Deep-Dive into AI and Digital Assistants for Collections

Kush Kumar:

Thanks and thanks a lot for getting all the names right. So good afternoon everyone. Thank you so much for coming here on a Monday afternoon, I know it’s been a long day. We already have the popcorn, Marinko, and Scott. So I thought about how I should start this session. So the best way to start the session is, I thought why not start with wine right? So what do we know about wine? And I’m going to talk about order to cash. It’s not a wine class but definitely, I’m going to talk about the order to cash.

Kush Kumar:

So the US is the biggest consumer of wine. What do we know about wine today? We know the five S’s right. The see, swirl, smell, sip, and savor. And we also have our own choices of grapes right? Which is Cabernet Sauvignon or Merlot or Pinot Noir. But do we know when was the first wine produced, any guesses or where was it produced? The first wine was produced in 10000 B.C. And it was in Armenia. So since 10000 B.C. to now, all these years of wine that you drink on a daily basis or on a Friday has come a long way. The technology behind wine has changed drastically. There’s a different science altogether in this stream that’s called ‘Artificial Intelligence’. With a V right. So when I talk about that, artificial intelligence has all these factors. When the wineries are using, you know the best fermenting technologies they actually, see you know, if you see here then what is the best crop. What is the best climate for that crop? What is the best temperature or what kind of soil will give us the best yield. Today data is driving all that. 5000 years ago and a couple of decades ago wine was actually made by crushing the grapes by your feet.

Kush Kumar:

That was a manual process. So the whole point here is if an industry like wine has completely transformed from being manual to using AI and ML, why won’t we in our daily lives, in order to cash embrace these AI and ML technologies right. And I think that is what Sashi was talking about in his keynote that we have to be the agents of change.

Kush Kumar:

So moving to a couple of years back, three years back we thought that a lot of our little cash vendors out there. But they are using point solutions or they offering point solutions, right. So they are, there’s a vendor for billing, there’s a vendor for cash application, there’s a vendor for collections, but a couple of years back we thought why don’t the order to cash professionals have a platform, right. And that’s why we move towards the integrated receivables platform, where all your order to cash functions are talking to each other and are not working in silos but are working in tandem. Then we thought about if you have one platform how can we empower that? How can we empower all the teams to take strategic decisions and the right decisions?

Kush Kumar:

That’s where we started with Rivana and that’s the AI and ML engine that we kind of used in our integrated receivables platform. So in collections just a use case in collections. We started with predicting the payment date. So as soon as an invoice is generated the system based on certain parameters like region, whether it’s A-Pac, it’s North America, it’s Armenia or whether it’s the dollar amount of the invoice or it’s some other factors based on that it will predict that you will get this invoice paid on this date.

Kush Kumar:

So that was helping the collections team if you’re familiar with deductions and disputes. In deductions, we use the same parameters to predict whether it’s a valid dispute or an invalid dispute, based on the whole pattern of the customers. So that’s where the whole AI and ML started to take pictures. I’m going to give you a formula that you can go back and use and that’s the secret source that we have. And you can just go back and start using it and all your customers are going to pay on time. Any guesses on the formula.

Kush Kumar:

I asked my data scientist for that and he gave me that one slide. You can note it down, It’s pretty simple, right. So this is the whole data science behind your AI and ML. But, to just dump it down in our language there are two kinds of factors. One is the invoice factors. And second is the customer factors. So the invoice factors. Okay. What kind of an invoice is it? What does the amount of the invoice, when was it paid, when was it created? Whereas, there are customer factors in like what percentage of invoices are not paid, what kind of percentage of payments are partially paid, when do they usually pay. So all these are factors that are taken into consideration and then we came up with three models. This is the whole crux of AI and ML. So the first model is the binary classification. As the name suggests, it will just give you whether the invoice will be paid on time or not. Yes or no, right. So that’s the binary classification. That’s how we started with it. But then we thought it’s not helping the teams, right because it’s only giving a binary answer.

Kush Kumar:

So we thought why not segregate the customers into various buckets and classes and then we use another algorithm out there called The Multi-class Algorithm. So the multi-class algorithm will give you again the yes and the no, but will also give you a date range. So yes, the invoice would be delayed and the payment is going to be made between 16 February to 20th February, there was a range. But at the end of the day, we in stealth thought that we didn’t want the range, we wanted a date, we wanted an exact date and that’s where you use the most famous algorithm in AI and ML, and that’s the Random forest algorithm. If you have played this game of 20 questions where you ask each friend 20 questions which is a favorite movie and they’ll give you my favorite movie is The Washington Post, then your last 20 questions why. Likewise, you ask each and every person and they give you 20 reasons. Then you take an average of that then you decide which movie to go to and watch.

