Success Story: Scaling up Receivables Forecasting at Danone North America with Artificial Intelligence

Struggling with spreadsheet-driven forecasting, and managing multiple entities and banks, Danone North America could only forecast receivables every six months. Moreover, the existing process failed to capture deductions and customer payment behavior, so they had to rely on assumptions. Learn how Danone NA has set out to perfect its receivables forecasting process where it can now predict with 96% accuracy across business entities.
Jacob Whetstone

Jacob Whetstone

Director, Invoice to Cash, Danone North America
Download Slides

Session Summary:

Takeaway 1
Top day-to-day forecasting challenges arising due to error-prone and manual spreadsheet-driven processes

Key Points

  • Time-consuming and inefficient manual process
  • No insights on the invoices that haven’t been generated
  • Lower visibility into AR processes across multiple horizons
  • Difficulty in analyzing the customer behavior changes
Takeaway 2
Top benefits of implementing AI-Based Cash Forecasting

Key Points

  • Improved accuracy of short-term and long-term forecasts
  • Enabled analyst to perform variance analysis and high-value tasks
  • Improved customized forecast models to filter early payments and deductions
Takeaway 3
Top benefits that Danone achieved through leveraging HighRadius end-to-end integrated solutions

Key Points

  • Increased accuracy up to 96% in the monthly forecasts
  • 30% reduction in the time spent on trivial manual tasks
  • End-to-end automation across regions added efficiency to forecasting
  • Variance analysis helped identify payments and improved collections
Show More

Jacob Whetstone 0:06
Alright, guys hear me. Okay. Yeah, thanks. Thanks for coming to this session. Being a part of this, I appreciate it. So that’s the thing. So we won the award for Treasury Excellent. I’m not a treasurer. I’m in the accounts receivable aside, I don’t know a lot about Treasury I’ve I’ve worked with them a lot, for sure. They’re a great department, but I’m not a treasurer. So this is our experience with Cash Forecasting from an accounts receivable perspective. And so maybe a little bit different take a lot of the slides are actually similar to some of the size that we just saw in the other presentation, but we’ll go through it anyway.

