Session Information:

Pushing RPA to the next level with Artificial Intelligence/Machine Learning

RPA can build solutions for financial transactions to train a Machine Learning model that directs corrective actions to robot workers. By combining an ideal mix of bots and humans working in tandem, significant improvement in process efficiency at much lower costs can be achieved. The AI-based models will be built upon financial data sets and recommend corrective actions in order to improve financial management.

AI-enabled robotic process automation capabilities will address financial system errors.  Currently, RPA, however, is unable to handle unmatched transactions, which are those where the invoice number, amount, method of payment or any number of details cannot be reconciled. It can take humans anywhere from a day to months to track down and resolve these unmatched transactions, adding up to billions of dollars in unresolved financial activity. If the models are successful, they will not only save time and money, but they will help increase financial compliance and get a better idea how its dollars are being spent

• Build intelligent tools to spot and fix accounting errors without the assistance of humans.
• Develop artificial intelligence/machine learning models that can go beyond the capabilities of robotic process automation and make decisions about financial transactions.
• The new automated ML models would “be able take irregular, complex financial information and data, make decisions and apply judgments and solutions to a high level of accuracy or confidence without having to have a human in the loop.
• Models will first learn to categorize unmatched transactions as either simple or complex.

Session Transcript:

Bryan Eckle is a partner summit to see, collaborates with clients, and leads the consultant team to identify and solve complex problems with clean and simple solutions. Brian expertise lies in implementing people, process, and technology solutions to financial management, procurement, budget, and overall data analyst, analytics challenges.

He's constantly evaluating emerging technologies across data analytics, process automation, and AI, and machine learning, in order to identify the optimal solutions to the client's business challenges, And with him, we have doctor Helen Jackson, drawing upon more than three decades of practice expertise in physics, engineering, machine learning software development. Doctor Jackson is currently sharing her expertise as a machine learning engineer, which summit just see?

Doctor, Jackson and Brian, it's a real pleasure to have you with us. Our audience is very much looking forward to your presentation. Thanks for taking the time, and sharing it with our global audience.

So good day, everyone, and thank you for attending.

I'm going to, our topic today is pushing RPA to the next level of artificial intelligence, Machine one.

Move on to, I'm going to give a little bit of background.

Um, the table of contents, here, I'm going to go with some basic things that, and a lot of technologies, I'm going to define everything that's involved. I'm going to give them a path that we've outlined here.

It's only to see the path to robotic process automation, and then I'm going to talk about what fits under the umbrella of an intelligent automation and artificial intelligence and machine learning, which are the compilers at arrival of this path.

And then I'm going to talk about the combination of RPA and machine learning, and then use cases.

OK, so what if we could harness the power of their products and their systems to Correct. The walkers and this is achievable was what we call 0 to 2. So, here's, uh.

A diagram we've made up, you start up your data.

As you can see, the data is, could be any number four, OK, and then you, and, by the way, this is an audit on an AWS platform, OK? Then, that, in turn, has fallen to ML model, which happens to be, in this particular case.

AWS, sage Maker, and then we will get dry and Hawaii.

And when you're dealing with AWS, in order to connect to it, you have to, you have to have an airport. And at that point, you, you, you, you are rich at you.

You are ready to go to your, Yup, take your predictions to whatever you have, which could be, oh, and for the purpose of this, Talk, RPA.

So I'm really glad I'm going to define the technologies, which, and probably most people in the audience, are very familiar with.

Artificial intelligence is actually an umbrella term that covers many of sub domains, OK.

And but it is describes the bar using machines to imitate intelligent behavior in order to solve problems ML, machine learning is a technology used to achieve artificial intelligence. Algorithms will all be able to load on itself.

A subset of the sub domain of that is deep loan, which is the further development of machine as natural language processing, which is the branch of artificial intelligence that deals with communications. And this is actually a colloquialism, all computational linguistics or semantic identification technologies. And then there's computer vision, which is the process of understanding digital images, videos, using computers. And it seeks to automate tasks that the human, the human vision can achieve.

We do use a subset of that quite a bit in document understanding, called CR.

Relative on an technologies, IA, which is also an umbrella term, intelligent automation automation, it's also call hyper automation and cognitive development, and it's a combination of artificial intelligence, Machine learning, process automation, or anything that is used to create smart business processes and workflows that line up on their own, Then, key concept feel about it. Processing. Information is a form of business process automation, technology used metaphorically.

Those are metaphorical software robots or on a telling artificial intelligent digital world.

And I thought I thought I'd throw in all the time I went to 27 55 as emerging technologies didn't need that a standard definition and some standardization.

And that's the purpose of Tripoli, 27, 55.