Kush Kumar:

So that’s the concept behind random forest algorithm. It will take each invoice, see what are the parameters of that invoice, see when was it paid, and then put it back to the algorithm and predict the next invoice. So it takes an average of that. And finally, we have the, you have already seen it in the keynote is the Freeda, is the virtual assistance of how virtual assistants can actually help us empower the use of the AI and ML. And here in the next section of the presentation before I talk to Scott and Marinko, I just wanted to throw some light on what is NLP, if you have not heard about that earlier. So that’s Natural Language Processing. When you talk about Freeda, when you see Freeda, there are two kinds of attributes to Natural Language Processing. I’m just kind of dumbing it down for everyone. It is called the entity in the intent and the intensity. When you go to Amazon and you search, Adidas shoe of size 8, so what is your intent? Your intent is to find and your entity is that Adidas shoe of size 8. So the NLP actually works on only these two attributes, intent and entity. So whenever you give a command to Freeda it is searching for that intent and then it’s searching for the entity. So just to give you an example, you will see what is the number of open invoices for this company. The number is the intent and the entity is the invoices for that account.

Kush Kumar:

Now, obviously, we’ve already talked about that, so now I’ll introduce the change agents, Marinko and Scott. So, Marinko and Scott, welcome to the panel. You have recently implemented certain technologies that we have been kind of talking about. So Marinko starting with you. Okay, before I ask you a question Marinko, I know a funny story where you got a call from a hospital, an automated call. And the call said something like, pay outstanding or fall sick again, something like that.

Marinko Marijolovic:

Hopefully, I won’t get sick again, but you know you got to pay the bills first. Yes, I live in Cleveland Ohio. And by us, we have the Cleveland Clinic, which is a major hospital system in the country. And so you know they send you a normal bill as everybody else does and then if you don’t pay, you will be sent to collections. Well, I don’t only pay the bills in my house, it’s not my fault, but I got sent to collections. And usually, the amounts are not very large because it’s either the copay or whatever is leftover from the insurance company. And you will get statements online and things like that, but it was kind of interesting that they, this has been for a couple of years now, but they actually have a robotic call and the system calls you and says you know your past due and it gives the option that you can actually make a payment on there, or if you want to speak to a live person you can also hit a button and you will do that. It’s kind of interesting because they’re dealing in really small-dollar amounts and if nothing else they pinch you, so like in my case I ignore them.

Marinko Marijolovic:

I guess I’m like okay, I’ll talk to my wife and see what’s going on but at least it’ll be a reminder for me, it’ll be a tickler to kind of say, hey you know what that’s right, I got the outstanding bill. But if you think about it having a human doing that works for such small amounts would never really be feasible. And they always call like they almost kind of know like when you’re home. So you know I just you can program the system to make the calls when you know, whenever you want. And that’s the same thing like, even outside of consumers if you’re dealing with small accounts that have a small-dollar base you can set up the system. And that’s kind of the initial, I guess like fake intelligence maybe not artificial intelligence, but that they implemented and I think you know we can do the same thing in our environment and especially what Sashi kind of showed earlier, you can really soup it up and take it to the next level.

Kush Kumar:

So you mentioned that when you talked about these robots for the small-dollar invoices. So obviously there are two ways to look at it. One is that either you write off the small-dollar amounts or, if you can have a robot do that, why should it fall through the cracks, right. So it initiates the whole process of, maybe I have to pay them, right.

Marinko Marijolovic:

Right. I mean you know once you put the system in, obviously it’s already in it can call 24/7. If you got different time zones whatever and you know you can obviously send dunning notice and automate that process. But if you kind of want the next point which is it is going to leave a voicemail if nothing else you know on the answering machine or wherever, a person can say I wasn’t aware, that’s like another pain point. And if there are small dollar amounts, you know if it’s 50 dollars or even a couple of hundred dollars, it may not be worth you know depending on how many accounts you have. But if you can get some of those low-value tasks out of the way, you people can then work on the higher dollar accounts.

Kush Kumar:

I just got a question for you. You recently implemented the in-app calling functionality HighRadius, right.

Scott Phillips:

Yeah.

Kush Kumar:

So what are the benefits you are seeing right now and how does it affect the team’s productivity, efficiency or the whole collections process? Could you throw some light on that?

Scott Phillips:

Sure. So we’ve had the in-app calling feature live for a few months now. And one of the benefits that we’ve seen out is recording the transcript of the call. Right now we’re actually recording the voice and we’re getting permission to record voices during that call. And we’re actually typing in what the outcome would be, so I think we’re helping build some of the use cases that will be in the autonomous collections.

