If you would like to earn CPE credit for listening to the show, visit earmarkcpe .com slash FPA.
Download the app, take a short quiz, and get your CPE certificate.
Finally, if you enjoy listening to FPA Today, please go to your podcast platform of choice, click the subscribe button and leave a rating and review of the show.
And now, onto the show.
From Data Rails, this is FPA Today.
Welcome to FPA Today.
I'm your host, Glenn Hopper.
Today, we welcome Nate K.
Mink to the show. For two decades, Nate has been immersed in some of the most complex, i .e., nerdiest math and statistics out there, starting his career as a mechanical engineer.
He honed niche skills in math and statistics, and these skills he later discovered, thanks to some insights from his MBA cohort of commercial bankers, could be a powerful asset in forecasting and financial modeling.
Nate's journey took him to product demand and financial forecasting at Cummins Inc., which is a Fortune 500 giant for those who don't know it, where he applied these specialized techniques to deliver next -level insights.
After serving in various roles, including as CFO, Nate saw a gap in the FPA and forecasting tools on the market, tools that couldn't keep pace with the sophistication that he knew was possible.
In response, Nate founded Better Forecasting, which launched its beta AI -powered forecasting product in April, and with a recent major update, they're poised to transform how FPA professionals spend their time automating the heavy lifting in financial modeling and creating new opportunities for strategic
insight. Nate, welcome to the show.
Thanks for having me.
What are these nerdy, niche math concepts, and why do they matter so much for forecasting?
Oh man, starting with the heavy hitters here.
Yeah, let's go right in.
It's probabilistic forecasting, which is different than deterministic, and we can get into that if we need to.
It's small data, not big data, because in financial forecasting, we don't have big data.
We have, if I have five years of financial statements, I have five Octobers.
That's five data points.
In inferential statistics, and the difference between why I say inferential statistics is because there's a big difference between descriptive statistics and inferential statistics.
So yeah, those are some of the key concepts and some things that I happen to have learned as an engineer.
And then when I went to start doing forecasting, it was like, oh, wait, I guess finance people don't get degrees in advanced statistics.
So there's maybe a need that I can fulfill here.
So yeah. A little off topic because this is just interesting to me because I travel around, and as we were talking before the show, talking about implementing AI in finance, and finance and accounting got passed by sales and marketing in machine learning and use of AI early on.
But to your point, if you have an e -commerce site and you have all this data, they actually do have big data where we don't have it in sales and marketing.
So we know these concepts and we hear them out there and we see what sales and marketing are doing with these big datasets, but we don't have big data.
But we still want to benefit from being able to use machine learning and make data driven decisions.
I guess what you're saying is we have to have a different approach.
We can't try to use the same math.
Yeah, totally. I don't know that many finance people are going to get into this, but a place to look for technology is our libraries and Python libraries that are available like through just packages you can install.
So a good example would be Facebook's Profit.
And I've actually seen some CFOs and FP &A teams that use that and they'll learn Python and they'll use this library.
The problem is that Profit itself...
That's a aren't apparent until they cause bad problems.
And so part of the reason for that is because it was built on technology that's used for big data.
The definition for big data for purposes of our discussion is anything with a thousand data points or more.
So there's some statistical convenient assumptions you can make when you have above a thousand data points that you can't make when you have less than 30 data points.
And so the issue becomes like, hey, it's easy.
It kind of works. So we're going to apply it here.
And it'll get you a good starting point.
There's a couple of major caveats or pitfalls that can end up causing some issues with your actual forecast down the road.
And so the reason I talk about those libraries and what's available is I look at a lot of them and most of them are doing the same thing.
They're applying big data concepts and they're trying to make it shoehorn it into our little shoe box and it does okay.
But when we're talking about...
We'll talk about this.
It sounds like we can get into this if we need to.
3 % to 5 % miss on a revenue forecast is a huge deal to our cash.
And so the difference between a big data machine learning versus an area -specific small data that we're applying could be that 3 % to 5 % difference.
And that's a big deal.
Yeah. So I do want to get into all this, but I guess maybe we need to level set a little bit more than what I did in my introduction.
And I touched on it in the intro, but tell me a little bit about your background and how you ended up coming to do better forecasting.
Yeah. Well, and I think you put your finger on it.
So I actually started as a mechanical engineer.
So at the time, this was 20 years ago.
So two decades ago, I was trying to solve a chemistry, physics, and mechanical engineering problem in terms of emissions for diesel engines.
There was no papers written.
There was no first principles research on the area that I was doing.
So I had to go and I couldn't find a science paper on this is the chemistry.
I had to use inferential statistics to help me figure out what's going on.