So, intelligent automation, very quickly. As I said, isn't, it's an umbrella term that can combine robotic process automation?

With a artificial intelligence, machine, traditional, robotic process, automation, if you stick it to business rules, combining it with machine learning increases in efficiency, RPA can be a means of injecting intermediate legacy enterprise systems, Which I think I showed in the previous slide.

Of the path from zero to AI, your choice of RPA product, Can your choice of RPA, can be made with the choice of a machine learning model in a secure cloud?

Next.

Well, a little bit more on intelligent automation platforms.

Um, we, UI path has the ability to teach software robots to intelligently process documents using AI.

UI Path also has something called a Fabric for Machine Learning.

However, when you are, you can include your own custom machine learning model on your path.

Document understanding can be achieved via machine learning within AWS, AWS, and a product of all of these.

So, again, a little bit more on the definition of artificial intelligence.

It can be cognitive automation, machine learning, hypothesis generation analysis, natural, language processing, and intentional rhythm.

Mutation producing insight and Linux at or above: human capacity.

The diagram that I have on the upper left shows that machine learning is a subset of artificial intelligence, and deep learning is a subset.

She ... machinima.

And, I just had a little box, the animation, oh, training, going on inside a deep learning neural network.

So, a little bit, real quickly, on the difference between traditional machine learning and deep learning.

The main take, one of the main takeaways. If you look on, the lower left, given, the plot on the lower left, is that, given the same amount of data, the performance is much higher deep neural networks as opposed to position.

Also, you can look at the diagrams on the right, for machine learning.

There are much less hidden layers, OK, so you have a hidden layer of output, whereas, with deep learning, you have many more.

You can have thousands of layers in between, input and output layers.

Now, what machine learning can do, as related to document understanding what can do a lot of things, It's huge, But, for the purpose of this discussion, we want to know what it can do is related to document understand.

Machine learning learns from the bowl, rather than through a given algorithm.

Predictions of classifications can be made. And this is somewhat similar to the first to the slide that I show, the path from zero to AI. We're bringing data into a machine, or algorithm, and output, OK, in a system of invoices.

It can learn to potential taxonomy features, and by taxonomy, I'm me, the taxonomy of a typical, typical invoice, for example, food, their amount a times, etcetera.

So, the difference between robotic process automation add, um, artificial intelligence, right, very quickly, is the robotic process, automation of the robotic processes, such as robotic desktop automation and robotic process automation, are data driven, well, some ashame learning.

They are, they, they, they are simulation of intelligence by hanging. On the most fundamental level, RPA is associated with doing well and ML.

We're thinking and learning RPA suitable for automating abrupt work of e-mails, for example, and putting them into the FIFO, creating a bill and then all the accounting software.

On the other hand, AI is required to intelligently read the invoices, extract the pertinent information such as invoices, invoice number, supply, an invoice, due date, and so on.

Again, a little bit on IP, I saw a few years ago in which there's a need for standardization in so many areas, and artificial intelligence, because it was just all over the place.

As far as the final, so I triple E standard, ways of thinking, has come up with a definition, or RPA, which is RP and refers to the use of a software that uses business rules and predefined activity, Cory Rafik to compete.

Yeah, I could claim the eponymous execution of a combination of processes, activities, transactions, and towels, and one or more related software systems to deliver service with human exception management.

So although this talk is about the combination of RPA and machine learning RPA alone, has experienced phenomenal growth over the past few years, it has a compound annual growth rate of 29%.

And if it is a, it is projected to have a market size of two point seven billion by the year 20 20.

So now let's look at what RPA can do.

We're going to apply RPA streamline operations controls.

RPA can just look at the transformation process, uh, of finance and accounting systems by eliminating manual processes to reduce costs, improve accuracy, celebrate the national reporting, and empower employees.

Software robots perform event driven computing tasks such as those with a mouse or keyboard.

It has tools that allow the automation of a series of repetitive tasks.

For example, open up a PDF file from an e-mail and in other applications, such as Oracle Panel, run a report out of an application, and enter the data into another application.

Search for source documents on the web site, Download them, and then store them into SharePoint sight.

one thing to be noted is RPA is not smart, Uh, so, as previously been shown in cruise line, you can automate processes that are rural, but it is not intelligent or able to law.

An example would be reading the content of an invoice and making a decision, as opposed to machine learning.

Which labs.

More RPA, which is why we're, this talk, is about combining it with machine learning, is an another example.

Currently, RPA is not able to handle unmanaged transactions, which are those where the invoice amount, number, method of payment, or any number of other details cannot be reconciled.

It takes humans anywhere from a day to month, to try and resolve these options, adding up to billions of dollars that are resolved by financial activity.