Jacob Whetstone 0:39
So just start off a little bit about so here’s the agenda. We’ll go through real quick, but a little bit about the unknown. And so just an overview of who we are. So, known to a lot of people in the United States, at least when I say denial, and they say what Who do you work for, we don’t know anything about to know. And so put the American spin on it as Danone, Danone yogurt. So so it’s the yogurt product. But did known as our parent company didn’t own is actually just a brand of ours, we have a couple of different brands. So all the symbols there, the boy looking at the North Star, kind of like our guiding star there. But some of the other brands that you may be aware of you may know a little bit more activity a yogurt is our oils, yogurt, we also are big in the plant base space or silk milk, we own silk milk now we produce that as well. If you guys like ice creams, try so delicious ice cream, ice cream, I think no matter what’s not healthy, but so delicious, is a little bit more healthy, so you can feel good about that piece of A, we also have Evian water. So some of those logos are just some of the different businesses that we support. I just support the United States and Canada is really where I cover but we are a worldwide company. And different countries have different similar people supporting them as well. So so this is just very high level and very simplistic, we wanted to give an idea of of how we we work with the Treasury Department as accounts receivable piece, and how it works with our Cash Forecasting. So as it makes sense, I’m sure to all you guys here that are there. So we’re just the inputs there. So the way that we wanted to show this is I needed provide the cash in forecasting, the accounts payable would provide the cash out forecasting, it would go to a team that we call our cash management team, they would put that together, and then present something up to our global Treasury team and to our corporate finance, that would actually report out to the shareholders about where our free cash flow landing is. But really, we’re just that a spot in that corner, which I say just that spot on the corner, I like to think we are a big deal. We always accounts receivable, say no one in the company gets paid unless we collect the money. So so we are a big deal. But I’m trying to forecast the cash that’s coming in gets very tricky. So really, you know, what we set out to do is we needed to be able to provide a good cash in forecast. So the company can make decisions of what it wants to do. If we have enough cash coming in, that they need us to get more or they need a pace amounts that we can land where we want to be make sure were as accurate as possible, and then drill down to the lowest level so we can get monthly, weekly and daily forecast out of it. So our initial approach, and again, this slide I know is similar to the one we saw from the other talk, but it’s really the idea. So we’re we were already a HighRadius customer, we have a HighRadius, we use them for our Deduction Management System, the Collections, and the Cash Application piece, so a lot of our information was already flowing back and forth to HighRadius. So HighRadius was one of our biggest inputs. But we were extracting that out of HighRadius, putting in Excel spreadsheets with our banking data and our ERP system, our SAP, and then trying to crunch all the numbers to come up with a Cash Forecasting all within Excel. And I thought we were pretty good in Excel, we did some pretty cool things in Excel or so I thought, but we were still not very accurate with it and the more variables that we thought of made it more challenging for us. So we would want to get it at the invoice level we are trying to forecast at the customer level, the invoice level really tried to understand as a CPG company, we have a lot of deductions. So we knew if we had a $1,000 invoice, we’re actually going to get $900 in cash, we wanted to capture that $900, not the $1,000. And as we added more and more variables to it, it just got a lot more complicated. So we’d have a credit analyst who when the time would come to forecast the cash would spend eight hours of his day just manipulating these spreadsheets. And then we would send it off to the Treasurer that our cash management team, we’d hit the send button. And we would hold our breath and we were dread because we knew we were going to get questions in return. And the only way we could answer those questions was to re-forecast and spend that time again. And so it got to be very complicated for us. And we didn’t have a lot of intelligence that we could actually talk about it. So it was just a very challenging position that we are in to really try to get and as our business grows, and we brought on other companies, we needed this more and more. So the other thing that I wanted to talk about here and maybe this slide doesn’t highlight it enough, but like I mentioned the different factors that we brought in, but for us, as we’re yogurt A dairy-based product, our terms are very short. So our product is very perishable. And so we have short terms. And so we would sell our product and the invoices that would generate in one month would be paid before the end of the month, or there were times when we would, we would need to invoice and it would become due before the end, the month would come when we’re trying to invoice before that actually invoiced. And so we needed a way to not only get visibility into our open AR but also to get visibility into future invoices that would generate and then payout before the end of the month. And we weren’t able to do that with our forecasting model that we had in Excel. So again, that’s kind of that bottom line there. But for me, that was a really important piece because we needed to add that variable that we didn’t have before and those invoices yet to be generated that we wanted to capture within our forecasts in that short term.

