hello and welcome to today's event thank you for joining us I'm Laura Dillon from MS Dynamics world and we're here for a session all about improving your accounts payable process with machine learning in d 365 fo and ax our presenter is Miko hi tonin CEO of do app and it is truly my pleasure to welcome him here with us today as we get started know that we do welcome your questions please feel free to enter those questions in at any time during Miko's presentation in the Q&A box on your GoToWebinar dashboard and I know that Miko and his team will be reviewing those after the session and getting back to you directly and without any further delay I'm gonna hand things off to Miko to get us started all right thanks Laura for for the introduction and well welcome everybody to the webinar today so as Laura mentioned we're going to talk about accounts payable process and what machine learning and AI have to do with that and particularly focusing on the 365 fnot and MAS before going into the topic maybe just a brief introduction of me so my name is Mika hi tonin I live in Austin Texas I get around 20 years experience in in erps finance and business roles when I got some spare time I'm really really into cooking and Asian fusion cuisine is kind of my specialty I'm also a sports sports fan and attic fan and obviously when it comes to AP and AP automation I could do that power for that so that's the app to go for for AP alright moving on to to the agenda so first I'm actually gonna go through a bit of an introduction to to AP automation at a more more generic at level and why I'm actually gonna do that is is really that that forms the basis for implementing any ml' or' or AI into your existing processes and once we've actually got the basics covered up then I'm gonna explain a bit what are the different terms that we're using what does our PA mean what is ml what else what is AI and what are the differences at between then we'll actually move into more more concrete stuff so we're gonna look at the opportunities what these different techniques actually offer for for finance roles and this is more at a generic level so we're not focusing on on API automation that much yet but that's that's gonna be then the next part of my presentation where I'm gonna give some examples how a male can be used already today when we look at the purchase-to-pay process and then maybe just a few words on on the future how we actually see what's what's gonna happen after we implemented all the stuff we can do with our PA machine learning and nai so moving on which is to look at their what is going on in the markets and this is probably something that we feel a bit sad about that if we look at already all the products and solutions available on on the market for ap automation there are actually plenty plenty good ones out there but still when we look at it especially for the users of d3 six five and ax at twenty twelve we've actually seen that a 75% of organizations haven't really done anything to to implement that ap automation and I think in in our mind that the case for doing that is pretty simple and straightforward because you're basically getting the benefits from day one so there shouldn't be anything that's actually stopping you for for implementing these solutions and maybe just a bit more background why this makes a lot of sense and I think these are also the core elements when we look at machine learning and AI but even some of the the our RP a use cases so I think when we look at the dynamics users manual data entry and offense the inefficiencies in the processes are something that we probably all all struggle with and then then if we go a bit more what happens next that's obviously the workflows and how things get approved and so forth so the manual of routing of invoices that's that's another clear pain point when we see when we've actually surveyed up the market around these topics and also if you look at the third one there which the high number of discrepancies and exceptions that's probably another topic where we can do a lot when we implement some of the automation techniques that I'm gonna talk later on but I'm essentially if we look at the pain points for Microsoft Dynamics users for processing ap really the manual data entry and the manual routing of invoices is something that we can actually create improve by implementing some of the machine learning and AI techniques then maybe just briefly because I think what what we're wondering at times is that up why does it take awhile for companies to move up into ap automation and I think what we actually hear quite a lot is that there are other priorities and I think that's gonna be the fact always there are always other priorities but I think how we look at things is really that if you think implementing a pretty standard ap automation process that's actually quite simple when you rely on the market standard and you're you're kind of ready to make some changes into your own process to make sure that you're really getting the best of class solution I wouldn't say it's it's too complicated so there may be other priorities but I would actually encourage everybody to to look into the benefits where you can really achieve when when you do to implement eight the automation and take take really take really the advantage what the solutions have a have to offer and I think if we look at what we hear at times is that I'm implementing ap automation is complicated not really because in the end of AP automation and processing invoices is a pretty structured process so what we would just say is that kind of focus on on the essential in your your process look a bit around what other people have been doing on the market and I think the outcome really is that it's not not too complicated and then maybe the third one here what would we here at sometimes is that don't fix something that is not broken what we tend to say here is that if you're running a pure manual of a process it may may work well but I think when when you implement even a bit more level of automation into your process it's actually gonna make your in everybody else's life a lot a lot easier who's involved in in the process and I said it is broken but I think we have a lot of solutions currently on the market which you can actually help help you with that automating the manual steps that you're currently having in in the process so moving on so what is really ap automation how to get the basics right before you can actually consider robotic process automation machine learning or artificial intelligence so there are some basics that you need to login into first and I think if we look at the overall big picture when it comes to AP automation in in our mind it's more a way way of thinking rather than anything else and I think for whatever products or systems you're using I think the ease of use