So.

In the next slide, I am going to the demo, a side-by-side comparison of the process of the processing of invoices by Heyman and a robot.

So in this case, the robot will perform the following step.

It's going to retrieve electronic invoices from an e-mail.

It's going to download the invoices folder.

It's going to extract the details from the envelope.

Yeah.

He's got a great field.

Will this look here, see if you see the difference in time?

Between the human and the robot.

Yeah.

Yeah, OK.

OK, so we see the robot is finished at the three minutes into Seconds.

I'm not going to go all the way to the end, where it shows that the human is going to take eight minutes.

I'll just them, As for the video.

OK, so, the difference here is Law.

Um.

Is that it took the Robot ... process different voices. Whether human took, eight minutes and NaN. Also, in this example, the human made an error when reading one of the bills.

OK, so now, RPA and ML, how does the mix of RPA and machine learning fit into intelligent automation, combination and develop artificial intelligence, machine learning models that can go beyond the capabilities of robotic process automation?

They make decisions about financial transactions aided by deep learning.

Automation of scripts, in some cases, can be accomplished in hours, as opposed to as much as months.

Mix of RPA and machine learning, the ... machine learning be incorporated into RPA.

Well, RPA can build solutions or transactions to train a machine learning model that works correctly, corrective action, to robot workers, and I end up taking A, the last part of the slide that showed the journey from 0 to 8 to show that this is a last part of it. By combining this ideal mix of working in tandem significant improvement and process efficiency at much lower cost can be achieved.

The ad based models can be built upon financial datasets and recommend corrective actions in order to improve financial management.

Erawan that AI enabled, robotic process automation capabilities, well addressed financial system.

Another thing that RPA and ML do in combination and RPA, also, is they eliminate the human in the loop, useful validation, machine learning models.
19:19
Validation is built into the model, as input data is divided into training.

Not only automated machine learning models will be able to take regular pamphlets, financial information, and data, and make decisions and apply solutions to a high level of accuracy or competence, what will happen to have a human in the loop?

So, now, Oh.

So now, we've gone from zero to AI.

So now, the next thing that I'm going to do is give some use cases.

Some of the most successful, I'll do three of them.

First, is solving unmet financial transactions by combining RPA, all machine learning.

So, the goal is to solve a match financial transactions by combining RPA and machine learning, unmatched transactions are repetitive or complex transactions that require elaborately search and decisions such the billions of dollars worth of unmatched transactions and the outstanding rating, inaccurate financial statements, an audit exposure?

Yeah.

So, here, I'm going to give a video, which is on our website, showing me the combination of machine learning and RPA, and you all path, which is invoking AWS, sage maple.

The handle on that.

Oh.

Good.

Yeah?

Yeah?

Let me think.

Yeah?

We do?

With them, Margaret, are the people there?

Cool.

Yeah.

Millimeter.

Yep.

Yeah.

Millimeter, I can do that.

Boom!

Whoa!

In the mountains.

Poland?

No.

We try to do that.

Aye.

What does one go?

What is display?

Don't worry and better.

Good, Good.

Yeah.

Bye.

Don't do.

Scott.

nine is fun.

Come on.

Yeah.

This was the Republican state.

No.

Prediction back.

Isn't that the victim?

The end goal.

I'm not this out to be go into the blog.

Or maybe even for humans.

That would be the case.

This is the pivoting.

They've been thought of either.

You can see that next.

Then see it on something like that disappeared along the way.

This makes it easy for us to do them all.

B We.

OK, This book.

They'll see you, too.

I love it.

Sopa a bunch of properties on it.

Thank you.

Yes.

Run it in the title.

Me, Stephanie.

Then, the output from this.

So, good.

That's, that's detail the CSV file.

This has been easy.

Opinion.

And now, need each night.

Man, yeah.

nine.

And Pass it to me.

The stage. Mika?

You get a prediction once a week.

Same obligation.

Now, based on that prediction, you go to PayPal.

This action, and, again, be anything you want. It can be looping.

I'm back.

It could be bought.

Are you sending them you?

Millimeter.

Now, there are the next line, you can see in the mix, context information.

Will teach me a prediction back again.

It's insufficient funding.

Man, you get the option called principally.

Yeah, that can call. So, I'm going to stop there.

And what this shows is that it's able to make decisions for large transactions.

No.

OK, so.

The next use case is Accounts Payable invoice entry with Machine Learning.

Started to see created an automation process to process PDF scanning, process PDA and scanned PDF using computer vision and Machine Learning.

