Skip to content

ON - Forecast Accuracy Deep Dive - 2025-08-07

Metadata

  • Date: 2025-08-07
  • Company: ON
  • External Participants: Yulia Ershova (Treasury Corps), Rodrigo Cabrera (Treasury), Amanda Mitt (Treasury Analyst)
  • Palm Participants: Christian Sobkowski, Art Koci
  • Type: Customer Call
  • Domain Areas: Cash Forecasting, Forecast Performance, Variance Analysis
  • Recording: https://tldv.io/app/meetings/68948763137b1600138478f6/

Summary

Context

Deep dive session with ON Treasury team to discuss forecast accuracy requirements, granularity needs, and how Palm's ML forecasting should be tuned to meet their goals. Art Koci (ML engineer) joined to understand ON's specific data patterns and accuracy targets. Session focused on understanding what accuracy metrics matter most, at what granularity, and how forecast vs actuals reconciliation should work.

Key Discussion Points

  • ON's internal target: 95% accuracy goal for Treasury (balance level)
  • Accuracy granularity debate: global balance vs per-category accuracy
  • Weekly accuracy more important than daily for most use cases (payment runs on fixed days)
  • Forecast horizon preferences: weekly for 13 weeks, monthly beyond 3 months, daily for next week
  • Cash pool view needed: aggregate forecast across pool header + children accounts
  • Intercompany transactions should NOT be forecasted (nets to zero globally)
  • Account 2967 challenges: repurposed account, behavior change, regional team management
  • Forecast update frequency: currently twice daily, once weekly model selection

Pain Points

  • Intercompany miscategorization - US team uses different descriptions, causes categorization issues
  • Behavior changes - Accounts get repurposed, historical data becomes misleading
  • Regional team visibility - US/Asia teams manage locally, Treasury doesn't see day-to-day details
  • Kyriba limitations - "Now pushed Kyriba but now we have there forecast" - switching to Palm

Feature Requests & Needs

  • Cash pool aggregate view with balance forecast across all pool accounts
  • Click-through from forecast bar to transaction details (drill-down)
  • Flexible view pivots - user-defined groupings (entity, currency, custom pools)
  • Snapshot capability for quarterly reporting (to show forecast improvement over time)
  • Open items/AP-AR integration for better operational forecasting

Jobs & Desired Outcomes

Job: Achieve reliable balance forecasts to support liquidity allocation decisions

Desired Outcomes: - Minimize the variance between forecasted and actual balances at quarter end - Increase confidence when reporting forecast figures to management - Reduce the time to identify which categories are driving forecast variance

Job: Allocate excess cash to investments based on forecasted minimum balances

Desired Outcomes: - Minimize the risk of overdrafts due to inaccurate forecasts - Increase the accuracy of available-for-investment calculations - Reduce manual calculations when making investment decisions

Job: Reconcile forecast variances to improve model accuracy over time

Desired Outcomes: - Minimize the effort to identify categorization issues causing variance - Reduce the time to re-categorize transactions and see updated forecasts

Domain Insights

  • Accuracy measurement: ON measures at balance level (not cash flow level), 95% target
  • Timing matters: "It doesn't matter if we forecast 25 million this week and we had 50 million... We need to forecast when it will happen" - Rodrigo
  • Payment run patterns: US/Canada on Thursdays, Friday for some, Monday for others - makes weekly→daily conversion predictable
  • Forecast horizon tiers: Daily (1 week), Weekly (13 weeks), Monthly (beyond 3 months)
  • Intercompany philosophy: Don't forecast IC - "ultimately it doesn't affect your global balances"
  • Two use cases for variance analysis: (1) Quarter-end reporting to management, (2) Finding flaws to prevent overdrafts
  • Forecast freshness: Want up-to-date forecast when logging in to make investment decisions
  • Snapshots: Needed for reporting/proving improvement, not for day-to-day business

Action Items

  • [ ] Art/Christian: Address account 2967 accuracy issues (new account patterns)
  • [ ] Palm: Build cash pool aggregate forecast view
  • [ ] Palm: Add click-through from forecast bars to transaction details
  • [ ] Palm: Open items integration going live "next week"
  • [ ] Rodrigo meeting Christian at PLC Tech Days Zurich end of August

Notable Quotes

"The more is okay. Less is bad." - Yulia (on forecast accuracy - overestimating cash is acceptable, underestimating is risky)

"It doesn't matter if we forecast 25 million this week and we had 50 million... We need to be into forecast when it will happen, the expense, not a week after or a week before." - Rodrigo

"I think that's very old school way of doing it... that you keep snapshots" - Amanda (on moving away from snapshot-based reporting)

"The current view already very good. Have a very good usage for our day-to-day." - Yulia

Full Transcript

Date: 07/08/2025, 13:00

00:00 Christian Sobkowski: breeze again.

00:00 Yulia Ershova: It's crazy.

00:04 Christian Sobkowski: How are you both doing? Amanda said You had some time off this week, they get the right.

00:12 Yulia Ershova: Rodrigo had.