Now, what was interesting is it was $10 ,000 per data point to collect a data point.
Now this was 20 years ago.
So $10 ,000 was a lot more money at that time.
I think at that time, it was a huge portion of my annual salary every time I said, oh yeah, I need one more data point.
So because of that specialization, when I went into financial forecasting and I went to get to do product planning at Cummins, which I got to play with lots and lots of zeros when I was there, even the finance people and all of it were like, where did you get this math from?
Because nobody's applied this type of math to our space before.
And so that's what got me thinking about this.
My dad was a business owner.
I always wanted to start a business.
And then it was like, well, how do I make it as easy as possible for somebody in finance to be able to use all this advanced statistics without having to go get a master's degree or a PhD in statistics?
And so that's what our software is doing.
I'll talk about a lot.
We could get into how nerder you want to in the math.
But the reason for doing that is highlighting a couple of things that we've already handled.
So if you want to use this technology we're talking about, you don't have to go learn R and you don't have to go get a master's in statistics.
You've already got a degree in accounting.
You already understand your business.
So we're trying to take that complexity off of your plate and make it just automatic.
So do you know the site Data Robot?
It's a drag and drop machine learning...
Oh, I haven't actually played with it.
I've heard of it, but I haven't.
When I was studying analytics, we got free access to this.
And it's like a $50 ,000 a year subscription.
So super exciting to get these drag and drop models that you could do and do anything.
But there were people using it who had no idea what they were doing.
They didn't know the difference between clustering and classification.
And they were using completely the wrong models.
And it's a very powerful tool, but it's a very powerful tool like driving an F1 race car.
You have to know the basics of how to do it.
And I'm wondering...
So I think it's important that people, if you're going to be using these tools that you understand.
But it sounds like, with better forecasting, maybe you have guardrails in there and it's like, we're going to take you down this path.
Yeah, and we do. And that's actually probably the part of this journey that's been almost the more difficult one.
And it's how do we make this usable for somebody that doesn't care, doesn't need to care.
Our goal is, if you're using one of our AIs for forecasting, it's going to do all of that stuff for you.
It's specifically trained in these methodologies.
And all you have to do is say, well, okay, for this line of business, consider these three line items in our CRM.
Or if you're trying to do a rolling forecast and you're looking at an expense line item, let's say it's...
I'll call it... Insurance is a good example.
All you have to do is tell it, well, here's our labor code.
This line has labor.
This has admin labor.
This line has direct labor.
And it does everything else for you.
That's all you have to do.
We're really trying to make sure we've created a way to capture the business knowledge of the FP &A person and their ability to partner with the business.
But then all of the number crunching and everything, we make it as easy as possible.
Now, we also try really hard to do integrations with Excel because I love Excel.
It's interesting how many times I see someone, oh, Excel is dead or whatever.
I always laugh when that happens.
Because Excel is such a powerful tool.
And I actually originally, when I started building this, I was a CFO and I was starting to try to build my own forecasting.
I actually built several of these models in Excel.
We ran into a processing limit at 30 models in Excel itself.
And that's when I moved over to Python.
And now it's, okay, well, we're just going to try to get you as accurate as possible driver -based forecast models for every line in the chart of accounts.
And then we also will do a three -statement model.
We'll create a three -statement model that somebody can use.
Or you can use your own and plug it in.
We're trying to make it as flexible as possible.
We're just trying to elevate the statistical power of your existing process and the existing tools that you're doing.
You mentioned something earlier that we had talked about before the show, when you said rolling forecast.
And I think that this is to this day...
And really, people in FP &A, for the most part to me, they kind of get the difference.
It's the rest of the company that you have to educate and explain this to them.
But the core difference is between a rolling forecast and an annual budget.
And I guess, why do you think it's so common for people to confuse the two?
And what do you think leads to that misunderstanding?
I'm going to talk about this the way that I would talk to a CEO or to a non -financial person.
Because I find, especially when we as finance people are talking, oh, that's how that guy explained it.
So maybe I can use that to explain this to my CEO.
So think of it this way.
I'll start with a budget, and then I'll move into a rolling forecast.
So a budget actually contains four different key pieces of information.
The first one is it has a target.
It's a target that we've gone through a consensus -based process to identify, hey, this is our target for the year.
We'll use some fictitious numbers that I threw together just to kind of think of a way to talk this through.
The second piece of information it has is a forecast.
So a forecast is going to be, well, here's the expected outcome.
Now, there's a forecast and a target inside the budget.
And when you finish the annual budget, they should be the same.
It should be, this is how we expect to get to the target.
There's actually two other pieces of information in here.
And that's the plan.
The plan for what we're going to change to get there.
And then there's also the expected outcome of the plan.