The challenge is an ecosystem accounts payable invoice formats from suppliers, PDFs and scampi, the challenge, the accuracy of standard OCR capabilities, OCR capabilities and event automation.

The solution would be combining RPA with machine learning in order to identify key AP invoice taxonomy, such as invoice number, what we've been looking at, slides, address date, amount, purchase order number, without pretty, pretty five business groups, the data is entered and to enter in an Oracle ERP for approval and payment.

This is combating RPA and machine learning allows for greater accuracy, less maintenance and higher return on investment across time.

We approach this with a change in the process of theory, document quality accuracy without writing Pacific rules, or changes in document format.

Next slide is going to be AI document understanding and another use case, which is similar to what I've been demoing, extracting data the PDF endorses analysis and update accounts payable systems. And this is built using UI path Platform.

So what you're going to see is, a bot reads the scan documents from a phone.

It digitizes every document.

It extracts the desired attributes, like account numbers.

Do they add rules, it opens a validation station for the user to validate the details, writes the details into Excel.

I'll start this.

Yes.

I'll run this for a little while, showing how it's extracting the desired attributes.

So, I'll for this to the point where, um, it's actually putting the information into an Excel file.

So, in summary.

Um, harnessing the benefits of the mix of RPA machine learning is feasible on several platforms.

Therefore, artificial intelligence, machine learning models, are able to go beyond capabilities, robotic process, automation, making decisions about possible.

We have demonstrated the feasibility of going from 0 to 8.

Sound literacy specializes in delivering data analytics, system integration, and intelligent automation by combining machine learning and RPA.

So that ends my part.

Brian, you'd like to say something.

Thank you, Helen. That was, that was great. I think the key there is, it's all about the data.

So the biggest driver to injecting AI into RPA is, is getting the right data.

And so if your organization has good historical data sets of images, could be database records, could be ERP systems, logistics systems, procurement systems, you can use the historical data in those systems.

Get that data AI ready by labeling it, and then use that to build machine learning models.

Once you have those models, you can easily plug them into existing RPA, business process management solutions or ERP solutions via those API endpoints.

Terrific, Terrific, thank you, doctor Jackson, thank you, Brian, for the presentation. Doctor Jackson: Doing doing your video, the audio was a little bit hard to hear on, on some of the videos on YouTube videos that were embedded. But I think people got the essence of the demonstrations that they, that you're sharing with us.

Are a number of questions that came up, so I'm gonna bring them up. And if you and or Brian would like to address those, that would be great. The first one that I'm gonna get into. It's a, it's a fairly technical one that that Misha identify this is one of our participants. She says, that UI path has document understanding AI fabric. And her question, is that, why was AI fabric selected. And not document understanding for the invoicing processing project that you shared.

Yeah, so I think you saw a couple examples there. The invoice example actually uses AI fabric, and document understanding.

The unmatched example was our own custom model, custom integration with machine learning, which was actually done before the UI path AI Fabric product was even released.

So that's, that's why you actually saw a couple of different ways that you can connect RPA to AI those examples.

Thank you for that. There is a perfect segue for the next question that comes from Cat. And Cat is asking, would you share your experience with adding your own custom machine learning model in UI path and how that happens?

Yeah. So I think you saw the graphic there, the zero to AI. And so it all starts, again, everything starts with data, So it's all about getting the right data.

A lot of the organizations we work with find that they are very great data locked away in all these back office systems.

So those back office systems have captured the data of these very manual, repetitive processes, and if you can unlock fat, take that data, label it, get it AI ready.

You can use that to create some great training datasets to build ML models from, once you have the ML model built, you're doing your running experiments, you're picking the best fit, so to speak. And then once you have those built, you expose them via an endpoint.

And then you can really connect them to any system, your existing back office business systems. You can use RPA as a connector, you can connect them into existing automation processes, or even business process management solutions, like a ServiceNow, an Appian, or those type of tools as well.

That's excellent. The next question that come, that has come up is around the cognitive services, and the question is, whether you leverage any third party, cognitive services, in the work that you do.

Yeah. We, I think, a lot of what we did, we labeled it, so. No. No, Cognitive Services. All of the, we do use as far as, I guess, if we're talking about how we build the models, that's anywhere. From using Python libraries to, there's also some, some really good auto ML services out there.

We use AWS, Sage Maker, we use databricks, ML Flow, those are some of the ways we build the models for on the data to build the data pipelines.

That varies.

We use Stream set's, Amazon Gulu, those type of things, is as far as like, process mining tools or things like that. We don't. We've experimented with some of those but we're not currently using any of those in this solution.

That's excellent, Thank you, and yes, please.

That and keep submitting your questions as we have time here to discuss them. Rose Blair had a question related to audio information.