00:13 Christian Sobkowski: Pretty good.

00:13 Rodrigo Cabrera: Yeah, few days. Yeah.

00:14 Christian Sobkowski: Okay. Nice.

00:16 Rodrigo Cabrera: Trying to enjoy the end of the summer.

00:20 Christian Sobkowski: Very cool. Did you did you do something fun?

00:24 Rodrigo Cabrera: Eh. It depends what you consider fun? No, but yeah, we went to the

00:28 Christian Sobkowski: I don't know what? What do you consider fun?

00:32 Rodrigo Cabrera: Now, we went to the A to the French side here in Switzerland for a couple of days on the weekend. It was nice, it was nice.

00:38 Christian Sobkowski: In Zurich.

00:48 Rodrigo Cabrera: A yes. Limited end of August, right? 20 something, yeah.

00:51 Christian Sobkowski: And of August. Yeah, yeah. Nice.

00:55 Yulia Ershova: If you're coming to Tori.

00:58 Christian Sobkowski: yeah, we are, we're gonna be exhibiting at the PLC Tech days in Zurich and in

01:05 Yulia Ershova: Nice.

01:07 Christian Sobkowski: Geneva, I think Rodrigo is you actually coming to the event, right?

01:14 Rodrigo Cabrera: Yes. Yeah.

01:15 Christian Sobkowski: A fantastic. Very cool looking forward to

01:19 Rodrigo Cabrera: It looks just looks interesting. We'll see. Yeah, I was talking with Gorget, the schedule something afterwards. The before the grab something to eat. Will we'll see.

01:32 Christian Sobkowski: We should we should 100%. Hi Amanda.

01:36 Amanda Mitt: Hi. Hello Christian. Hi everyone. Hi

01:40 Christian Sobkowski: Very cool. Hey

01:40 Amanda Mitt: Sorry.

01:43 Christian Sobkowski: you both met but nobody else did. Right. Are you gonna say like to see words

01:54 Rodrigo Cabrera: Yeah.

01:56 Christian Sobkowski: about yourself?

01:57 Art Koci: Hello. I'm Art. I'm a machine learning engineer, AI engineer. I'm not quite sure nowadays, but one of those. Yeah, and I have been working recently with forecasting and the data And we are specially focusing on your your particular data because there's some cool behaviors. We want to capture better and forecast better.

02:21 Art Koci: Um, yeah. So I'm just joining this, this chat to see how you guys are using palm and what would help for you to for well for both of us to make the predictions better. And yeah, the data generally more useful. Yeah, that's that's basically it.

02:45 Christian Sobkowski: Awesome. Maybe, where, where you've been before? I think that's always nice

02:46 Yulia Ershova: Nice to meet you.

02:48 Amanda Mitt: Nice to meet you.

02:51 Christian Sobkowski: context.

02:52 Art Koci: Yeah, I'm quite a big fan of your company as well. So

03:52 Yulia Ershova: Nice nice it impressive. Good for palm, right?

03:53 Amanda Mitt: Right.

03:57 Art Koci: Oh, I hope so.

03:57 Christian Sobkowski: when I said, Good for us, Whenever said good to us.

04:01 Amanda Mitt: What is opening a lot of machine learning AI positions as well? So,

04:08 Art Koci: Nice trying to be post in front of my CEO.

04:09 Rodrigo Cabrera: Don't tell that the Christian.

04:13 Amanda Mitt: Just, I think it's too big already.

04:17 Christian Sobkowski: Amanda no coaching, please.

04:17 Art Koci: If I get a if I get a nice shoe discount, I might

04:27 Christian Sobkowski: I'll try to scope this session a little bit and then a very, very curious to show and discuss and Dive in. I think. Actually is it is it okay if I share my screen with with your instance?

05:08 Art Koci: so,

05:12 Christian Sobkowski: Yeah. Okay.

05:13 Yulia Ershova: Absolutely.

05:16 Christian Sobkowski: We are we are creating these forecasts. For you right now. Really right now. Sort of, what's, what's being highlighted in, in the application is just a subset of all the things that sort of We're calculating in, in the back. And I want to use this session to give you a bit more of a view of what's happening behind the scenes on this.

24:18 Christian Sobkowski: and then, I want to talk about two things, I want to get two things out of it, a Which of those? Which is those patterns and things to discussing today. Should we actually highlight in the main application?

07:20 Amanda Mitt: Oh maybe we can let you know if something comes up during it if that's okay. Or I don't know if you want to present and then we yeah.

08:03 Christian Sobkowski: And they then stack up into balance forecast for the account. Which you've seen here. And then, all those balance forecasts stuck up together into global balance forecast. Does that make sense so far? Okay. Cool. Any any questions or remarks in that approach? First of all.

10:16 Amanda Mitt: Okay.

10:17 Christian Sobkowski: proxy that we can talk about today already on the categories itself. Okay. And so now, let me now let me go back then. so, What? the way this is filtered now is And Amanda, I took five of the accounts that you sent me on Tuesday. For this view. I think these are like, five of your bigger USD accounts.