Now, I'm using these four words, and I'm going to walk through a word picture so you guys can have an idea how to talk about the difference between them.
The first thing, let's say we've got a $12 million a year line of business just to make the math easy.
So it's a million dollars a month in revenue.
But we have a budget session.
And in the budget session, we say, hey, we're going to hit $24 million in this next year.
Now, I know a little bit out of reality, and there's caveats to all of this, but I'm just trying to make a point here.
So that means we're, come January, we're going to hit $2 million a month in revenue.
So we put together a plan.
Here's the forecast was 12 million a year.
This plan is going to get us from 12 to 24.
The end of Q1 comes around, and we have our Q1 actuals.
And we actually hit 1 .5 million a month.
So we only achieved 50 % of our plan.
That means we're at 4 .5 million at the end of Q1.
And the traditional way that I see people do this, we all do this, right?
Is you do your 3 plus 9.
And you say, like, okay, well, we've got 4 .5 million plus I'm going to have an 18 million for the last 9.
That means we're on track for 22 .5 million.
The issue is that's not actually going to happen because your plan is only 50 % successful.
So the forecast shouldn't say 22 .5.
The forecast should say 18 because you're at a run rate of 1 .5 million.
But now, but your target's still 24, your forecast is 18.
And what we have to do at the end of Q1 is update our plan.
We've got to have a new planning session.
We've got to have a way to make up the $6 million gap.
Because we all know that we agreed to the target at the beginning of the year.
That's what our bonuses are based on.
If you're publicly traded, you've already told your investors.
So it's important to understand the key differences.
And that's that, okay, well, now my second plan, call it a different plan.
But it's going to be, here's my $18 million.
This is my forecast.
I have an updated plan for the $6 million that I have to make up to hit the target of $24 million.
So it's complicated, but we have to recognize that a budget contains all four of those pieces of information and a whole bunch of assumptions that it's all going to work out and aggregate into the consensus that we agreed to.
And so the rolling forecast, when I'm doing a rolling forecast, I always have to ask people like, okay, is this a forecast?
Or is this a plan? Because they're different.
A rolling forecast, to me, the forecast is a peer.
This is the trajectory we're on.
This is what is going to happen if we don't make any changes.
So no changes, no management changes.
What we're doing is working.
The existing plan is there, which is different than the target.
If you're lucky, they're the same.
If your plan is 100 % effective, which they never are.
But if your plan is perfect, then your forecast is going to say you're going to meet the target.
So I always have to go into those.
And I think what I say when I go to a company and I say, hey, where's your forecast?
Somebody will give me a budget.
And I always say like, no, this is a budget as a forecast, which is BAAF.
And I call that baffling.
It's really important to understand the decisions that need to be made around, okay, we need to change our plan because the forecast is not working or the existing plan is not working.
The forecast is telling us we're going to miss the target.
You're giving me flashbacks to my early days in FP &A.
And back then, I don't even think it was called FP &A.
I was just the finance guy, but I was in telecom and we had very aggressive budgets, a lot of M &A activity.
And we would give the annual budget to our investors and we would go, we'd get through the first quarter and we'd be 20 % behind budget.
The CFO, which we're not going to recast the budget after one quarter.
So, you'd have your actuals, you're behind on budget, but then the budget for the rest of the year, you still, when you're showing that full 12 -month view, the budget stays the same.
So, you're not even making a forecast to say, this is the reality.
It's just put the pressure on sales, we've got to make it up or whatever.
And you're intentionally skipping the fact that you need to update your plan.
To me, it's really important that the forecast is, this is the most likely outcome.
Now, the rest of that process is important.
It has to happen. But if we don't acknowledge that the pieces of data are different, then we end up just ignoring the problem.
And that's not useful.
So, I always say bad news early is good news.
Because if you get the news early enough, you can fix the problem.
Now, all of this puzzle piece pulling together is one of the reasons why a rolling forecast is so important.
Because a rolling forecast should be like, I always like to show forecast versus target or forecast versus budget.
And so, the reason that companies fall into this trap is because, like we said, because the budget is the forecast.
They call it, they're not recognizing the different pieces of the puzzle that are in play in this.
And again, we're preaching to the choir here, but sometimes that's what's fun about the show.
So, I think it's important that we talk about what the purpose of an annual budget is and how that supports the decisions and alignments.
And then on the flip side, the pitfalls to traditional budgeting, and it's really, it's that Mike Tyson quote.
Everybody has a plan until they get punched in the face.
I think the annual budget is an opportunity, I think it's important to get consensus on what we're going to do.
It's important to set the target because that's when you're saying, okay, this is our priority.
This is what matters the most this year.