Perhaps the question that I'll ask is that if you could provide links, I assume those are YouTube links to the videos. There are available for consumption. If you could provide links to those videos, that would be quite helpful. And if you're able to do it. Now, you could put it on the chat, or if it's not easy to get to it now. If you could let, let our conference leaders know about it. And that we can send to the participants after, after the conference, when we send them other materials next week. But there's a question about, you know, can I get to the YouTube videos that you share? And I guess the question is, are those public videos or their private videos?

I believe all the ones we shared are just, they're publicly out there, there.

We, there kinda janeiro's sized, so to speak.

I have been able to bring them up on you, too, OK, great, great! So we should be able to get, for those in the audience who are asking about it, we should be able to get you the links for that there. And then either through the chat today, or later on, the post event communication will, we, should be able to get those links to you.

Very well. The other theme that has emerged on the Q&A is related to the concept of smart robots.

People know, the RPA vendors, if you will, they have been talking about smart robots for, for some time. And most of the time, this is the context of adapting to dynamic environments, and now you, the website changes, and the robot smart enough to, to kind of adapt to the new reality.

What is different from kind of that dynamic environment adaptation to what you're prescribing of what you're sharing here today?

Yeah, so, you know, I think, traditionally in RPA, if you're just writing a bunch of business rules, those are very static and don't change over time.

If you can limit those business rules, or include, include those business rules, an overarching Machine Learning model, or, or some type of AI, it just, allows your workflows to be so much more dynamic, and more resilient over time.

That's what we've found.

And most of what we've done is really driving dynamic automation, related to these back office business processes. So, very resilient as to the ability to grab data from an image, You know, even whether it's poorly scanned.

And the reason why we can we can be accurate with that, is because we've trained those models using your data.

We've pre labeled them using the last, let's say, three years of your data.

So that that's made those models very custom to your organization, to you, to the type of data you're getting, and very resilient when they see new data as well.

Excellent, excellent. And the 1, 1, 1 final question here, we have, from the audience. And this is an interesting one. Natural language processes is an interesting field because, I am coming from a semiconductor background, I was always very surprised that it takes a whole lot more processing for natural language than it does. For machine vision, for example, the chips for natural language processing have to deal with many, much more, in terms of processing.

Then, machine vision, which you wouldn't expect, that should be the case.

So the segue is that, are you, how, how, what do you see in terms of an evolution for RPA and intelligent RPA applications.

Are we seeing a bit more of natural language processing, becoming part of that, if not now, in the next 12 months or so. Or do you still see that as kind of like, not a really good fit for RPA applications?

No, I think it's, I think it's a great connector. I think the text, either text or voice interaction with these automations is crucial.

Very important component and so that's the NLP piece of that, is a vital part of it. Absolutely.

So, I think it is time. Yeah, go ahead.

I like to say that applications that we're developing the machine learning, but document understanding, what we do combine computer vision, use it in the first apps, and then we do natural language processing, we do some combination of it.

So, because the fact of identifying techs accurately good point.

That's, I think that's a good, good wrap on that.

Farewell farewell well. Ladies and gentlemen, our time is up here, I want to say thanks to John Driscoll, who actually found the videos on YouTube, has already posted as a question. And I hit reply to everybody. So before I close the session, look into the message that I sent, it shows a reply to drawn Driscoll and it has at least one that the Human versus RPA Bot Invoice processing video link is in that answer. So copied that. And so, you have at least one of the links already. And I want to thank, again, doctor Helen Jackson and Brian Echo for being here with us. Tremendous view of what's going on in the world of RPA and intelligent automation. Thanks for sharing your your expertise with our global audience today.

Thank you, JJ.

Thank you for having me.

Thank you, ladies and gentlemen, that was summit to see leadership. Sherry, the greatest insights from RPA, machine learning, and artificial intelligence, Of course, there's so much more that we can talk about. And in our time, was limited, but I hope that you got some good insights from that. And there is much more to learn. We're going to be wrapping up the session, and at the top of the hour, you're going to be meeting me back here.

And, I want to talk about, these are transformation, business transformation and cultural transformation. There is no successful digital transformation without business transformation, and there is no successful business transformation with our culture transformation.

So, we're going to talk about the elements of excellence, innovation acceleration in gradient during organizations. How do they bland ideas, methods, technologists, and the people to achieve great enduring performance? So it's going to be a higher level presentation on what sea level teams are focusing on, to change and adapt in the decade ahead. So I'm gonna be sharing that benchmark with you at the top of the hour I hope that we can engage at that time, and looking forward to, to our, to our interaction. So, closing the session for now. See you back at the top of the hour. Thank you.