39:18 Christian Sobkowski: Is that right?

11:21 Amanda Mitt: Yeah.

11:27 Christian Sobkowski: My gut feeling would be and curious to hear. Yours is first of all, that something like this is acceptable. where something like this is not

14:01 Yulia Ershova: It was exactly. The more is okay. Less is bad.

14:04 Christian Sobkowski: Less is always bad fair.

14:12 Amanda Mitt: because like we said, if we're not considering Intercompany just operating for example, like Categories. Then I think, Even low. A lower number could be good enough for us but if you want to be precise in every category and of course, as high as possible. Do you have also a target in mind when you build? The forecast.

42:47 Amanda Mitt: Like, what is realistic?

15:12 Christian Sobkowski: Is it's a 95% number. Something you're setting was in the fantasy.

15:17 Amanda Mitt: We have this goal. No, we have just seen this our Treasury goals. We have like 95% as this kind of Target. But That's it. Not in the finance team just in Treasury. Yeah.

15:36 Art Koci: Saturday, 95% matching results with your forecast or like 95% confidence.

15:46 Amanda Mitt: I think it would be like, in Categorization and then for example, 95% of the categories accurate and then of course I don't replicate into the forecast, exactly. But I think like,

15:55 Art Koci: Okay.

16:00 Amanda Mitt: above 90s, something that we aim for in, for example, the balance Forecast. I see it as good but by category don't have a number. We just yeah, usually we

16:07 Art Koci: Yes.

16:11 Amanda Mitt: say above 90% because it sounds like a good target to aim for but I don't know if that's realistic.

16:19 Art Koci: but yeah, maybe like if we look at these accounts, like, I think they're mainly within the 10% mark apart from the 2967, which is Quite off. Yeah, I think I think that's something we can discuss Christian probably. Just gonna make a note of this.

17:16 Christian Sobkowski: I, I agree art good. I think I'll play it back one more time. Amanda. Do you? Do you have a definition of the 1995%

17:27 Amanda Mitt: Yeah, I can share the goal that we have with you but it's basically when we

17:30 Christian Sobkowski: Mm-hmm.

17:33 Amanda Mitt: and then we also have palm. So I think the 95% is to get like how Reconciling the plan and the actuals between these three tools like to get to this kind of above 90.

18:00 Christian Sobkowski: Yeah.

18:00 Amanda Mitt: Acres between all of them.

18:03 Christian Sobkowski: Yeah. What what I meant is, um, Would you be, would you be calculating that on a Cash flow level or would you be calculating that on a sort of a global balance level?

18:17 Amanda Mitt: I think balance.

18:19 Christian Sobkowski: Balance right.

18:19 Amanda Mitt: but,

18:20 Christian Sobkowski: Got it. So, so basically you would want to, you would want to be able to say that we are, we are confident that our balance one quart from now is within 95%. Of forecasts.

18:35 Amanda Mitt: Yeah, for example, right? Yeah, we have to target off. Sorry.

18:46 Christian Sobkowski: You're a mutually.

18:49 Yulia Ershova: at least, I don't know if there is a possibility to split those two indicators because from the perspective we don't have around and we are not running a risk of having more cash, but we're running risk of not having cash at the right account at the right place.

20:05 Christian Sobkowski: I think we can look at We can look at two things I want to show you two things. I think the first one is maybe I want to go into two of these accounts, maybe this UBS here and maybe this this used to the account, is that good those two good accounts to go into

23:11 Yulia Ershova: I think the bottom one is relatively more predictable isn't it than our

23:12 Amanda Mitt: Yeah.

23:15 Yulia Ershova: investment account? As we can, we can start of course, like really.

23:21 Christian Sobkowski: How we can we can also is see this one and or one of these two, a more interesting one.

23:27 Amanda Mitt: I don't investment. I think it's that master. I see the cash flowmaster. No, the

23:27 Yulia Ershova: I mean, no cost.

23:34 Amanda Mitt: night. From on AG.

23:38 Yulia Ershova: Well let's dive in. We will see then definitely costs are more interesting

23:40 Christian Sobkowski: Okay.

23:43 Yulia Ershova: that's for sure.

23:44 Christian Sobkowski: The forecast was created. And here is the seeing then forecast versus actuals by week. For the inventory time series. This makes sense. What you're seeing here.

25:02 Rodrigo Cabrera: This is only not only for the account, but only for the inventory, category of the account.

25:08 Christian Sobkowski: Yeah. Yeah.

25:15 Amanda Mitt: One second, biggest. Let me.

25:19 Yulia Ershova: Doing, do we have any other categories actually on this account because it's quite stretch that there is just one.

25:25 Rodrigo Cabrera: transfers, right?

25:49 Christian Sobkowski: Exactly, exactly right. So we're we're not forecasting into company movements.

25:53 Rodrigo Cabrera: It makes sense. That I mean this,

25:58 Amanda Mitt: Yeah, makes sense.

26:01 Christian Sobkowski: Does that make sense Julia?