Jacob Whetstone 5:48
So again, a lot of challenges with our Excel spreadsheet, even though I thought we were pretty good with it, we still had a lot of errors with a we couldn’t get into any root cause we really wanted to use our Cash Forecasting, to try to get to be more proactive. So we could say, Okay, we want we know that this invoice is forecasted to be paid this day, it didn’t payout, why not and be able to send our collection team out to go and try to understand why it didn’t get paid. So we really wanted to get to those root cause analysis and understand where we missed our forecast and what we could do about it. And then being able to also take into more variables, like I talked about the deduction piece, but also the early payments fee. So as we were looking historically, we would see where customers were paying early either because they were paying early for their cash flow purposes, where we were requesting early payments, and that was screwing up our forecast as well. So being able to take into different customer behavior changes also. So then we worked with HighRadius, and we were able to implement their AI forecasting module. And so this looks very similar to the other one I showed before. But the biggest piece is that little cog there, which represents Artificial Intelligence, I guess. But that cog piece is really where we took out that manual process and had the system do it for us. And we were able to input a lot more different variables into our forecasts. So we could take in different things that we weren’t thinking about or that we weren’t doing. And the system was actually doing it. So it’s pulling some of the same data as well as extra data. But instead of a person doing it, it was all about the machine doing it for us. And now our credit analyst was sitting on the other end of the little chart, and instead of him spending all of his time, putting the forecast together, he was then spending his time analyzing that forecast. But like I said, trying to be more proactive and being able to understand which customers were key that we worked with to ensure that the collections were coming in when it was projected or forecasted that they would come in as well. So we really wanted to leverage this tool to help our collection process and help us be better in collecting and being able to improve our DSO. Oops. So yeah, some of the things I just talked about, there’s, I should have put this slide as well. So I apologize. But these are just some of those highlights that I just talked about as being able to really dive in a little bit more and spend more time analyzing the forecast rather than calculating the forecast out. Alright, and this, yeah, so sorry. So just sort of how our implementation went with this. So again, we were already customers with HighRadius. So a lot of the integrations were built, and they were already there, we were already sending a lot of our data. So for us, we decided we were going to go live. I think it was like a February timeframe. And that we were up and running very quickly because of all those integrations that were built. And then we were able to do the UAT as a parallel testing. So for us this semester and the year-end were a big deal where we really needed to have the cash forecasted. So we were going we were shooting for the June month-end is where we wanted to have this ready to go. So they did all their deep dive all their analysis and everything so we were ready. So the start of it was actually like the middle of May, they were going to start their forecasting and we are going to start our forecasting doing the old way that we had. And so our credit analyst, he thought it was great. So he thought he was going to be HighRadius so we put a challenge up he told me he asked me that I would have to buy him a big price he actually wanted a car I told him there’s no way I’m gonna buy him a car but he said if he beats HighRadius forecast, I owed him something big so so we put it out there. So he tried really hard to try to beat HighRadius forecasts and he didn’t even come close. So we were able to hire radius using there. And again, it was because of all the different variables that we were able to put in. I didn’t owe him a card and or anything like that. But he was great to work with and helped us out with that. But it really very easily just with a push a button, we’re able to get a much more accurate forecast and we were able to get without him so and then here’s some of the actual results that we had. So want to see where it really was. So prior I would say with our we did a pretty good job before. Without HighRadius when we were doing it with the Excel spreadsheets. We were getting around, I would say 83 to 85% accuracy where our cash was coming in. Again, a lot of work out there just to get that to that piece of it. But when we were implementing HighRadius, we were able to get both at a weekly accuracy at 91%. And then a monthly accuracy at 94%. And then as we included some other variables, so the variable that we’re going to show here is just with early payments. So as, as they, as HighRadius was trying to understand historically, what had happened. So customers starting to pay early, they were able to build that in their model and adjust the AI model to be able to input behavior into that piece of it. And we improve that to 94%, or to 96%. So again, very high percentages for us. And this is looking at what was forecasted versus the actual cash that came in. So it was a great win for us that, again, saved a lot of time, but something that we could feel very confident with that we could pass on and be able to have a lot of confidence in what our accuracy was with the forecast overall.

Jacob Whetstone 10:51
So again, just some of those numbers. So up to 96%. Accurate, I put here that we save 30%. In Manual time, it probably was a lot more than that, especially in that June and December timeframe, where we really were trying to focus on that cash forecast and get as accurate as we could a lot of time there, we were able to get daily visibility. So prior because it was such a manual process, we didn’t forecast daily, now we just push a button and click get a daily forecast. And being able to, like I said, this variance analysis. From my standpoint, this was one of the biggest benefits that we had from it is being able to leverage our forecast to help us with collections and be able to understand what’s going on with that piece of it and work with our collections team to help them collect better. So you know, what’s next for the unknown. So again, we’re going to continue to work to try to get it as accurate as we can and try to scale the product up as much as we can. Like I mentioned earlier, one of the big things for us with our short terms was not only understanding the forecast of when those invoices would be paid that were already opened but understand the invoice sales that were yet to happen. And so being able to understand and break down the forecast, when HighRadius gives us a number, whether it’s six months out three months out one month out, I can know okay, this fort, this part of the forecast is based off, the invoices that are open this part of the forecast is based off what those sales are. And I can make changes if I need to, I can just understand that piece of really trying to dive down a little bit deeper to understand what’s driving that cash forecast and be able to make changes and action plans around that. And then as we continue to grow, so we built the model with our denote US business this year, we have Nutricia up there, which is our medical business that they we just brought them on a couple of months ago, and then being able to leverage them as part of this Cash Forecasting also, and being able to easily bring different entities in. So it’s an all-encompassing thing. We don’t ever want to go back to our manual way of doing things. We don’t ever want to resurrect those spreadsheets. So any company we bring on, we want to make sure that we run this through as well. And then like I said, again, I don’t mean to hammer this, but to help our collection activities. And again, from an AR perspective, that was a big piece of that. So that was really fast. I know I probably talked a little bit fast. So I apologize. But that’s yeah, I mean, so that’s our story. That’s our story with HighRadius and the Cash Forecasting aspect of it, and really where we came from. And again, it’s something that we’re very proud we were able to do and something that’s helped us out a lot that now we just really push a button and we get a forecast. We’ve also had our treasury team. Now for their daily forecasts, we gave them logins as well, they just log into the system. And every day they get the forecast from their for that cash in piece of it. So it’s been a great benefit for us. But anyway, any questions from anybody? Yeah?