Scott Phillips:

By going through that we’ve been able to go back and look through some of those calls and find certain calls where we may disagree or we would have put a different outcome or a different payment commitment or have forwarded to our sales team. So it’s actually provided a good base to develop some training for our team. So that’s one of the unexpected benefits that we’ve gotten from it.

Scott Phillips:

It also helped us partner with the commercial team and giving them this insight, in some of these activities that the credit department is performing, that they may not have known in the past, that they should be having more interaction with their credit manager that’s assigned to their accounts.

Kush Kumar:

So do you think as a logical next step, the autonomous mode makes the most sense because then you’ll have promised to pay that will be automatically logged and that can help you go back to your credit team or the order releasing team.

Scott Phillips:

Yes absolutely. So we’ve been live with the HighRadius collection module for, I think February is actually twelve months, so one year and we came from an environment where collections were being managed on spreadsheets and just being tied to notes and things like that. So the initial feedback from our 30 person credit team was that they were very much valued down. Now having a tool is very much valued. Having a tool and being able to have an instantaneous picture of the entire AR portfolio and see what was happening with that and how it was being collected. But one part of the feedback that we’ve gotten as we continue to use is that there are multiple screens, there’s a lot of features in the collections module, but those require clicks, those requirements go into extra screens.

Scott Phillips:

So now that they’re in the mode of using a tool they want to make it more efficient. So the autonomous collections tool is very impressive on the demos that we’ve seen and we’re very excited to take that next step and to give our credit team what they’ve been asking for.

Kush Kumar:

Thank you. ShurTech was one of the first, I think one of the earliest customers to embrace this technology when HighRadius came out with AI and ML, right. So you piloted it with HighRadius, the collection simulation.

Marinko Marijolovic:

Right, yeah, we did a simulation with our data, they created a model and, to kind of see if we followed the recommendations for the payment predictability. And we noticed, you know, a significant benefit to us here. So, if you see on the screen there, you know, we kind of categorized at 1 through 15 and 13 or 15 plus. The biggest difference would be for the older items, like for us, selling to a lot of the retailers, they kinda go of date of received versus date of invoice. But on the older items, we believe that that’s where we could get the biggest benefit. And especially if we can catch him earlier. So this system can properly predict when one invoice will be paid. We kind of design a strategy to say hey if it’s only going to be five days ignore it, but if it’s going to be 15 days column earlier based on the dollar amount kind of bubble all that up in the worklist would be an advantage. And so what we’ve done now is we actually have it live.

We have incorporated it in you know in real-time but it’s still kind of early. The system’s learning so you know we don’t have the results yet, but we’re hoping that throughout the year we can actually move forward.

Kush Kumar:

Okay. So you recently went live with that feature right, payment day?

Marinko Marijolovic:

Yeah, we did it late last year. And so it’s just been a couple of months I guess at this point.

Kush Kumar:

Okay. So let him go just to follow question on that. What is the biggest difference that you see between the proactive collections versus the reactive collection that was happening earlier?

Marinko Marijolovic:

Well, so what we’re trying to do is one of the things that we’ve talked about, and you know HighRadius has talked about this and I mentioned as well as to be able to predict when an order will be paid. It’s really the order, not even the invoice. So when the orders entered, we can actually see if they’ll be paid on time or not. So it’s actually not looking at the account it’s just looking at that specific item and the whole goal is to go from being reactive, so if you think about it you know in the old days if you’re using spreadsheets or even index cards or whatever way back when most of the time you were reactive either the account was past due or you had an order that came in and that forced you to take action in account.

Well, the whole point is how can you kind of flip the tables and be more proactive. So all of these system solutions are allowing us to do that more so. And like I said if we can alter our, you know our strategies and enhance them because one of the things that we were having problems with before is our sales team would be more on the side of the customer and they’d be defending the customer kind of question you know when was the last contact anybody tells the customer they want to hold all this. And if you recall, if you mailed your dunning notice for example customer may have gotten the main ad and went through the email or through the U.S. Postal Service. And then you kind of moved to email but now that you have more technology you kind of paying all these customers and if nothing else you can say hey they’ve been notified that there’s a possibility an order will be held. And we’ve noticed a big drop off in our salespeople not being as defensive or not causing us to be as defensive I guess. And so we think the predictive payment is like another step in that direction because if we can you know tell the customer and our own internal people with a fairly good amount of accuracy that this customer will pay late for this product you know for this order we can take action with them we can ask the customer to negotiate or ask the salesperson negotiating with the customer to say, “Hey we need a commitment that we’re actually going to get paid.” So even if you get paid on that order the next time around and it comes around and you want to hold you you’re saying, “Hey we can’t release this large of an order.”