must always be in the core because that is for one thing driving the adoption for for the solution so if things are complicated it's it's very unlikely that false in your organization will get super excited around watch what you're doing but if you just make sure that the user experience and the ease of use is there from from the very beginning it's much more likely that you're you're gonna run a very very successful project in terms of let's say some of the areas of automation what we see think there are two different types first of all there's kind of the the pure transaction handling automation but secondly there's the process automation and maybe just to explain a bit the difference is around these two two areas so I think for the process automation when we have all the data available in in our systems we know what to do we have the abreu overflows there there's no point of having extra clicks for for anybody in the system and that is really what we mean by process automation so everything that canapé kind of in terms of the steps within the process automated you you should of course at the same time you need to have the relevant control points follow your delegation of 4:30 and so forth but basically the ones were just clicking the button it makes a lot sense to to automate those and when it actually comes to to the other part which is the transaction handling automation so we know especially when when we look at purchase invoices and invoices we receive that there are actually quite a lot of repeating patterns you get the same invoices from the same Wenders pretty much every every month and even at times the amounts are the same so there's no reason why you couldn't automate your invoice goating and and so forth and I I think this is also where the machine learning really really comes in into play because when there's a pattern that is something that you can easily utilize to to automate the things that you're doing or are being done in in the systems and I think one one important thing what we're really looking into in in terms of machine learning and partially AI as well is that a water kind of the behavior patterns of all the different users and what I actually mean by this is that I'm let's say if you're an approver user you don't actually need to see a lot of the data that is normally available in the system you can actually just focus on on the areas which are really relevant for you approving the invoice and really the system shooter automatically highlight those areas when when you log in a for example to your mobile app to approve your invoice you should only see development data there based on your your past behavior so this is really what it means when we're trying to learn from what what the users are doing in the system and then maybe just to focus on the first step app when when you're automating your accounts payable process really we need to get rid of the paper because I think getting something automated applying some of the machine learning AR techniques it really means that we have access mostly to structured data but at some sometimes also on unstructured data and that really automating paper is complicated but when we actually have all the data on our invoice in electronic format we have all the metadata correct that we need for processing the invoice then we're we're basically good to go and also take take the next steps in automation so mainly relating to the more advanced techniques of machine learning and AI this is something that I I mentioned already earlier I think that we see quite a lot that our clients on the market wanna customize the AP process quite a bit we we tend to say say to our clients and prospects that actually if you just think about the whole AP process from a very simplistic approach it's it's quite simple and I think also the process as such should be quite simple so when we just look at some of the core areas within the process and receiving invoices you receive you process and approve and then it goes all the way to payment so if we kind of really simplify the thinking the process should be simple and straightforward of course we know that they're quite like things around different types of purchases and purchase orders and so forth but I think really we should be focusing what are the similarities and how how to improve the process because that is really when when you're getting all all the benefits from the automation as well and going to the benefits I think automation obviously that will save you quite a bit of money it will save you a lot of time and I think one of the key areas what could even be the most important thing is that you actually came with ability and control to your AP process do you really know what is going on in you you know where your invoices are at if there's a manual approval someone needs to go through you can really figure out where's the universe stock and take that the necessary action although in our mind many of the actions that are currently manual will be automated going going forward but I'll dive into that a bit later in this presentation then maybe just briefly to explain our way of looking at things how we see ap automation and I think this is pretty much similar for all all the other players on the market so as mentioned earlier the first thing is is to get rid of the paper and if it's an email invoice similarly we need to capture the relevant data so that is why we have the scan capture and validation sir which is which is basically the first step in in terms of getting the the data in in a digital format then obviously when you captured all the relevant data then you have the approval workflows up if it's an on pio invoice there's the non P or coding and obviously for the pío based invoices will automatically match against the peels that have been is used in in issued in your d3 6 5 4 ax at 2012 and why I am actually highlighted these parts in in the process is really that I think these are the areas where we can get most of the benefits from from the automation and I'll go back to the user case is use cases a bit later in the presentation but this is kind of preamble for for that so there are a lot of areas which we can automate and I'll let you know the concrete steps a bit later obviously we're looking at a solution there are a lot of different areas they're like reporting archiving obviously you need to capture the the audit trail all the other compliance requirements and in the end when when you're done with the AP part oviously then the invoices sap get transferred to a 2012 md d 365 so let's move on to the RP a machine learning and AI part and I'm not going to technically in this presentation this is this is more morph of you how how you should look at things from from a very concrete approach to Ward's are they the AP automation and this may be something that you've already