#39 The Microsoft Power Platform and AI Builder Episode with The Automation Guys

Hello and welcome to another episode of the Process and Automation Podcast with the Automation Guys. Yeah, since we kicked off with this podcast several months ago, Arno and I talk about, yeah, we talk about all things automation, really intelligent automation, hyperautomation, process mining, low-code, RPA, chatbots, and so many other topics in regards to automation. And yeah, we basically covered all the different automation pillars companies can look at to get the most out of automation. And yeah, I know you have read a lot of that feedback as well, what we receive. We receive so much feedback, which is really great, and we would like to really thank you all on sending the feedback and suggestions in. And yeah, by the way, please keep this feedback coming and please rate us on iTunes as well. 5 stars, if you would be, would be really great if you like our content. We really want to make this the place to come back to for everyone looking into automation and get more done with automation, get more done with less in the business. Yeah, and Microsoft Power Platform seems to become very popular with our listeners and we have been to cover recently more on that specific topic. So this was our first episode about the Power Platform and with the request to cover more about the Microsoft platforms, we have been asked to cover also more on the AI and the capabilities and some really practical use cases. And yeah, that's why we thought it would be really great to combine these two areas as part of the Power Platform. We now look at the Microsoft AI Builder as well. So let's get started. So Arno, shall we give our listeners a bit of a background on the Power Platform first?

Yes, of course. So obviously the Power Platform is the Microsoft platform in this high productivity, low-code business app space. So what that means is if you're a citizen developer and you're interested in building productivity apps for either yourself or your department, for example, you could use the Power Platform to build very powerful applications to solve problems and doing so without writing any code. Now obviously, the Power Platform is nothing new. It has been evolving over the last few years into something that's actually quite powerful. At the minute. We have had quite a lot of requests from our listeners really to unpack some of the features of the Power Platform, just to sort of demystify what's inside it, how can you use it, the types of scenarios of business problems you can solve. And I think today is a really good opportunity to kick that off and really to high-level touch on the capabilities of the Power Platform, what's in it, just at a very high level. What we also want to do specifically today is really focus on unpacking the Microsoft AI Builder, which is a set of artificial intelligence capabilities that ships with the Power Platform, which is really powerful. Where you have scenarios that you have to, say for example, turn unstructured data into structured data, plug that into a workflow, or make that available on the form, a mobile device, do some predictions, some sentiment analysis. So before we get started on that, just a quick roundup of, of what the Power Platform is. Obviously it's part of the Office 365 platform. It is licensed as a standalone by each of the constituent capabilities, which includes Power BI, Power Apps, Power Automate, and of course, our Virtual Agents, which is the Microsoft equivalent of intelligent chatbots. There's various licensing models for, for these various capabilities. So for example, you could license per flow or per user. Of course, if you have an enterprise license, that's all taken care of. But again, in a nutshell, Power BI gives us a very, very capable platform to create data-driven reporting insights and visualizations of your data. Really powerful. It's been around for quite a while. Very mature. And of course, you know, as you would expect with Microsoft, it integrates with a variety of data sources, you know, data services. So very, very easy to use. Like I said, quite powerful. You know, it's a no-code environment as well. So it's very easy for citizen developers and business process owners, or, you know, people that's enthusiasts inside a company that wants to create reports. To put this together. Of course, with Microsoft, the community and all of the support, the help, enablement for using Power BI is quite strong. So it's really easy to get started with it. Things like Power Apps where you can build custom apps that solves business challenges using drag-and-drop, no and low-code solutions. Of course, Power Apps is nothing new. It's been in the Power Platform stack for quite a while. Microsoft has added quite a substantial number of features to it and also quite a few updates. It's quite easy to use. So it is becoming more mainstream compared to, you know, your enterprise no-code and low-code platforms out there. Some people still think there is some limitations in what it could do. I think it coexists pretty easily. With other enterprise low-code platforms, we see coexistence with Power Apps, with other no-code and low-code platforms within quite a few organizations. We've got Power Automate, and that's everything to do with automating processes. So this is workflow automations using Flow and also the, the RPA side of things where you could use, um, attended and unattended robots. That's within the Power Automate platform to automate some of those mundane repetitive tasks out there. And of course, all of this integrates well with your Power Apps. So if there's a handoff from Power App application to a flow, which is either RPA sequence or a workflow, everything, as you could expect with Microsoft, is fairly tight to integrate it, very easy to use. Very drag and drop. And there are some, you know, capabilities if you compare the, let's say, the RPA side with something like UiPath or Blue Prism or Automation Anywhere. There's some limitations, of course, and the platform itself is not, not that mature. But I think the target audience for this at this present point in time is really sort of that citizen developer that wants to create productivity apps some automations. You know, email comes in with an invoice, perhaps the invoice gets downloaded from a shared inbox, invoice gets OCR'd, for example, and then the data gets pushed in a database somewhere or surface somewhere on a Power App user interface. So it's these, these types of automations that you can Of course, the Power Virtual Agent, that's a place that you can easily build your chatbots to engage in conversations with your customers and employees. It's got pretty strong natural language processing capabilities built inside it that could detect intents. Of course, it could also trigger Power Automate processes, including RPA and workflow processes.