Attendee 13:42
Your payables. Oh, I’m sorry. Any plans to put your payables into that model as well?

Jacob Whetstone 13:46
Yeah, so we went just with the receivables. First, we are looking at our payables team now they saw the success that we’re having. So they’re still stuck in the old way. And so when we get together and we work with the Treasury team, and we just push a button and get a number and they’re still crunching spreadsheets and they’re making originalist stuff they’re really interested in analysis so yes, that’s the plan is to try to get them into it also.

Attendee 14:10
How many entities domestically? Are you guys forecasting?

Jacob Whetstone 14:15
So about? So we have one some that are split and kind of two But about six entities between the US and Canada? Okay, we’re working with.

Attendee 14:24
And in terms of so you mentioned your payment terms are kind of short by the nature of the product beyond the point where invoice data is valuable so say I don’t know two months are using different methodology beyond that to forecast so like AI machine learning and what’s like that switchover point were no longer invoice data something else?

Jacob Whetstone 14:41
So yeah, our terms are short enough that within the same month that we can have invoices there so yeah, after about I would probably say like six weeks or so then we can’t really rely on any of the open invoices. So then it’s all yes, it’s using the historical data, the AI learning trying to understand by and it’s by customer level. So it’s also trying to understand that customer level different growth rates with the customer other things like that to try to determine what that invoice sells is going to be out there.

Attendee 15:04

Jacob Whetstone 15:09
Anything else? Yeah?

Attendee 15:20
Do you have a sales system? That’s another integration point? Or how do you

Jacob Whetstone 15:22
Yes? Don’t with HighRadius, it’s just using the AR data. So it’s just looking historically at AR AR balances at that customer level. And figuring out, you know, eventually all those turns all those sales turn into AR data, but that’s all we’re sharing. We’re not sharing anything else outside of just our AR data.

Attendee 15:39
Got it? So it’s not like a Salesforce feed.

Jacob Whetstone 15:41
It’s not there. No, not with us. No.

Attendee 15:42
And then one other question I had real quick. So one question that comes up from a lot of our potential prospects and customers is, how do did you think about how COVID may have affected the way your customers were paying you when we first looked at the historical data, like, you know, a big disruption that could be sitting in that data set? And how did you guys overcome that or factor that in to get to, if it was a factor at all to get to the level of accuracy that you got?

Jacob Whetstone 16:09
Yeah, so we, we did take into account like some of those things. But also each time we do this, there are other factors that we know that the system doesn’t know. So even if, like one of our major customers in December had an EDI error, where we weren’t getting them any invoices at all. So the system was projecting that we’re supposed to get paid. And we knew we wouldn’t. So there’s a place in the tool that we can go in and make those manual adjustments or Yeah, or we would work with HighRadius to say, Hey, this is going on, or we’ve had customers behavior change behavior as well. So as we’ve done better on our collection job, we’ve changed someone from a check to an EFT. And we know we’re gonna gain, you know, five to 10 days or whatever. So then we work to try to build that in the model. But it’s also learning as we go. So some of that is automatically captured. But some of that is adjustments we make based off what we know.

Attendee 16:55
Got it, thanks.