heard and are familiar with them but anyway I'll open up our thinking a beta here and I think this is along the lines that the market in in general is thinking but maybe maybe just to give you a bit of an over you so if we look at the the RP a the robotic process automation which is typically something where you would get started and this is not typically directly related to the automation either because you can actually automate a lot of different areas in your processes but as as the kind of term goes it's really around process automation and you could actually think that the robot which you're having clicking things in your user interface is more of a virtual employee so based on rules that you've actually said and it doesn't need to be in the same system it can be between multiple systems as well you kind of have a virtual employee which will click through through the different screens and so forth in your systems to make sure that your process is automated and it moves on as as planned I think with our PA it's pretty easy to implement if you take one of the solutions that are available there on on the market go live can can really be be in days so when you just know what are the areas you want to automate in your process up it's pretty simple and straightforward to to get going and obviously for this part you typically use your existing applications and processes so you're you're automating the different steps that you're having in your process and of course this is very rule-based so you define the rules and based on those rules you will actually start to automate your your process I think when we go then to the next step the machine learning so I'd say that's really learning from the data that you have available from from the history it's also identifying the repeating patterns that you would have in your system based on that data so I think these are kind of the key things what do you need to think for machine learning so you have the data and you identify the repeating patterns and based on that you would then be building your application of course it's it's a bit of an iteration so you always have a feedback loop there so you go back and see that okay this is what the Machine thought that should happen and was it actually in line what what the real life is and what we've actually seen seen in in the past and of course as with many data driven application quality data is really the key because if you don't have a quality data you won't get the quality results either so I think this is an area where you need to pay constant attention because you just want to make sure that the data in your systems is high quality because then you will also get high quality results there are quite a quite a lot of out-of-the-box solutions available on on the market already for for these but of course how it normally goes is that if you want to go very very advanced with with your learning algorithms and so forth then it will for sure require some custom development and as with many of the modern solutions it's really about trial and error and continuous improvement when you got the first steps covered then then obviously you will move into more advanced used cases and also better better results as a result of the learning learning process I think the artificial intelligence is then then the third level where you really try to simulate at what a human being would do there and that obviously that would then include all all the behavior patterns not just the data and obviously pretty much all all you have available to try to try to predict the behavior and that is kind of where the algorithms tend to get at get really really complex and I think even as the terms artificial intelligence I'd I'd really kind of emphasize the intelligence when we're at this level also the machine should be able to interpret its own own actions and and learn learn from that not just the data and I think when we look at these solutions there are some available on the market but this is typically where you really really go deep into the algorithms and technology and really see how that would best support your your business and obviously in the end the target is get too close hungry person level of autonomy and when we're talking about these things and I'd say specifically the the AP automation in d 365 and ax really the machine learning and AI they are already there so many many of the products that are linked to the AP automation of the Microsoft erps are already using a lot of these techniques or wisly including our our own solution and I think the RP a part that is something that you can use more more generally in your process automation so it doesn't need to relate just tap let's say to AP automation but it can actually relate to to the link links between the different systems and maybe just briefly how we see the AP automation evolution I think when when we got started it was kind of nearly pretty simple making sure that we have the right of airflow to get up the purchase invoice approved and maybe based based on a certain wind earth the approval routing would follow certain pattern and so forth so kind of then the next step would be the rule-based hub automation so essentially about one good example of that is that if you receive an invoice from a certain Wender the coding of that invoice app would have always follow a certain pattern or if it's a certain vendor there would be always certain approve overflow which food which would result so we're very much based on on the rules available and I think the rbass mentioned earlier on that is something where you can then take more more complex that use cases and also also automate between different systems but then when we go further in terms of the level of automation really the machine learning and AI are are the ones that are getting you closer to the 80% or 100% mark when when you're automating your accounts payable process so this would then be not not just rule-based because you can get to a certain level with the rule rule based applications but really you always have a human interaction you have the exceptions in in the system and this is something that needs to be taken into account and this is where machine learning and AI are are helping big time then maybe some of the use cases for for machine learning and AI so what what can you actually do with these with these two and I'm starting with the beta wider context in terms of the finance organizations and and then then moving on more to the AP specific use cases I think this is kind of the classic the forecasting so when you actually have a lot of data from from the past including the behavior of your customers you can combine that to external data sources like whether that a demand and so forth that is where I'd say and what I like to call the classic use case and that's that's the forecasting so