#93 TAG talk about AI and Document Extraction

Hallo and welcome back to another episode of the Process and Automation Podcast with the Automation Guys. Yeah, another week, another podcast to record. So today we have decided to jump on an AI topic. No surprise, AI is still very hot. OpenAI came up with a big keynote earlier this week with lots of new features. So AI will be huge for the rest of the year, next year, and then going forward. So yeah, so today we thought let's look at a very specific topic of AI, which is the document extraction, which is very, very important for companies, especially companies who are dealing with documents, invoices, contracts, purchase orders, all these documents a lot. So we thought we will cover 4 very important areas of those AI document extraction tools. And yeah, so we would just jump straight in. So it was our first very important feature. And yeah, one of the important features is the ability to read digital documents, printed documents, and sort of the very special discipline is the handwritten documents. So, you know, when you're in business, there are lots of documents flying around, especially when you're like a bank or an insurance, and these documents reach you maybe via email, or maybe they reach you into your mailroom where they have to be scanned. So, but before artificial Intelligence, extracting information, reading that information from those documents, and even handling those documents was, was a pain. But now, now, now with AI, we can process these documents and it needs the ability to recognize these documents, what it is, is it an invoice, and, and for that optical character recognition is used. So Yeah, we will convert, or we will convert these documents usually, and businesses have to convert these documents from maybe from letters into digital documents. And then that OCR will work on these documents and it can find out images and readable text in these documents. Yeah, and as well with this OCR technology, it can even identify handwritten text. So that's very important and very difficult to handle because not everyone out there is using digital channels and creating PDF documents in sort of in a structured format. There's still lots of people out there, they're filling in information on a sheet, maybe within a bank branch. So all these kind of things have to be recognized. So with that ability of processing digital documents, print documents, or handwritten documents, yeah, a business can really achieve really, really seamless processing, maybe a good end-to-end processing in some workflows. Yeah, so, so that's, that's a key feature every AI document extraction tool needs to have these days.

Yeah, and I agree with all of that, Sascha. I think strong AI that can look at OCR, not just for digital but also printed handwriting, handwritten documents, you know, very key. There's still a lot of forms out there that we see that's filled out by hand. They are scanned. So, so, so that's definitely a sort of a feature, strong feature to look out for. And no surprise at the top of the list. The next one, when you look for a tool that specifically, you know, for document extraction and sort of the AI-driven document extraction, is look for a tool that allows you to have custom document classification models. And traditionally, if you look at intelligent document processing solutions, there was a flaw in the way they work because the models they used to actually extract the data from the document were prebuilt. So if you classify incoming documents, you know, it was prone to a lot of errors because there was a lack of precision in the and these moments. And again, you know, this, this actually led to quite a lot of frustration where human workers had to really manually review the results after the extraction happened. And, you know, there's, there's lots of reclassification that, that constantly were happening. If you look at sort of the modern tools, sort of OpenAI tools, what you what it allows you to do is, you know, you could leverage machine learning to effectively build your own AI models for these tasks like document classification. So what you could do is you could just go and sort of upload a batch of business documents, for instance, contracts, and then the platform itself can train a custom AI model to to recognize these documents and also to recognize them in the future. So if it's more of the same, so it actually reduces the sort of level of interaction you have to have with the actual document itself, you know, when classification happens. And, you know, this custom model approach, it really ensures that, your document processor is able to pinpoint the exact document types and, you know, and, you know, and then it allows you to sort of extract the right type of information from the document. So if it's a contract, you know, it's quite intelligent in recognizing that it will know sort of certain areas of the contract. So if it's a, you know, non-disclosure agreement, the model will know sort of what areas of that is important. And, you know, and it sort of funnels that information to the right places. And, you know, even goes a step further where it could sort of do an analysis for you if something in a document perhaps doesn't seem right. You know, instead of saying it's a nondisclosure for 2 years and it says 25 years, you know, these are the sort of things that the intelligent part of intelligent document processing comes into play. So it's not simply lifting and shifting data from unstructured source to structured source, displaying that to people. You know, it can sort of learn about what is, you know, with insight, the structure of the document. And also the, you know, in terms of the content, you know, you know, what's accepted content. And I think this is very relevant in the legal space, for example. And again, you know, if you choose the right AI-driven document extraction tool, you know, what it should allow you is to sort of version your various models and sort of create new models. If your documents change, it changes maybe in layout. So again, this ability to create a custom machine learning model is really, really important. So it's not pre-canned, you know, models where you can't really change the thing. And that in itself is a major sort of capability that you look in for or looking for when you look for a tool to do document extraction.