26:06 Yulia Ershova: Actually, thinking a little bit about it. Meaning that if it is a cash poll and that's a well we can let assume all entities need need cash, right? So they will take from the master account And if we don't capture it, Wouldn't it give us in reality lower but

26:31 Rodrigo Cabrera: we know that the full Picture.

26:49 Yulia Ershova: Okay.

26:49 Rodrigo Cabrera: Not just individual account the details.

26:51 Christian Sobkowski: Yeah.

26:53 Yulia Ershova: But at the end of the the day, we need the cash on the master account, not on the local account. so I mean since I can I can check the entity entity shows, let's say minus

27:02 Christian Sobkowski: Correct.

27:09 Yulia Ershova: We're not gonna finance the entity. We're gonna find the well if if there is no cash the master I'm just thinking if this the allocation of pool, Forecast is correct for the and it, I mean, for one side, Of course, I understand that we need to see how much entities is consuming for this.

27:28 Yulia Ershova: This part of the forecast makes sense. And but for the from cash perspective, like

27:29 Rodrigo Cabrera: it's

27:33 Yulia Ershova: where we need to have it,

27:38 Christian Sobkowski: Great question. I I can I can try answering of sort of what what we are looking In, it's sort of a cash pool view. Where you're having for all pool for all the accounts in that pool, including the header and the children, you're having the forecast of all the categories, but more crucially, sort of an overarching balance, forecast.

56:45 Christian Sobkowski: For the pool.

29:04 Yulia Ershova: Exactly. Yeah, exactly.

29:05 Christian Sobkowski: Is that right?

29:07 Yulia Ershova: So it's basically changing the view, right? So the correct for us is done on the

29:08 Christian Sobkowski: Good. Yeah.

29:12 Yulia Ershova: local entities and then we can like it would be it would be great. Exactly for how flexibility if I want to check it. Just the entity with all its accounts at once. Or let's say we send it to this currency combination and then if I just

29:21 Christian Sobkowski: Yeah.

29:25 Yulia Ershova: want to drill into the what's happening in the cash pool together, exactly. You mentioned that. That would

29:31 Christian Sobkowski: ultimately, You're looking for a view similar to this one here. For all put right now. This view is for all these five accounts, but really you would be looking for this for The 10 or 15 pool accounts, you have in that pool, right?

30:05 Yulia Ershova: Exactly.

30:06 Christian Sobkowski: and then, equivalent balance, balance you

30:11 Yulia Ershova: Yeah.

30:12 Amanda Mitt: Because I assume these are the top five that generate most from operations I think from all the company, right? I think so because It's JP Morgan in the US and invent meaning cachines in the US and then cash out. From inventory, for example, from on agf, I think, yeah.

00:00 :

30:34 Christian Sobkowski: it's

30:35 Amanda Mitt: You didn't even.

30:46 Yulia Ershova: It actually I think I think it is a very nice function going forward, I mean not maybe as a priority but so that the user can indeed set up its forecast view. So for instance I have just the pivot's moving like I need this entities or this accounts, a compile, my pool of whatever is important for me.

00:00 :

31:02 Christian Sobkowski: Yeah. Let it 100%. It's it's in the works. I promise it's gonna happen, Julia.

31:12 Yulia Ershova: Sure, I didn't.

31:12 Christian Sobkowski: But okay, so So, this would okay. Amazing. I do have one more question on this account. Oh, I quickly it. Did that answer? The question on cash, pulling? I think I have a good idea of what what you would want to see.

31:35 Yulia Ershova: Send it again.

31:37 Christian Sobkowski: Okay. Amanda anything you wanted to add?

31:42 Amanda Mitt: No, thanks.

31:42 Christian Sobkowski: So I think, right? One of the things that's important is this headline number over the next, you know, Four weeks. The next 13 weeks. How accurate are you getting? How important is getting it right in time for you. Right. So When, when you're looking at, When you're looking at for it, right? These, these overarching weep a week views of inventory on this account.

51:19 Christian Sobkowski: This is obviously on a week by week basis. This is not yet at 90%. Accuracy. But overarching. Over the next months, it is. How important is the timing for you? To get super accurate.

32:50 Yulia Ershova: Do you mean the timing to reach the accuracy or the frequency? How the forecast is showing?

32:56 Christian Sobkowski: The. The yeah, the frequency, right. So so basically Is, is it important how important is it? That we are getting each of these two bars aligned within 90%.

33:21 Rodrigo Cabrera: It's pretty. important, it doesn't matter if we forecast, 25 million this week and we had 50 million not going it and next week, forecasting, 50, and having to 25 We need to be. Into forecast when it will happen, the expense? No not a week after.

33:42 Christian Sobkowski: Yep.

33:43 Rodrigo Cabrera: Or are we before? I think it's pretty. Important.

33:51 Christian Sobkowski: Good, good. So, the the question then, the follow-up question is Right. Now, this is on the week by week basis. one of the one of the questions we have internally is Do you need this accuracy on a day by day basis? Do you need it on a week by week? Or do you need it on the ones by months? Because obviously obviously day by day is amazing but it's really really hard problem, right?