basically when you know what happened in the history you can pretty easily get a prediction to award some future and I think unfortunately even though this use case has been pretty much around I think I would even say forever still still we see quite a little application in into this use case of course if you look at companies there are companies at different levels in terms of applying this but we really see that they're actually a lot of let's say manual face is currently where you could take the advantage of the forecasting use case is just to make your your everyday life easier when you're putting for example your your financial forecast or your demand forecaster together I think another thing what happens always in a finance organization is analyzing the data and nowadays the machines can actually do that very well of course when the use cases that get very complicated the algorithms get complicated as well but let's say if you want to analyze standard deviations for your having let's say in your in your revenue line or your expenditure there are a lot of solutions available already who can actually do that for you and of course when you have the basic analysis app conducted by a solution applying some of the machine learning or AI techniques then you can really focus on the deltas and exceptions which are maybe something that the machine cannot yet figure figure out and I I think also if we look at just the quantity of transactions that typically go through in in a finance organization there are a lot of items which can be automated and I think one of the benefits is really that then when you automate the basic stuff you can really focus on managing the exceptions because as we all know there are always those there are things that we couldn't predict there exceptions that we didn't have visibility to so basically what what we're saying here is that you wouldn't automate the basics as as much as you can because then you can really really focus on the exceptions because what tends to happen is that you're putting a lot of your time any four in into the basic stuff app meaning that you don't have the necessary time that you would actually need to spend on looking at some of the exceptions I think on one area if you'll get a modern systems many of them unfortunately not all they have pretty good API send integration layer so if you look at and I'd say this is probably mostly the RPA techniques you can actually easily automate also between different systems so if you just have access to the interface that you know what's the next thing to happen in in the process the robot can actually do it do it for you and I think there are a lot of use cases going even even beyond the finance organizations where you can really really get benefit just just by putting the necessary rules in into place so that you can take the first first steps in terms of automating your your processes and then I think the fraud detection is something that we see nowadays quite a lot and I think we've all heard heard the stories when there has been a fraudulent invoice or payment request or something like that and unfortunately these are something that go through our systems and checks and controls at at time but I think when we have access to all the data we have the necessary tools that we can use we can easily identify the patterns and actually tackle the fraud before it can even even happen because I think fraud Asajj she's all always a result of certain anomaly which you haven't seen in your process earlier I'll go be deep into this one we talked about the fraud detection when it comes to 2ap automation and then maybe something that we probably haven't paid that much attention to if we look at finance organizations obviously if if you go into any any website nowadays there's the the chat bot which which you may or may not like and it may or may not kill the answers you're after but I think also for finance organizations both for the internal internal folks but also for the outside companies that have questions I think it makes a lot of sense to implement chat BOTS to actually address some of those questions because then again it will save a lot time from the organization to focus on different things when you can at least let's say the more more simple questions or the simpler questions you can actually address by by a chat board which then again is is using some of the data the machine learning and AI capabilities that can be easily embedded in into your systems and and solutions then if we go a bit more into the AP automation or or the purchase-to-pay process I think some of the most simplest use case is obviously matching your reigning rowers against the PIO automatically and I think this is something that it that is very very standard in in the AP automation solutions but obviously I said earlier for you to be able to match the universe against the beo it means that you need to have the universe data in electronic format because then when you have access to the data on the invoice you can easily match it against that the PIO and obviously if there's a purchase receipt or sorts that can be masked in in a similar manner or you actually get get the data from from different sources and of course I think if there's some that's against the PIO that is probably something that we always wish for to make our process automated but I think with the ml and AI techniques even though there isn't a total matza we can actually apply those in in the systems to automate it even though there isn't isn't a perfect match I think a good example is freight charges and so forth some miscellaneous charges which you would have on an invoice and the email and AI in the system can easily identify that hey actually this was a similar charge that I had in the previous invoice so it probably makes sense to order approve this and obviously you can then set certain thresholds let's say if you have an invoice of ten thousand dollars and you have a miscellaneous charge of fifty fifty dollars that's probably okay to go go through an automated workflow but let's say if the miscellaneous charges are like 20 percent of your invoice value then you probably want to have have a manual step there and I think these are all worthy ml and AI capabilities in in the system can can help of course the non Pio invoice coding based based on own rules or machine learning from the existing data you can easily make this more automated so basically if it have if it has happened in the past in a certain way it's it's very likely that this particular invoice from the same vendor pretty much with the same amount would be something that would be coded according according to what you've done done in in the past so pretty pretty simple in a way but we all know that these use cases can get pretty complex and that is why why we really think that the mesh learning