Yeah, so that classification is very important. And having the AI automatically classifying it will be huge because AI can I can remember that hardwired pre-built configuration in the OCR tools, obviously very powerful back then, and but still you have to, you basically have to go through lots of variations of invoices and pre-build it there and search for certain fragments. Yeah, definitely, definitely. Good that we have AI these days to support us was that one. Because without it, it was always very clunky and error-prone, as you mentioned. And that still was then the reason why we couldn't properly end-to-end orchestrate the processes. So there was always something to be reviewed or too much to be reviewed. So yeah, it's very important that actually that capability of AI document extraction as part of the end-to-end process. It's great to have that tool, I guess, sitting on one side, but it is even better if it's part of an AI process automation platform, intelligent process automation platform, where it can actually enable that end-to-end processing. So once statements are maybe classified and extracted so that that end-to-end processing maybe into a finance tool, that would be really cool instead of having one tool here and then sure you can speak to another tool over there. But if it's sort of seamless in one platform, then you get really the big, big benefits. And yeah, and there's so many processes which are document heavy and the more you can extract The more intelligence you have in that extraction, the more processes you can enable for better processing. And yeah, it could be hiring processes, background checks, all these processes where documents are flying around in digital form. It could be just an email as well where you extract information, or it could be just a PDF where you extract information. But yeah, having that as part of the wider orchestration, workflow orchestration, very, very important. Yeah. And when you currently decide for platform which should do workflow orchestration, AI document extraction, process mining, etc. So yeah, look out for platform rather than individual tools you need to plug together later on.

Yeah, yeah. And I guess the next thing is you need to look for a tool that's got very, very strong data foundations. And this is where you can sort of take document extraction to the next level. If you have a platform that includes what is called a data fabric architecture, and that offers very, very easy data connections. So just briefly, you know, the data fabric is really a sort of architectural layer. And I guess you could view it as a toolset that connects data across disparate systems in your organization and it creates a unified view of it. So effectively, it's sort of a virtualized data layer, you know, so it means that, you know, you could have data that lives in the ERP system, combine that with data in your CRM system, could be on-premise data, cloud service or multi-cloud environments. And when you have a sort of a data fabric in place, it really allows you to combine all of your business data in sort of new ways and allow you to surface that data, like I said, in a unified view. So again, data fabric. So if you look at, say, document extraction, once you extract the data and it goes into your sort of data fabric layer, it really then ensures there's a sort of seamless flow of that data across your various environments, the type of software that you plug in. Because it all kind of points to a sort of a virtual layer. You don't have to extract the document data into one database and then move that data to another database because there's an approval process. Everything kind of connects to sort of unified source. So again, if you're looking for a tool that helps you with your document extraction, it is really, really important to ensure that sort of the data fabric architecture is supported and is built on top of that. Because again, you know, it just means that any information that you pull from documents can then easily be passed into other systems. Other systems can reference it if you have workflow that perhaps got orchestration for moving the document around, especially in sort of approval scenarios. It just helps the fact that you don't have to duplicate the data in sort of various silos in order to achieve that, you know, and then again, you know, Data Fabric gives you a very kind of complete view across the whole data ecosystem. And if you're gonna add loads of documents into this mix, you know, you want to make it easy for employees to sort of gain access to sort of the datasets that these, the underlying datasets of these, you know, documents. And perhaps you want to enrich it with customer information, right? So if a document is for a customer, you could pull that information into a dataset, goes into your data fabric, you might want to actually enrich that by actually looking up the customer reference or the customer record. So it might be from a sort of a processing perspective, and especially in a sort of document approval scenario, you want to hang some additional information off that particular document, for example. And this is where bringing all of that information from the various systems and representing that in one data set actually makes a lot of sense. And again, if you think about API connectors. So you might want to have API connectors between the different systems and you want to move data around. If you have sort of one source of the truth, then that's very, very powerful. Especially when perhaps you have third-party connectors to your documents where third parties are allowed to sort of integrate via APIs. Again, it's just sort of one source of the truth. It just keeps things sort of very, very tidy. So definitely want to look out for, so it's not just that individual silo, the document data gets extracted, it gets dumped in the database, and then there's all sorts of ETL scripts that then need to push that into various places, mirror that everywhere. You want to have sort of one platform where it natively supports this. Because otherwise, I think you're going to run into trouble very quickly with data duplication and, you know, sort of old ETL jobs that runs and sometimes it doesn't run and then, you know, it just creates a very, very unscalable solution.