Everybody within your organization is going to be supporting you because the customer is already reneging on the previous promise. So everything is about you know being proactive versus reactive.

Kush Kumar:

Makes sense. I think the biggest advantage I have seen with all these customers is and the team that is not involved right now but the Treasury team I’ve seen they have they’ve benefited a lot because based on the predicted payment date they can do a lot of cash forecasting right?

Marinko Marijolovic:

Right, and it is one of the things I talked about is. So if you incorporate your predictive payment. You’re gonna be fairly accurate as to for your cash flow and then you know you guys also have the skip tracing tool as we talked before, if you incorporate that, you should be able to get like a 90-some percent accuracy rate and especially if you’re doing like cash forecasting you know on a daily basis versus a weekly or monthly basis it would be a huge huge improvement.

Kush Kumar:

So Scott likewise. You also recently implemented the in-app calling feature as we discussed. Would you like to walk us through the UI if you know what everyone, like you would like to just kind of tell everyone how you are collectors are using this apart from just you know logging the call notes and transcribed? Do you think that this is something that is very beneficial for you or your team? About the DSO, like what is that what are the softer benefits that you’re getting with this feature.

Scott Phillips:

Like I mentioned before, before HighRadius in the collection module we were using spreadsheets and folders and I think Sashi had a good slide where you showed the lady on top of a stack of folders. That was our credit department before HighRadius. So just simply having that tool and now having visibility and putting metrics in place that the credit managers are held accountable to. We have far exceeded our line on the collection module reducing our DSO by two days and we’re looking for ways to take that down even further.

So that has been substantial, I won’t tell you the dollars but there has been a substantial impact on the company. And I also have treasury understated services as well so I can confirm that. That the tool has also been enabled our Treasury Department to improve their cash forecasting as well.

So we’ve just seen a number of benefits just from the collection model.

Kush Kumar:

So one question for both of you how do you create a culture in the company to take pilot initiatives and partner with organizations such as us and take pilot projects and kind of internally sell this value proposition that we should be doing pilots and POCs to see the benefits or to see the long term results.

How do you create a culture like that in the company?

Marinko Marijolovic:

Well like for us our CFO is kind of big on technology. He pushes for automation. So I’ve already said and I don’t know how many RPA and blockchain sessions are demos and things like that.

That doesn’t mean obviously we’re going to jump and implement everything but we definitely don’t want to be left behind. And in my case, so I’ve got a retail side of the business. And then an industrial side so we can only commit so much time to either side but we have to obviously take care of both. And you know we do need technology to make sure that we’re balancing that and it’s kind of funny. So like on the retail side we need more help with deductions and automating deductions, where on the national side we need more assistance with you know collections and predicted payments and things of that nature. So without technology, we would have to keep adding people.

And we’ve actually reduced the number of people that we have. And that was not our goal when we started out as we were just trying not to have to add people every time we grew a little bit. And that’s that has been the case.

Kush Kumar:

And Scott, for you?

Scott Phillips:

Yeah. I mean some works with Marinko that is. Implementing the collection piece has made the credit managers’ jobs more efficient, has created more time that they’ve been able to reallocate to value add. So instead of reducing headcount, we’ve chosen to help them re-prioritize their day and re-prioritize what they’re focusing on.

So instead of calling an account efforts passed due like you alluded to earlier we’re now looking at those accounts where we are more focused on looking at credit line utilization and looking at accounts that are 85 plus percent utilized on our credit line and starting the conversation with the commercial team of what’s on the horizon and what are the future sales look like and the timing of that to determine when we can start reviewing those customers earlier than we normally would prepare them and have that credit line available when the sales there so there’s not an impact to the process and timeliness of it being us being able to deliver our product.

Marinko Marijolovic:

And just to add to that you know what Scott’s talking about, that is the key like you want to free up your people to be a collector, an analyst to manage their process and with some of these tools that are really what you’re allowing them to do because then they can work on those projects that Scott mentioned. And they’re not as stressed and they’re not kind of walking outgoing “Whew. You know I survived” versus “OK. You know it’s 10:00 in the morning I’m gonna take a little break. I’m going to work on this account which I know it won’t take me a couple of hours” and that’s really what you’re trying to do with a lot of this technology.”

Kush Kumar:

Any questions will open it up for Q&A, any questions for Scott and Marinko about what they piloted.

AUDIENCE:

Did you mention that you have a skip tracing technology?

Kush Kumar:

Skip invoice tracing. Yes, we have it there.

AUDIENCE:

It was. What is it?

Kush Kumar:

Skip invoice tracing.