and AI can help here - to create create extend and I think if you look at the first two AP automation use cases here that I present did they were actually also the two pain points that we got out from from the survey from the organizations that are are using D 365 if F&O an as1 it well and don't have AP automation implement it because when you implement already these two use cases which are actually available in in the AP automation solutions off of the self you can actually already reap a lot of a lot of benefits and and cut back the the manual work that you would have in in the process otherwise then I think an area of course we want to capture the relevant data from an invoice but what we're investigating at the same time is that what if you captured all the data from an invoice what you can actually do do with that because then really what you can do it's it's kind of limitless because if you have very powerful machine learning and AI available in the system and you have access to all all the data that you have on the invoice then that kind of the automation capabilities are are limitless I think in in terms of looking at the processing power and and so forth you need quite a lot of that but I think that's also something that now in the cloud era is more easily available and I think in the end it's probably not all all the data we want to capture from an invoice but it's all the relevant data and there again the the machine learning and AI capabilities can can help to create extent I think again quite a simple use case I would say is that okay you have a different approve overflows in your system based on on the allocation of authority and of course when if pattern has happened in the past it's most likely that the pattern is something that we should be following now as well so as as mentioned when you're getting a certain type of invoice the machine learning can easily identify what should be the approval workflow being in the system and if there are any other steps that you need to take so I think this is one of the basic use cases but yet again something that can help you to automate a lot because if you don't need to manually think about the very flows rather they are driven by the automation and the system based on the different levels of approvals that are needed in in the organization obviously it will make everybody's life a lot easier I think this is something that I I mentioned earlier on when we're talking about the user experience so where we actually want to get out these that you would only see the things you need to see in a system so if you're approving the invoices you probably need to just see the relevant data for you it may be the invoice picture maybe the comments saying that you actually approve the same invoice about a month ago so it looks like that it you're okay here and it actually fulfills all all the criteria for you approving it so I think what we mean here is that in terms of the user interface and user experience we want to get very proactive so basically the relevant data and and screens and so forth should be there for you not the things that that you don't need of course again if you're an AP user probably the activity is that you need to go through they are more more more complex but at the same time we want to make sure that even those those areas are something that are as as simple as as they can so basically whoever's processing the invoice that the automation couldn't take care of there would be just the relevant data and relevant screens available and I think a few more version on the fraud detection I'm I think this is kind of one of the most powerful use cases for AP automation as as well if if you kind of think it from the risk perspective so one one fraudulent invoice could end up costing a lot of money and unfortunately we've seen seen that happening so I think with the machine learning embedded into the AP automation solutions it can really spot the anomalies so we if there's a change in account number if there's a minor change in in the Wender name or or whatever whatever that hasn't happening in the past it's very easy for the machine learning to actually spot those and then just kind of bring a red flag to the dashboard and tell the user that this is something that you should be looking into at more detail because it doesn't seem to be be normal and I think this is very very similar what what do you see from the consumer business so if if there are ever suspicious transactions within your own credit card obviously the bank will will contact you immediately and ask that do you do you think that this is something normal were you actually the one responsible for the transaction and it works pretty much in a similar manner in the AP automation solutions that we just wanna wanna spot the anomalies and then then actually flag those before anything anything can happen yeah then maybe just to conclude here what we really think is that when we look into the future and it may not be two years from now it may not be five years from now it may be ten years from now but I think really looking at what's happening with AP automation and embedding all all the machine learning and AI techniques and kind of in the wider context to your ERP s as well I think when we look at the basic transactions and and the level level of automation that you can achieve achieve there it's each really hundred-percent and I think for many subsets of transactions we're already there and of course the the more complex the transactions are getting the more complicated it is to automate but we we really and truly believe that with the help of machine learning and AI capabilities this is not just a dream it's actually gonna be a reality in in the future and I think this is something that I would like you to take with from this webinar as well when you think about your current level of automation when it comes to AP automation and that we're good we be it doesn't need to be the hundred-person but if you're getting closer let's say to 70 80 90 I'd say that that's already a huge step forward from from some of the levels of automation we're currently seeing on the market but this is really what I had for the machine learning and AI part and if you have any any further questions going forward you can always reach out to me or reach out to us at dewlap common and we're we're they're there to help so thanks for for joining the webinar today and I hope you actually got some food for thought I'm out of this so thanks for joining excellent Mikko thank you so much for that great information and yes if you did ask a question he and his team will be getting back to you we did record today's event as well so you'll be getting a link to the recording of this very informative webcast so with that I think we're gonna wrap things up have a great day everybody thank you bye-bye