrubbed. See these are just like sample  bank statements and we will upload that .zip file   containing the financial statements. So once these  documents upload, I can then go to the Run Batch   tab here on the left-hand side. I will then choose  the model that I want to apply. In this case, here   we've only got three models currently uploaded  to my little interface here for demo purposes,   so what I'm going to do is I'm going to select the  statement model. I will then choose the .zip radio   button here and I will select that .zip folder  that I have just uploaded just a second ago. Once   I've selected the .zip file and I've selected  my appropriate model, I just hit Run Batch and   this will kick off a process where Equidox is  currently ingesting the documents. It is looking   at them kind of page by page, one by one and it  is taking all of that model training that we have   done ahead of time and it is automatically tagging  these PDFs. So we'll see these green lights start   to light up here as we work through the process.  So you can see one of them is already complete,   and we will see more of these green lights kind  of lighting up in just a few seconds. Now keep in   mind that this is just a demo environment of this  technology. Equidox AI can be scaled to meet much   higher demands and throughput. Of course, certain  organizations might require literally hundreds of   thousands of documents every single day to be  remediated. We can just dedicate additional   resources to it in order to power it to meet those  types of demands for speed. While these documents   are going through the process, I'm just going to  open up one of them here just so that you can see   what these documents look like. So if I open  up like randomly statement number four here,   what I can see is that this is a sort of  generic-looking bank statement. We have   images up here, we have text, we have headings,  we have tables. But this document is completely   untagged in its current form. If a screen reader  was to interact with this document it would tell   them that this is a blank PDF. So they would not  be able to interact with this. They would have   no idea what their monthly transactions were.  They wouldn't know what their balance due was.   Obviously, that's a huge problem because you want  to… some people really need to make sure that they   are being charged appropriately for all of their  transactions for the month and they, of course,   need to know how much they need to pay for their  credit card bill. So we need to make sure that   these documents are being tagged and delivered  in an accessible way. So with that said while I   was talking there the whole process finished. I  can see the status is now done. So all of those   documents finished processing. I can look at the  individual PDF files. So if I were to let's say   just download one of them I will download this  and save it to my desktop. And if I were to open   this up in Adobe Acrobat, I will see that this  version of the document is now fully tagged. So   if you're familiar with tag structures, you know  how complicated it is to manually set up these   types of tag trees in Adobe Acrobat. If you're  doing it manually it's a very time-consuming   process. It requires a lot of technical expertise  and even if you are the best Adobe Acrobat user in   the world it is still very very time-consuming.  So at the scale of things like statements, where   there could be literally hundreds of thousands  of them produced every single day by a single   organization, it is completely unsustainable to  expect human remediators to go through the process   of tagging these elements manually. But now if I  look at this version of the PDF, I have the logo   has been described. We've trained it to apply  this company logo alt description to the logo   whenever it sees this particular image. We have  text identified so the entire document is tagged.   As we can see here, we have text identified, so  the bank name and the customer name. We have our   figure tag which is just the image, we have our  heading level one. As we tab through we can see   we have an H2 for the account summary, the table  which is nested below it has been properly tagged.   So you can see all of the rows of the table are  properly tagged. We jump down to another heading   level two for the deposits and credits, another  table that is properly tagged, another heading,   another table… All of these different elements  have been accurately accounted for because of   the initial training that we have put into this  document template. So at the scale of a company   that is mass producing these types of documents,  we can now fully automate that process so none of   your staff has to spend countless hours of their  week remediating PDFs. This type of process can   be applied to other templates. So it's not just  for bank statements. It could be explanations of   benefits, it could be bills, like utility bills or  medical bills, or medical statements and reports   and test results. There's an endless amount of  of use cases for Equidox AI. It really just comes   down to the customer and the types of documents  that they're producing. But we would certainly   love to chat with you based on your specific use  case and your specific template to see if Equidox   AI would be a good fit for your organization.  So closing out of this document now I can also   download all of these documents in a batch as  well. So if I go to my Batches and Jobs Tab here,   this is the batch that I just ran. So I would  be able to extract all of these documents in a   .zip file as well. So I'm able to download  them as a .zip just like I delivered it to   Equidox. But keep in mind you don't always have  to run this process through this through this   interface. We can run this process for you as a  managed service, or the technology that's kind   of going on behind the scenes can be seamlessly  placed within your existing workflow so that we   are not disrupting a longstanding process that  you have for generating and delivering these   PDF documents. Okay so just as a quick summary.  Equidox is a leader in PDF remediation. This is   all that Equidox does. We are entirely focused  on PDF remediation. So we can solve your PDF   remediation challenges whether it's a large or a  small project. Many of you are here and this may   be your first introduction to Equidox. Keep in  mind that that is only part of what we offer. So   the Equidox AI solution is our newest technology.  But we also have Equidox Software that is meant   for more unique one-of-a-kind ad hoc documents.  So there are many organizations that are producing   you know marketing flyers or just documents  that are meant for kind of one-off use. They   might not be producing them en mass. So Equidox  software is designed to dramatically speed up and   make more efficient the process of going through  and tagging those documents that have completely   unique elements where we can't necessarily train  artificial intelligence to fully automate that   process. We have tools built into that Equidox  technology that are powered by artificial   intelligence to speed things up, but there are  certain examples of documents where you do need   some sort of human intervention to analyze the PDF  for context and be able to properly remediate it.   So if you're interested in those types of use  cases you can also feel free to reach out to us   after the fact. We would love to talk to you about  both or either or use case that you have in mind   for your organization. And of course, we did just  see the Equidox AI demonstration there. So you   have a general idea of how that technology works.  We will be sharing this slide deck out to everyone   on the call after after the fact. So there are  some relevant articles here if you are new to   PDF accessibility or digital accessibility. If  anything of anything presented today was confusing   we will certainly give you some resources of some  relevant articles and we do have a really large   library of information on our website as well.  So feel free to interact with these articles and   check out some of our past webinars if you're  interested in seeing our other technology. But   also please feel free to reach out to us for a  more personalized demonstration and discussion   around your use cases. So with that said I'm  going to pass it back over to Pat. We just have   a minute or so left to to kind of wrap things up  but thank you everyone for for attending today.  [Pat Needles] Thank you, Dan. Great job great  presentation. I hope that all of the folks here on   the call today saw the great value in the Equidox  AI tool. And we'd love to hear from anybody who   feels as though there might be a proper use case  not only for our AI solution but as Dan said our   PDF ad hoc remediation software tool. And you  could reach us at EquidoxSales@equidox.co,   of course, you can always reach out to  us by phone you see the number there,   or if you would just like general information  feel free just to visit us at www.Equidox.co.   And that concludes today's presentation. For  more information about how Equidox Software   Company can help you with PDF accessibility,  email us at EquidoxSales@Equidox.co or give us   a call at 216-529-3030, or visit  our website at www.equidox.co.