As well.

So fairly, fairly solid offering from Microsoft. As you can see, it's becoming a lot more popular with businesses and enterprises out there. You know, for, like I said earlier, sort of citizen developer type automations. Of course, nothing stops you from, you know, doing large enterprise-grade deployments as well. But of course, you know, there are certain limitations still in, you know, certain capabilities of, you know, compared to, let's say, enterprise low-code vendors. So that's sort of a, I guess, a very high-level, you know, what is, what is with inside the Power Platform. That's all right. Very good.

And it seems like lots of companies, especially the bigger corporations, they can make good use of it already because lots of that stuff is part of their big corporate license agreement anyway. And they have sort of lots of credits, cloud credits available when it comes to the consumption of these services, the AI and everything which is volume-based. Azure-based. So yeah, they have plenty of capabilities. And yeah, I think that's why it also becomes quite popular. There's a good framework there, there's a good set of tools there, and it is easy to get started with. You don't have to go to another sort of procurement layer to get something really substantial, substantially capable built, isn't it? So yeah, that's, that's why it's popular.

Yeah, and I think one of the reasons why we want to touch on Microsoft AI Builder today is that if you look at the Power Platform, including the Power Apps, Power Automate, and then obviously the virtual agents, I think the Microsoft AI Builder is really a differentiator here. Where it is, it's adding a lot of value to the Power Platform because Microsoft AI Builder is an artificial intelligence platform and it's part of the Power Platform stack. And it's really, it takes a low-code approach to developing and introducing artificial intelligence into your projects. Without the need for coding experts. Of course, there are some limitations in what you can do, but at a whole, I think what it's trying to target is a really large proportion of problems, common problems that exist with inside the business that it can fix. So really, I think what we want to do today is sort of dive a bit deeper into what the Microsoft AI Builder is. And I think, you know, the best thing to start really is, is, is, you know, what can AI Builder do? And again, I think, you know, AI Builder provides you with the ability to optimize your business processes further using artificial intelligence. I think that's in a nutshell. You're making your processes and your process automation more intelligent. So you could use AI to automate processes and gain really good insights from your data and then combine that with the power of Power Automate to obviously create the automations and then the Power Apps to create some of the user interfaces. Now, of course, the Microsoft AI Builder, all of these are built on the Azure AI capabilities. Which is a standalone service as well. The basic idea is that you could train models and build models to enhance the intelligence of your business apps using data you've got, for example, in Dynamics 365, your Microsoft Dataverse, which is, which is of course repository, data repository, a singular repository for all of your data objects. So really with AI Builder, you can build custom models tailored to your needs, and you can also choose pre-built models which is ready to use for common business scenarios. So, and there's a couple of examples we can touch on. Yes, yes.

I think a very popular is always the making sense out of documents in, it's letters, invoices, which are not necessarily available in electronic formats like a PDF. So the text recognition capability of the Microsoft AI Builder is potentially a very popular one. And this was this model of text recognition. You can automatically process text from images, basically scanned images. And there's so much document, there are so many documents still out there And effectively you use optical character recognition to find these text blocks and read the text within the— as well as photographs and all sorts of scanned documents. So the key here is as well to convert that unstructured data which sits in lots of databases and convert them into structured data. So very, very important when it comes to working with RPA. And you could say text recognition or your OCR is sort of the eyes of a robot. So without that structured data, a robot can't really do much, or the workflow can't be as effective as if you have lots of structured data available. So this is why it became Again, a really, really important discipline when it comes to the— or skill, how it's called these days as well— and when it comes to automation, to make sense out of documents and extract information. Like, it could be very simple, it's an invoice, you identify the VAT number, the invoice amount, the supplier name, all that kind of stuff. If it's not electronically available, This is what you would look at, and, or documents, contracts, medical documents. It could be all sorts of documents. And by identifying specific words which are just relevant for your business, yeah, you can make it highly effective over time. So when you train a couple of documents with very specific terms, you can really Yeah, gain lots of efficiency here.

