Dynamics 365 Business Central: Using AI To Predict Late Payments
All right everyone back from the dead so, if you saw the Halloween special suffered a terrible fate from Java 3000 but we updated it again or actually rolled back the update. So, we're good.
Just to get back into normal things, what we're going to talk about in the screen share today is after visual intelligence inside business central. So, we're talking about late payment prediction, today if you get an invoice or you're sending out invoices you're always wondering when I get paid the due date is maybe 15 days from now, but you might get paid 30 days you know 45 days after. There might be some customers that always pay late and some customers that don’t, and your controller just knows right?
Taking that away taking that human element away and having the system automatically predict the payment date for that invoice based on the payment history for that customer is something that a neural network can learn, and Microsoft released an update on that and we're going to look at that now.
Late payment prediction extension
Let's get into the system a little bit and explore this, if I go into extensions right here there is an extension called late payment prediction. This is from Microsoft that's relatively new and it basically hooks up to Cortana for predicting payments for customers so let’s take a quick look at how that works. If I go back here into Cronus USA, go into the search and type in late payment prediction, I can get into this set up and to get this going you have to enable predictions and you can go ahead and create a model, evaluate the model, scheduled it etc.
So, it's again going into the whole artificial intelligence model where you can tweak it, you can also connect this to your own Azure subscription for, you know, setting up the model again. If you need help with this we actually as a partner have now an artificial intelligence department so, or I'm sure your partner has one; if we go out of here and into customers, and I'm actually going to go into the School of Fine Art, this is the demo database and I have seen that these guys have a lot of data or significant number of transactions. I can get here and to navigate, history, ledger entries and if I look at the ledger entries here it's highlighting the ones that are late. The invoices, I’m looking at everything, not just the open entries. If I scroll a little bit to the right I have this payment prediction column and payment conference so, it’s predicting that these invoices here which are due on the 30th of April. We are we're on the 8th of April right now in the testing system, that they are going to be paid on time, but the prediction confidence is low. So, the threshold is probably set high meaning that it's going to give you a prediction but it's not very confident on this.
If I go ahead just to show you how this all works, if I create a document, a sales order and let's just go ahead and do new one to this customer: school of fine art and so what do date is the 30th we were just going to sell them a mug 1 mug for five dollars like so, go ahead and hit post. Another thing that you might be thinking okay it's very low dollar amount and they are used to paying high dollar amounts is that doesn't make likelier that they'll pay the low dollar amount and that's all in the model, in this session right here I'm not really taking the model apart and it might actually be fairly complicated to understand that in the mode.
From the user perspective, we just come in here, and take a look at School of Fine Art , go in to navigate history, ledger entries and we should see that new posting for $5 and you can see that it's actually not predicting anything for the new invoice, but if I go into functions, here, actions, functions, update payment predictions. Again, this is an extension so, you install it. It's installed on a new instance automatically, it pops in with this is going to be on time but again it's a low prediction confidence. It's kind of cool, you should be able to get this to a high confidence probably as more data is in here the confidence will go up and it will tell you if it's going to be late or on time using Cortana artificial intelligence in machine learning, so it's kind of cool.