34:21 Rodrigo Cabrera: I think if we if we have it weak on a weekly basis, it's easy to go to get to the To the daily basis because we work on payment runs or at least on the payouts. We only always run the payments for Most of the entities on Thursdays for on US and Canada on Friday and for only on Monday.

34:44 Rodrigo Cabrera: So we know When are we expecting the cash out? So, if we know that, On the on the first week of October. There's a 50 million, we know for sure that those 50 millions will be on a Monday or on a Thursday depending on the entity. So I think if we have a accurate on a weekly, we will have it also accurate on a daily basis.

00:00 :

35:16 Yulia Ershova: Maybe also to add. So we I think it also depends on how forward-looking we're talking about, right? So I don't need to know weekly in a year from now.

35:23 Christian Sobkowski: Mmm.

35:26 Yulia Ershova: So and that kind of far away period. It can be months because I need just anticipate. Maybe something like the similar, what we have, let's say 13 weeks so which is the closest period. This is something that worth having weekly. I don't know. One week actually, one week from now daily, that would be the the most important as well.

35:47 Yulia Ershova: Then if this kind of a combination of frequency of the forecast is feasible. But starting, let's say three months from now, I would say monthly should be absolutely fine because we it the current point of stage we don't really care if in December, it will be beginning in our end for instance.

36:05 Yulia Ershova: So let's in January

36:10 Christian Sobkowski: Except. Make sense. Good. I'll one one final question on this part because this is so, right. It's I think it's a little bit around. If you had to choose priorities. So, When when Art, and I go back to the team. Do we go and tell the team? Hey, it's really important.

36:42 Christian Sobkowski: It's most important to get the global balance within 95% for all. Or Hey actually it's much more important to get every single one of these bars across accounts and categories within 90%.

37:04 Yulia Ershova: I'm in the second option is more granule, Of course, if you ask me this is the priority but I think It will kind of go for the stages right in order to reach the second option. You

37:13 Christian Sobkowski: Yeah.

37:14 Yulia Ershova: pass anyway through the first one or let's say the first one is reached by default.

37:18 Christian Sobkowski: Yeah. Okay. Good.

37:26 Yulia Ershova: I mean, it's always the option which is more demanding, right?

37:30 Christian Sobkowski: Of course, of course. Very good and no, but I I read a, it's a, there's no black or white. It's it's more of, you know, and Yeah, you know. But I'm hearing from you and what I'm hearing from some others as well as ultimately. There is quite a bit of use of just getting to the top level numbers, right? That's the first.

37:58 Christian Sobkowski: Like low hanging fruit or lower hanging food, use case? And then sort of right going into this which requires a lot more work on our side. But also to be honest in your side because you need to steer a lot more That, that comes sort of. A little later.

00:00 :

38:21 Yulia Ershova: I mean, in these case, basically, it's then the absolute target is to have the second option of granularity and that maybe the first global balance. This is

38:29 Christian Sobkowski: Yeah.

38:31 Yulia Ershova: the first step right at the same time. I'm just thinking in order to To bring the accuracy for the global. It is also kind of all the underlying must be potentially in line with the same percentage. Unless if it is just as you say low hanging fruit, meaning like to take first to clear all absolute extremes.

38:52 Yulia Ershova: And then going further inside. So I think it is The target potential for everyone is the last one, but from the process perspective, it makes sense to start with the global to clean it up and then go further to improve it.

39:08 Christian Sobkowski: Great. Agree. Okay. And I would love to go to the account that didn't perform. Well, you can chat about that. Okay. Amazing. So um because I think right like even So let me let me go into this. Again I think even we are already like one level deeper at least around.

39:32 Christian Sobkowski: You know, this sort of on an account level. I want to, I want to get it very much accurate which is already one at least one level more granular. Then global So, let's go into 2967.

39:53 Rodrigo Cabrera: Yeah.

39:54 Christian Sobkowski: And I think is it right? That this is like it's some what new account for you.

40:06 Amanda Mitt: let me recap, because this Is yeah.

40:09 Rodrigo Cabrera: I think it is, yes.

40:12 Christian Sobkowski: Yeah, right. Okay. And I think we as art and I looked into it earlier and I think there is like a little bit of that newness of not having a year of full data yet. That we need to account for better. So it's expect more changes here but let's let's quickly go in.

40:41 Christian Sobkowski: And so, so right but by the way. So I filtered a filter into the account and these are sort of the forecasters actuals per category for this account. And you're seeing that actually. Both both collections outperform significantly. But also payables were much much higher than predicted

41:08 Amanda Mitt: I think it can be danger company. Just my first step.

41:11 Christian Sobkowski: I think so too. Can we quickly go in?

41:13 Rodrigo Cabrera: Yeah.

41:17 Christian Sobkowski: Sorry, Rodrigo.

41:18 Rodrigo Cabrera: No. No. I think it's it's intercompanying. It it's actually not a new account, but it was recently,

41:30 Amanda Mitt: Be perfect.