Yeah, and it's all back to, I guess, the old-school OCRing, but this just ships out of the box. So if you do have an application form that needs to be OCR'd, this allows you to upload, let's say, 5 examples or 10 examples of typical forms that get submitted. Let's say it's a a mortgage application or, you know, any type of form where data is in a predicted place. The AI model will then consume the actual documents itself, or it could be images, could be PDFs. It doesn't really matter— scanned, scanned images. And it allows you then to pinpoint certain areas in the document that's of interest. So it could be input fields that you want to map to specific entities and fields. And the model itself, there's a training routine that you train it, you provide it with the, the mapping effectively to extract the data. And once the model is trained, you can publish it and it's ready to be used in Power Apps or a Power Automate workflow, for example. So it's very easy to use. And of course, if it changes, your existing model is still available. You can then retrain it based on changes to the form, and it would recognize those changes if there's additional mappings, for example. And, you know, it makes all of this available without any coding, which is, which is really powerful. I think that's the appeal with all of these things. There's a minimum of 5, 5 documents. That's a requirement. But of course, if you add more, model gets more accurate. But that's, that's basically text recognition. The next thing it's useful with is object recognition. And, you know, this allows you to train an AI model to recognize specific objects or types of objects. So for example, it could be used in inventory management to detect equipment that is otherwise very difficult to identify by sight. So you could take pictures of products, specific products. You can train the model to predict that these pictures actually belong to a specific product. Again, very useful for inventory management type scenarios or in manufacturing, for example, website ordering. So you could take a picture of a specific product, upload it to a Power App application, and it would use the object recognition to understand what the product is. So it could recognize that is something you wish to order. So it's again, very, very powerful feature. No, no code required to do so. All you would need to do is just train the model to index these pictures or these products, for example. And it could be anything really. It could be damages to a car, for example. If you have insurance claim, you could create a model that predicts damages, for example. So, so there's, there's a lot of use cases for this type of object recognition within Site AI Builder. For real-world applications.

Prediction is another really good one. If you have lots of data available, you can use the prediction capability or the prediction skill within Microsoft AI Builder to identify patterns which otherwise would be very difficult to identify manually. You can't go through 1 billion rows of Excel spreadsheets. Yeah, you can, but it will take you forever. But with this skill, you can identify these patterns and use that knowledge to predict the future. And it can answer questions about your data very specifically, like it can come back with a yes or no, true or false, pass or fail kind of output. So it's It's very, very useful for— so one really good example, we use it for predictive analytics to maybe figure out on sorts of IoT data captured in trains, for example. It will monitor all the sensor data and push it all into this huge dataset. It captures every minute what is sort of in these IoT devices. It could be sensors in a train, all that kind of stuff. So it all flows in there, and based on, based on your prediction model, you can then look at, for example, under what circumstances maybe a train might go into maintenance because of some sensor data says it's basically out of the threshold, that kind of stuff. So if you, if you have something like this running, you can prevent potentially a maintenance outage by just knowing the future. And that can then trigger again a workflow, maybe a maintenance workflow where some engineers looking at a specific problem. So combining all these things together will then really produce a powerful solution.