Jacob Whetstone 17:17
So probably more going over the different variables. So I don’t know, like I don’t we were one of the first going with his CashApp. I don’t know, you know, where we were in there. But there are some variables that we didn’t think about. And I don’t think the HighRadius thought about, so one of them was very easy for us was, so the year-end was a big deal for us. And we noticed that the forecast wasn’t taking into account the holidays. And so like Christmas, or New Year’s, wherever we were expecting, the forecast was putting cash that was going to come in on those holidays, and some of those holidays, would actually push the cash to come into the next week, where we are trying to forecast that week and get that information, we were showing way too much cash coming in, because it was saying it would come in when it actually wouldn’t. So just some of those different variables. But then also this whole idea of being able to understand the invoice sells yet to come versus that open-air, we should have spent more time understanding that piece of it and really breaking it down so that we can know. Like the question, they’re being able to understand the different variables and make adjustments, being able to split that forecast out so we can understand that piece of it. But really, yeah, some of it is we are growing with HighRadius with this, not knowing what we didn’t know, I think we almost had to experience it and realize, okay, why were we off when we were off, and then being able to understand what variables drove that and then being able to work to make sure that that was built into the model. That makes sense. But yeah. So all right. Any other questions or anything? Alright, like I said, I know I went fast. So I apologize but give you some time back, okay.

Attendee 18:50
Without understanding your AP, I mean, doesn’t that kind of mean, how do you understand what your working capital is if your AP isn’t actually up to par with what your AR is?

Jacob Whetstone 18:58
So that is somewhere where we need to get to is to make sure that our AP is there. So right now, they have their forecasting method is just very manual. So they get that piece from them. It’s not as accurate, especially from a daily level as what we can do. But it was something that, you know, because we already had the integrations and everything with HighRadius. From an AR standpoint, it was very easy for us to get up and going from that AR side. And so we could easily trigger that where the AP is going to be a little bit more work to do. And hopefully, we can get them on that piece of it. But it is still needed.

Attendee 19:27
What is the daily reason like the reasoning for having a daily was I mean, at some point in the past, there was just like, mistrust with what was happening.

Jacob Whetstone 19:36
And then so we never, to be honest, we never intended to have a daily one, it became so easy to get the daily one, but now we do it. And now that’s something that our treasury team actually wants. We were just trying to get it for the month-end is really what we were driving for. But yeah.

Attendee 19:52
I’m sorry, no, you’re using HighRadius before for some of the AR modules. And I’m just curious, you mentioned you know, now With the forecasting can better support the collections and the targets? And just wondering kind of what was your functionality pre that versus post? Like, what have you gained with the forecasting to support collections teams on target setting? Driving cash in the door?

Jacob Whetstone 20:15
Yeah, so it’s really just so with the forecast, it was more just understanding what because the forecast was built to be able to understand customer behaviors, like their average days to pay and so we’re just, there’s a report in there that we can, when we get the forecast, we can dive a little bit deeper and see that invoice level which invoices by customers are supposed to be paid on which days, and then we just compare that with our open AR and see which ones actually have not been paid that should have been paid. And now we can pinpoint our collectors to do that. And we can act a little bit more proactively to the other way around that we can see. Okay, by the end of the month, we’re supposed to get this much from this customer, we really need to get this amount of money if we want to hit our cash forecast, our free cash flow targets. So can the collector ensure that all the invoices are on the portal can they do things prior to ensure that that cash comes in and there’s nothing stopping it? So prior, it was more just always chasing you know, sending the collection letters after it came due and now we’re trying to move it and say this should be paid out. If it’s that important. Let’s try to make sure that we know it’s there or act a lot quicker. So instead of waiting for so many days before it’s past due before we send that notice if it’s not there the day, it’s forecasted to be there. We want to contact them that day. So it’s just to get a little bit quicker with it. Give you some time back, guys, go get dessert.

Enjoyed the Session?
Here are some other sessions you might be interested in

Uplifting Treasury’s Mandate to Strategic Partner of the CFO’s Office
Greg Carlson

Greg Carlson

Managing Director, Riveron
Radiance speaker's company logo
Success Story - How Employer Flexible Leverages Cash Forecasting Insights for Strategic Mergers and Acquisitions
Raymond Dobrosky

Raymond Dobrosky

Controller, Employer Flexible
Radiance speaker's company logo
Kelly Grieshop

Kelly Grieshop

Treasurer and Small Business Owner, Employer Flexible
Radiance speaker's company logo

Experience the Power of AI-Enabled Receivables
& Treasury Applications

Schedule a free demo