41:30 Rodrigo Cabrera: it recently become repurpose and now it's a set ba account to A

41:35 Christian Sobkowski: Mmm. Okay.

41:36 Rodrigo Cabrera: So that that might be why?

41:41 Art Koci: I think it is not in your account, but some of the categories or relatively new So from the forecasting perspective, it's a new account category but yeah, I think there's payments from 2022 or something as well.

41:57 Rodrigo Cabrera: Yeah. Yeah.

42:12 Amanda Mitt: Yeah, so I remember it from the bulk updates. Maybe this account was in the past creating a lot of Issues. Because I think we were not using standardized. Descriptions in the statements, when paying their company? So

42:26 Rodrigo Cabrera: Wait.

42:29 Amanda Mitt: it's also an account that's managed not by us, but from the US team, So I think that also might be why we're still kind of learning some of the descriptions as well.

42:46 Yulia Ershova: Don't hear you Christian.

43:07 Amanda Mitt: Yes. So we have like an accounting team in the in the America. So in the North America South America and also in Asia, for example, and they all have

43:15 Christian Sobkowski: I'm, I'm

43:15 Amanda Mitt: Because the company was growing so they had like the accounting team doing treasury. Accounts payable. um, And then of course it's a different market, different banking. So it's always better to have someone locally. And then, we usually just manage excess liquidity or banking relationships with them, but like more on the deep dive on categories, and sometimes like salaries, for example, things like that, we don't really see day today.

00:00 :

43:39 Christian Sobkowski: Just making noises of Yes, and Aha. And

43:42 Amanda Mitt: as much as we see, for example, on Ag or on holding And also it was like I think of new process Rodrigo right to get them to send them more frequently the access cash from the US you on AG. So I think maybe if you see past data it was not like they were probably Keeping more cash there.

44:07 Amanda Mitt: And now it's more frequent where they repay. Intercompany. Much more. So it was a big change of behavior in this account. I think, in the past year, So tricky. Yeah.

44:33 Christian Sobkowski: How is that? How is out of interest? How's that set up at all? What's, what's

45:10 Amanda Mitt: Yes, I think historical data in some cases will not help that much.

45:15 Yulia Ershova: But it actually for it for intercompany. So let's say if it is not will it forecast, behavior of intercom, and on the local entity account,

45:30 Amanda Mitt: I don't make. No, I think maybe I know what the question is. Like, if it's Is it because it's being miscategorized as payments life.

45:42 Christian Sobkowski: the US team doing? Do you have like a local US financing? How is that? How is out of interest? How is that set up at all? What's what's

45:45 Amanda Mitt: So, I see. so,

45:48 Yulia Ershova: But then there should be in other category where the intercom is, right?

46:52 Christian Sobkowski: Got it. Yeah. Good. Understood. Okay. Yeah, but tricky but not not impossible. So I think, what's, what's in? This is a sort of, I think the second Way of questioning that, right? So this is so, there's two things that happen on this account from From what I'm understanding is a it's, you know, you you changed behaviors on the account, right? So Reasonably recently within the last half year, 12 months or so.

47:20 Christian Sobkowski: Then B, there is this is actually not. Not a fully clean time series because I'm assuming right? This this is an also

47:26 Amanda Mitt: I think makes sense. I think maybe the only, I see category that we could try.

47:29 Christian Sobkowski: into a company transaction actually. Right. Yeah.

47:31 Amanda Mitt: It's like the AP account, for example, this very

47:34 Christian Sobkowski: Do you think we should? Right. Yes. So here, hmm, I would actually be interested in I can quickly share sort of how

47:37 Amanda Mitt: Operational one that it's basically. the thinking that on ages like a vendor and for example, because on the

47:41 Christian Sobkowski: you're thinking about it. I'll be very keen to hear how you've done it. Before.

47:43 Amanda Mitt: so, Our approach has been. To not forecast intercompany. because it ultimately, It doesn't affect your global balances. And because that's what we did, when we did the direct cash flow model with the

47:48 Christian Sobkowski: Right? It's always going from one account to another, the same is for anything.

47:51 Amanda Mitt: FNA team. That's the only category that they would able to forecast. Because, for example, it's like the subsidiaries are paying a vendor, right? They pay inventor. It's

47:58 Christian Sobkowski: Vaguely resembling into company investments, fx are similar things. All of those

48:00 Amanda Mitt: pretty

48:01 Christian Sobkowski: are sort of actions that in the widest definition of word you make,

48:02 Amanda Mitt: Render, the behavioral behavior. We cannot maybe some sort of growth assumption

48:03 Christian Sobkowski: Either manual or automated. and they don't affect global balances, and So that's also why you're seeing for example right now, the the pool children

48:06 Amanda Mitt: into that and that's what we can, but all the kind of financing ones, it's very

48:08 Christian Sobkowski: accounts. Having a balanced forecast that goes away from zero. Is a positive or negative

48:10 Amanda Mitt: hard to predict. So the ones used. Yeah.