Yeah, and I think for prediction, another common scenario is in manufacturing where you have to predict based on your sales orders what orders you need to place for parts, for example. And a lot of companies do have a lot of historic data, but that's still quite a tedious process. And it's probably for a lot of manufacturing scenarios we've seen, that's still, you know, very, very driven by Excel for a lot of businesses because it's so complicated. And, you know, people want to have flexibility to, you know, put in some formulas to to make some predictions when products needs to be, or components needs to be ordered. With this, you could pull in that historic data, historic decision-making, and then it could make predictions based on your sales, what type of components you need to order to fulfill your sales orders, for example. So again, a very, very powerful no-code solution for prediction. I think the simplicity here is, is, is really powerful. It's easy to get started, and like you alluded to, Sasha, you could obviously plug this into other workflows. Um, for instance, if you need to place an order for, for some parts, all of that can then be streamlined. Um, another powerful capability, um, and this is very similar to the text recognition, is form processing. Now, the difference with form processing is that it's more advanced in the fact that It can predict or extract tabular data, repeating data, repeating sections. So for instance, if you have an application form, perhaps there's just some data in there which is, it's not very complicated with the form processing. It actually detects or detect repeating sections like invoice line items. Again, a very powerful way to train a model to, you know, detect, you know, very complicated data with inside PDF documents, for example. And again, turning that unstructured data into structured data that could then be, you know, used downstream to be keyed into a financial system, for example, using a robot. Or, you know, plugged into approval process where somebody can actually look at the data, ensure that it is actually correct. For example, if it's an order or invoice that's above a certain threshold, for example, it can go through approval workflow. So there's endless opportunities, and of course what we've removed using form processing AI capabilities, the fact that we've already extracted the data, there's no re-keying of information. It's a very, very, very powerful feature, really, really good application for that within cycle automation space.

Yeah, nowadays more and more of these documents come through electronically, isn't it? The PDF documents. Yeah, so this is a very wide use case. Yeah, and the next really big, big element of Microsoft AI Builder is classification. So think about a text coming in, an email with all sorts of content. It might be a customer request, but with classification, you can actually figure that out. Is it actually a customer request? Is it a complained. So yeah, it finds out these patterns within your language and tags and classifies content. It can also tag something which is potentially an angry customer. So with the sentiment analysis, it figures out someone is not very happy, and then you could take that insight, that knowledge, and maybe route a specific email going through or customer requests coming through to a specific agent which deals with maybe angry customers, and or you can classify a specific request coming in as spam, all sort of stuff. Or you can identify this is a name change request when it comes, for example, to insurances and people got married and then they need to have a name change and suddenly that needs to, yeah, needs to process some name changes in the backend system. And combining classification and then again was the Power Automate, was the RPA capabilities, you might be able to go into backend systems and do those things directly, or with specific classification, you will trigger some other workflows, maybe trigger the virtual chatbots, the assistant. So yeah, once, once we know what it is, we can do all sorts of stuff with it and trigger all the relevant processes to achieve more end-to-end processing. So yeah, very, very important capability and And it also learns over time, isn't it? So more and more texts come through. At some point you have the system classifying all sorts of stuff in your company. Yeah, yeah, yeah.

I think maybe as a roundup, we could just cover a few business scenarios. Of course, we talked about the pre-classification, form processing, prediction, object recognition, and also text recognition. But I think it would be good for our listeners if we just provide a bit of a summary on the business scenarios and the type of model you would typically use with inside the AI Builder. So the first business scenario I want to touch on is you have automation to automate, I don't know, a customer application processing. And, um, the model you'll use for that is your form processing engine, because that is something that you want to extract information from a form. You want to turn that into structured data and you want to potentially plug that into a backend process. Another example, um, you want to automate an expense report. And of course with expense reports, there's some receipts. That's typically associated with that. So you'll use the receipt processing model within AI Builder because that has got a very, very good capability to detect any type of receipt and extract the structured data from that. And it's really powerful. I've seen examples where receipts are scanned, they were crumpled, and it's really good detecting you know, the values within site. Another one, and I'll give you a chance as well to run through a few of your own sessions, where you perhaps want to categorize user feedback and what the type of focus is of the feedback. Are they angry? Are they happy? Are they ecstatic? And of course, we could use the category classification because we could, we could look at the the text itself, and the AI Builder would be able to categorize the sentiment with inside the user feedback, for example. And also, if you want to extract insights from your product reviews, when people type in free text on your products and they, they again just in natural English, they, they want to just leave a feedback. If you want to extract that into structured entities. You could use the entity extraction, for example.