Yeah, very, very important to get that data foundation right, 100%. Because here we were talking about digital transformation, getting lots of applications built, every department goes digital. And if that data foundation is not right, I think it is easier said than done really in larger organizations. But when that is right, so you can immediately start building apps and you tap into the data which is driving the value for your customers or was basically along the value chain of your business. So yeah. It's really, really cool to have that achieved. I know for lots of big companies it is such a big job, but at some point I think everyone has to get started. Yeah, you have these data warehouses. I think that was sort of an early version of that kind of thing, but still we are so far away from achieving that for most companies. It's very important to have that on the agenda. And yeah, and then as we say always, like, do it bit by bit, isn't it? You don't have to achieve it straight away, everything. But if you have a specific project, building that data foundation around that one is already a good starting point. Yeah, so many, so many companies out there, 100 different systems, legacy systems where data has to be updated. Yeah, it is, it is a huge, huge achievement to have that in place. But then, oh yeah, then the future will be, will be great.

Clearly, and that's what you want. You want a scalable future where if you, you know, hundreds of more of types of documents, that you're not having to actually, you know, do exponentially more work, right? It, it needs to scale. You need to have the models in place that could represent the few types of documents. With Data Fabric, that sort of allows you from an architectural perspective to have the data in one place logically. Physically, it might be distributed, but that's taken care of, you know, Fabric architecture. And, you know, that's where scalability comes into place. And if you then layer tools like workflow tools, RPA tools, report tools, you know, all of these things to actually automate the processes around documents, then it's a lot easier. You can focus on the problem then. You don't have to focus on actually fixing technical issues because there's some architectural issues or models need to be recompiled or models is not a good fit. So you have to have workarounds in your process to say, well, send this to a human now to actually do the classification. And that sort of just undermines the whole thing. We want to have a future that's easy.

Absolutely.

Yeah.

And that's what AI should do, isn't it? So it should really support businesses with that kind of stuff because that would enable these businesses to do so much more. It will reduce all the backlogs we see in lots of companies, public sector places. Everything, or not everything, but majority of workflows and tasks are around data and documents. Yeah, and we probably can have several episodes of the podcast just about document processing, but in a nutshell, yeah, it really simplifies how things are done within companies. Manual processing of lots of paperwork is not required anymore when once that is implemented and Yeah, everyone should leverage these tools. And yeah, if you, if you like to further play around with those tools or get started with those tools, you have to check out our community, theautomationguys.net. There we have a sandbox where you can play around with different, different things. And yeah, you can also exchange your— with other like-minded people in this community. So So that's the episode on document processing, AI document extraction tools. If you have any thoughts on the topic today, yeah, share that with us direct or in the community. Looking forward to receive all the feedback. And if you like that episode today, yeah, give us a 5-star review on the platform you're listening to this podcast. And yeah, of course we will be back with another episode very soon. And until then, let's automate it.