48:11 Christian Sobkowski: because it's just forecasting the operational activity right now and then once you start aggregating, on then then sort of all of artists canceling out

48:16 Yulia Ershova: Yeah fair enough to balance I I agree. I think that the Intercompetent doesn't

48:16 Christian Sobkowski: Yeah. Yeah.

48:19 Yulia Ershova: make sense because technically we should not do any payment if we see that balance is long so it shouldn't kind of be a risky point. So the intercompany should happen, only when there is enough cash and so on Okay.

48:35 Christian Sobkowski: Good two to that. And Operational, I see. Would you benefit from us forecasting? It

48:47 Amanda Mitt: I think yes, this one. Yes because Right.

48:58 Rodrigo Cabrera: Yeah, I think so. The but at the same time, if at some point, we start sharing our our open items and forecast with you, all of those invoices or or amounts, they will come in there to a

49:17 Amanda Mitt: True.

49:18 Christian Sobkowski: Okay, okay.

49:19 Rodrigo Cabrera: So, if it's yeah,

49:22 Christian Sobkowski: Then let's, let's solve it to that. It's I The team is hoping to get the first version live next week. Amazing.

49:34 Rodrigo Cabrera: Nice.

49:36 Christian Sobkowski: Okay. Um, Good. Let's let's solve it to there. I think I do have one more follow-up question on this account because so Doesn't matter if this is like a more challenging account or not, I do think no matter. What, what you do? There's always going to be. There's gonna be discrepancies.

50:05 Christian Sobkowski: Right. That's that's a whole reason you want a view in the first place. So right? So if you if you look at this this so in a way, My, my gut reaction would be that. If I go back, the right workflow is that Right. You're you seeing? You see in forecast with the actuals.

50:36 Christian Sobkowski: Imagine you only had five accounts. in all of on you seeing that, hey actually my global balance is, is reasonably accurate, but off by 50 million, And and you you would sort of then very quickly, see it through this and you would see Hey, the actual underlying problem is not that we We had a massive change in payables, but this is categorization issue and I Recateorized stuff and actually okay my My.

51:17 Christian Sobkowski: Accuracy is now back to 95%. Is that the right workflow that we should model out in the tool?

51:32 Amanda Mitt: I think that's what we have. In mind to set as a process. For to reconcile the variances yet.

51:43 Christian Sobkowski: Okay, then maybe the last last question for me on that. How often would you be doing this? Because this is not or Right. Is this something where? You would want to. On a daily basis. Understand, where where we are? This is something where At the end of the quarter, you're figuring out.

52:10 Christian Sobkowski: And I'm going into my forecasters actuals.

52:14 Yulia Ershova: I I think there are two two business cases for this one, is the for the quarter end, let's say if a result because definitely management would like to know, okay what are our Clothes or whatever, what our balances, by the end of the, the quarter or year end, especially and then we'll give them the number.

52:34 Yulia Ershova: And if this number number doesn't match, they will come back and say what's up. So this is, this is like, for, for the reporting line, for, for the Treasury day to day. I would say it depends how How? Detailed. We're gonna do our liquidity allocation. So for instance, if we we see the forecast for a week and then we put the funds in the correct spots to to be there.

52:58 Yulia Ershova: And then the next day, we will learn that there is an overdraft, Of course, we'll come back and see, like what's happened, why the forecast didn't meet? I mean, that's the worst case scenario, right? Like so,

53:12 Christian Sobkowski: Of course, yeah.

53:13 Yulia Ershova: But that can be so it's not like I don't think that we on daily basis will be logging in and check out what's happened in the past. We will be logging in checking, what is the forecast? And then potentially the the back date is needed either for reporting or finding the flows and adjusting, so that the future doesn't doesn't repeat it.

00:00 :

53:36 Christian Sobkowski: It's got a. Good. And then maybe as a follow-up on that. We need talk about the forecast. Would you like? Would you prefer to have sort of like, you know? And up to date forecast, where there's a forecast gets recalculated every day. but I'm assuming you're sort of also, You're looking, you would be looking at like, quarterly snapshots or something right to determine accuracy.

00:00 :

54:17 Amanda Mitt: Yep.

54:19 Yulia Ershova: Is a kind of is a result. Yes, but I would, I would say so for instance, if If on a daily basis, we see, let's say receive information that there is a big payment coming or something or, or in general, we need to see can for instance, I would like to invest money not to keep them on the count.

54:53 Yulia Ershova: So the first thing,

54:40 Christian Sobkowski: Yeah.

54:40 Yulia Ershova: oh, as I will open the forecast and see, Do we have a money? Like, What is the minimum balance will reach and what is the extra we can take. So, technically, I

54:44 Rodrigo Cabrera: M.

54:49 Yulia Ershova: think this is the question like so that the the view of the forecast is updated on the day when I have entered it or all and it

55:02 Amanda Mitt: I think that's the most incredible like Wow, because it's super online. So I think that's what's always a challenge, like a challenge if you're doing focused and they have to do like quickly manual calculations.

55:15 Christian Sobkowski: Yeah.