You mentioned some really, really good use cases there, and we touched briefly on some where we went through the different skills of AI Builder. But still a very common use case is indeed sort of accounts payable, where an email comes in with an invoice attached. We have that example in so many places, isn't it? So, but it is still a very common use case, and sometimes you get the PDF like Arno mentioned, you can then use that extraction there, but sometimes you still get stuff in from external where it is still a normal sort of a really scanned invoice. Some government places, they're not sending email emails with invoices. They're sending really scans, and then maybe an outsourced partner, they bring that stuff into sort of an email with a scan. And for that, we, in very, very many cases, we use OCR. OCR is really sexy again to really extract that information and to allow that further automation. Otherwise, someone has to look through it and coded manually into the finance system, and this is not the right way. So this is really helping accounts payable departments. It could be as well a purchase order which is coming through in the old-fashioned way. And we— yeah, not every company is modern, unfortunately. So that's a really, really big use case still to tackle. And and was making it that easy from the Microsoft side, it is really worth a look. Otherwise, yeah, you have very old OCR software which doesn't give you that kind of learning, isn't it? So it's usually very rigid, very strict. You need very strict documents, and if a new document comes through, you need to restructure it. So in this capability here, gives you more flexibility. It learns and is dynamic, and this is something you really need. Yeah, and I'm really a big fan of this building, building this prediction models. Yes, I think companies with the data, they should get it in, into this system and really let the the algorithms work on it for specific use cases. And it really, really brings so much insights. And with all this IoT data, I think we don't do much enough. So this was the example I had earlier. There's so much stuff in building management where so much IoT data is in there. So you could prevent so many issues, maintenance issues, and I really want to reiterate on this one, by really analyzing that data. It could be as financial departments as well, they look at data and they can actually predict certain outcomes, and you can as a business maybe get ready for certain events before they actually happen. And some tools bring that already within that tool, but some tools are still very old-fashioned. We're working with very, very old ERP systems as well. To plug that on it will immediately really put your old ERP system or your other old finance system sort of, yeah, it's sort of finance system on steroids by putting those kind of tools and models in front of it. In front of it.

Yeah, exactly. And I think, you know, the prediction element of this is quite relevant because we are generating a lot of data in an exponential fashion these days. And, you know, if you, for instance, look at an example where you want to detect fraudulent transactions, it is an AI capability that you need. It would be very very hard to spot fraud within massive datasets. So again, with, with the prediction model, with, with, with AI, you know, that, that's within sight, the, the reach of, of, of organizations. And of course, with AI Builder, it doesn't just stop with detecting things. It's also, again, on the, the text side of things. It can translate text into lang— you know, any other language. I think there's about 90 languages that's supported. Um, you know, it's got things like, um, extracting, uh, sort of text from a photo and saving that into a database. Object detection would obviously help you to automate inventory tracking like we mentioned earlier. Um, you know, there's even things like business card readers, just give an example of the sort of ad hoc nature that it can read text, but there's a model for that. Although I personally think business cards is a bit antiquated, but again, it gives you an example of the fact that, you know, it, it is very intelligent in, in terms of the variety of types of information it can detect and then save in a database, for example. And in key phrase extraction, very powerful one where you might monitor it, you might monitor social media feeds and you want to detect whenever your brand was referenced. So key phrase extraction would, would help you do that and then create the required automations to deal with that. Of course, that's very hard to monitor that manually. Lots of use cases in this one. I think it will be useful for obviously all our listeners out there if there's any additional ones that that you might have on your mind, run it past us. More than happy to discuss that with yourselves. The Power Platform is in itself quite a broad topic. I think it's high time we actually dive a bit deeper into this because there is a lot of noise out there and a lot more people that is asking, you know, what is it, should we use it if we've already got Office 365 in our state. And, you know, how does it coexist with other low-code platforms, for example? So definitely a topic we'll cover in our future podcasts as well.

Yeah, for sure.

The pursuit of covering anything automation.

Absolutely. Yeah. Yeah, thank you very much for everyone listening again. And as Arno mentioned, yeah, we cover everything, all things automation. And yeah, if you have further suggestions, please visit us on our website, theautomationguys.net. There you will find more, more resources on all things automation, and you can leave your suggestions there. And yeah, we will be very happy to, to pick that up in the, in the upcoming episodes. And yeah, of course, make sure you're subscribed to our podcast and to our newsletter on theautomationguys.net, and that, that will give you sort of all the information when we have events coming up, special webinars, special demo sessions, hands-on labs, that kind of stuff, what we're planning for the podcast community and our loyal listeners. Yeah, thank you very much, and let's automate it. Unfortunately, that's it again. With this episode of the Process and Automation Podcast. If you like this episode, please give us a 5-star rating and don't forget to subscribe to this podcast so you don't miss any upcoming episode. We hope you will tune in next time, and until then, let's automate it!