AUDIENCE:

And, how does that work?

Kush Kumar:

So the skip invoice tracing, obviously we have certain customers, your end customers who would post the invoice details on the portals so the skip invoice tracing technology is a robot will actually you go into that portal pull all that invoice statuses and bring back to the collections module so that when a collection specialist is planning to make a call or send the Dunning correspondence they will know that this is the status of the invoice but whether that has been paid or not paid or will be paid in future. So that’s the skip, so no invoices are skipped. Did I answer a question? Okay.

Marinko Marijolovic:

So if you’ve got like a lot of large accounts. So if you’re selling like say to Walmart or somebody you know where you’ve got thousands of accounts right now. Somebody’s got to go out there. On their website upload all the items they have set up for payment you’ve got to dump your data out of your system and then basically do a v-lookup there is really very little value for that just to bubble out like the 10 or 12 invoices that may be missed out of the thousands.

So if you can incorporate you know the due date. Into the open items list and then based on your strategy that you would incorporate it would come up on your worklist just like any other account to say “Hey here’s the, you know five or six items you need to chase”. But the other key thing with the skip tracing is that if you know early on, so obviously Wal-Mart’s not gonna go out of business like you can get paid but it if you know that that item has not been set up or payment like five-six days out but everything else has there is some kind of issue even though your EDI folks may have said “Nope everything looks good on our end”, they can address it right there. I’ll be a lot easier a lot quicker and the customer is going to be more willing to speak to you about it as well. So it’s it would really be beneficial in those like really super large accounts where you may end up having a few hundred thousand dollars floating out there for no reason.

AUDIENCE:

So my question is more for Scott. The last thing you guys were talking about was reallocating your department resources to do other projects because now they’re spending less time calling customers and e-mailing customers. So what type of skill set shift is that are you looking at a different skill set for the more analytical side or is it the same skill set that people used to have for the collections?

Scott Phillips:

So for our credit department each credit manager manages the portfolio and by managing the portfolio I mean they do the credit analysis they do the collections they manage the payments through the EIPP portal. So it’s more reallocating their time of where they’re spending that. So we don’t have a dedicated collector. Does that answer your question?

AUDIENCE:

Then I guess it was it specifically for you I thought that. You had a collections team. No?

Scott Phillips:

No, there is no dedicated collections team. So what I was referring to is just reallocating that time that they’re spending on collections to managing other parts of their portfolio versus the collection piece.

AUDIENCE:

So then do we have an idea of what the transformation will look like for a collections team implementing this piece of it. The today team and then the tomorrow team of how they would be spending their time instead of calling customers and e-mailing customers for collections. Now the tool helps with that so they have maybe five hours back in their day. How they’re spending their time. What skillset would look like?

Scott Phillips:

So for me, if it was a dedicated collections team and we did our own ROI analysis on the collection piece and then we met and exceeded that ROI. If I just had a collections team I would have looked at either repurposing that headcount to other activities or reducing headcount.

Kush Kumar:

We have time for one last question before we end the session. Any questions?

AUDIENCE:

Sorry about that. Okay. Lindsey from Mattel, I’ll go back to the skip tracing invoices with we have Wal-Mart, Amazon, Target and we have a lot of problems that might sometimes be EDI related. Maybe it just didn’t go through this flow through the system correctly from ours to theirs and it results in past due to invoices how does HighRadius notify you if their system just simply doesn’t reflect that invoice number?

Kush Kumar:

Correct. So we have the notification mechanism. So if there is an invoice number in your open AR and we have a created date right we have an invoice create date and we have a past due date. Right. So we go to the portal and if you don’t find the status of the invoice we can send a notification to the admin collection manager as well as the customer that we have not seen the status of this invoice. When are you planning to pay this invoice? And you, your team will also be notified that these are the list of invoices that we did not find any information on the portal. But these are the invoices that you need to kind of ask the customer.

Marinko Marijolovic:

I believe there is also a demo out there, don’t you guys have a-

Kush Kumar:

Yes. So we have a demo booth, you can actually see a demo of the same in one of the demo booths.

Marinko Marijolovic:

But I’m glad, by the way, I’m glad that there are other people that have problems with these large retailers besides me because everybody else says “Oh, there’s really no problem if your EDI goes through, it’s like no kidding, really?” You know, but when it doesn’t go through it’s a problem.

Kush Kumar:

So, Scott and Marinko before the session ends, so, I just want to kind of ask you one question, with all this extra money that is coming in with this technology, what are your next plans? Are you buying the other modules? No, I’m just kidding.

Marinko Marijolovic:

Big vacation.

Kush Kumar:

Thank you so much, have a nice evening. Thank you.