55:16 Amanda Mitt: I think it's really nice but then of course, if I'm the one doing the reporting and then I have to see how The model is performing, then I'll have to have some sort of snapshot to prove that we're improving. The forecast somehow. But I don't know how.

55:31 Christian Sobkowski: Good.

55:33 Amanda Mitt: Important, that is to us now. Like I think if we have Nice categorization like nice you know I don't think we need to like focus a lot and review me snapshots for now because I don't think that would provide most value. I don't know, but maybe I can check with.

55:55 Amanda Mitt: the team, but I don't see us like Going pretty on the snapshots and Yeah. Because that's not the goal, right? I think that's very old school way of doing it. No, that you keep

56:13 Yulia Ershova: On only for the target of improving the forecast, right? Especially if this machine learning, then the more let's say, Information, we give on this discrepancy, potentially the better. It should become. But not for not for business. I would say. Yeah, for business. It's only the

56:34 Christian Sobkowski: Good.

56:37 Yulia Ershova: reporting as you said. Also to show the management that whatever figure is we give them, they are really well reliable with a certain confidence level.

56:51 Christian Sobkowski: Understood. Okay. We just spent an hour in this. Thank you all so much for all of that. Any any closing words from you?

57:06 Yulia Ershova: Can I ask one question before we move to the closing for, for this, for this

57:08 Christian Sobkowski: You mean if it's possible to click on the bar,

57:11 Yulia Ershova: already, which you have a, so that the majority of the times you are selecting, let's say count or category in the upper filters and do you plan to keep it there as the categorization, or as let's say, when you click, let's say on the bar and then it opens the transaction little details.

57:33 Yulia Ershova: and the question, because I find that sometimes quite convenient, then instead of, you know, going to another report and going the through the details, I can have kind of prefiltered view, but it's, I mean, otherwise, I think the overview which we're getting from Palm is is very good. I love this view that we have the categories and so it's not only the the prairie count and then you kind of need to to see yourself but also the category straight away.

58:05 Yulia Ershova: It shows you like which one is off which one needs to to concentrate. So I think it is this very useful so even even there are a lot of comments. Let's say what we

58:10 Christian Sobkowski: It's, it should be coming. Yes. I would.

58:12 Yulia Ershova: would like or not the current use already very good. Have a very good usage for

58:13 Christian Sobkowski: Agree. Amazing.

58:20 Yulia Ershova: For our day-to-day. So I think At least from my side, I haven't advanced. I haven't been using it a lot. So we're now like kind of pushed kiriba but now we have there forecast so I'm checking it. And I think that now it's the, the perfect time to kind of align also with Parliament to see what's.

58:39 Yulia Ershova: If there are lots of discrepancy in what is what is driving them. So, Thank you.

58:46 Christian Sobkowski: Thank you so much, and

58:48 Yulia Ershova: Amanda you have anything like that from from your side, you think?

58:51 Amanda Mitt: No no that you just mentioned the how frequent the forecast is updated. Currently. How frequent is it? Is it like for the

59:03 Art Koci: So we generate forecast twice daily and once a week, we determine what models to

59:03 Amanda Mitt: Okay, thank you.

59:10 Art Koci: use on the forecasts. So once a week, we determine what we're using and then two times a day, we generate new forecasts. And yeah the frequency we might be going to one forecast per day since most of our users only update stuff daily but currently we do it twice a day.

00:00 :

59:39 Yulia Ershova: I think one or twice. It's it makes sense. It shouldn't be a very critical. I

59:39 Amanda Mitt: Dear, thank you.

59:45 Yulia Ershova: don't know if you have dependency on the import of bank statements because in our case we have all the type of time zones and the cut of times from the banks. So, us especially will receive, I think 12 or one. So, I don't know how it's the best for the system at which point of time to update.

00:00 :

01:00:05 Art Koci: Yeah, so that's I think the original reasons why we have twice a day and nowadays, we have fixed that but we it's just like a C. Yeah. There's no no real reason to have it twice a day but it it's very easy to do so that's why.

01:00:26 Yulia Ershova: I mean if it is not an operational or let's say the the processing pressure for the system then of course twice they would be more reliable because then in the morning it can check. I know that I can check all the regions, let's say except us and then later the day potentially us will have the updated.

01:00:45 Yulia Ershova: Values.

01:00:50 Christian Sobkowski: Very cool. Thank you so much for this. I think quite a bit more to come both cache pool views, as well, as another video was curious on currency balance, forecasts and well, that's coming. And I think focus is actually, you should be a proper one, that's really nice and easy.

01:01:11 Christian Sobkowski: Support should be going live in the coming. Okay, so weeks as well.

01:01:18 Yulia Ershova: Perfect images. Thank you very much. Yeah. You know the wishes

01:01:18 Amanda Mitt: Thank you so much, really cool.

01:01:20 Christian Sobkowski: Amazing. Thank you. All.

01:01:24 Yulia Ershova: Thank you.

01:01:24 Christian Sobkowski: Appreciate it. Take care. Bye.

01:01:25 Amanda Mitt: Thank you so much.