Here's how we're going to go about doing it.
And I think I've ran into this problem quite often.
It's a similar one to what you're talking about, Glenn, and that's this, okay, well, hey, the forecast update says we're going to miss the budget.
And here's the market driver reasons for why.
And so, a lot of times, the operating team will try to say, well, that's our new target.
No, no, no, no, that's not our new target.
That's not okay. Our target is the target.
And I think it's really important to set a target.
I think that that's probably the most important piece that comes out of the budget of, here's the target.
Here's how we're going to get there.
And then when you're doing your quarterly reviews or your monthly rolling forecast updates, you should just be tweaking the existing plan.
You're not writing a completely new plan.
You're just saying, okay, well, hey, we've got to fix about 20 % of this plan to still get to that end target.
I think the year -round budget is somewhat arbitrary.
I know we're used to doing it that way.
I'm probably again preaching to the choir here.
But I think the real purpose of the annual budget is here's the target.
Here's where we want to go as a company.
And people spend a lot of time thinking about that.
And then I think one of the other mistakes that I see is the assumption that a rolling forecast update is a rebuild of the budget.
And that's why this distinction is important.
When you're doing a rolling forecast update, the forecast itself should be really simple.
The thing that people run into is they say they try to rebuild consensus around the forecast every time they do a rolling forecast.
And that's not what we're trying to do.
The forecast is just, here's where we're most likely to end up.
And then you have the option of saying, do we need to update our plan?
I'm thinking of how many board meetings I've been in or even with the rest of the management team where you start to see the reality.
And it's really weird from the office of the CFO.
You start to see the reality that you're going to miss the budgets that everybody planned to, but then nobody wants to recast the budget because that's, I know we said, we all agreed upon this.
We're going to do this.
So you're trying to set expectations, whether it's figuring out cashflow and just all the parts of where we actually are going to be.
Whereas sales and marketing are driven by, we've got to have this much revenue and Ops has their own goals and people lose their minds on it.
And then somehow the finance person becomes the bad guy, shoot the messenger.
And I agree. And I also think that one of the reasons people have a resistance to changing the budget is because it's a lot of work.
It is a lot of work to put together an updated budget in Excel with those tools.
Now, that is also one of the reasons why we're building the tools that we're building.
And it's because you can do this with some machine learning.
You can do it with a really good sophisticated Excel spreadsheet, but it's this idea that, okay, well, actuals for March just came out.
And I've heard all kinds of stories of, well, how long does it take to update my rolling forecast?
So I've heard it takes two days, which would basically mean you're dumping the data and you're not doing any model updates or reviews, to it takes eight weeks.
And on the eight weeks end of the spectrum, you're probably updating all the models.
You're actually doing a good job.
The problem I kept running into is like, well, I would like to do that level of sophistication every month, but it's too much work.
And so I'm stuck with doing a sort of an update.
But in reality, because plans not working, I need to change my model for a plan and that adds time.
So we're stuck in this continual trade -off between, well, how much time do I have to update the forecast for how accurate can I get it or how much capability can I provide?
And so if you move over to like a true machine learning or an AI assisted based process, then what used to take, let's call it a week out of every month for an analyst to go and do an updated, if you're doing a pretty good update to your rolling forecast, AI can do that in minutes.
You're not doing the consensus building, but you are at least rebuilding the picture of the most likely outcome.
That's very easy to automate.
So with that, I mean, maybe this is something that even in finance, I think we may be lose sight of or forget.
So we know the main purpose of the annual budget and you've got the, this is the strategy for the company.
This is the sales and marketing plan, how we're going to get there.
This is what everybody agrees to spend and what it's going to cost and all that.
But then the rolling forecast almost becomes like the excuse for not hitting your plan.
But walk me through, I mean, there is, obviously there's advantages to rolling forecast as well as pitfalls too.
But walk me through the main objective of these rolling forecasts and especially in helping the organizations kind of anticipate issues.
Yeah, totally. Two main areas that I always try to use a rolling forecast.
The first one is cash because you'll put together a pretty sophisticated, okay, if we hit budget, this is where cash is going to end up being.
But to me in a rolling forecast, it's actually really important to know, okay, cash in 9 months or cash in 12 months is going to be this.
And I want to update that forecast every month.
I'm not talking about the 13 -week rolling cash forecast.
I'm talking about a 12 - to 18 -month rolling cash forecast.
And the reason we want to do that is because if you're not hitting budget, or even honestly, it's probably the other way around, if we're hitting budget or we're exceeding sales budget and our receivables are 90 days, then we're running the risk of using more cash than we plan for or needing more cash
in working capital.
We all know how that works, right?
But to me, the updated forecast is the rolling forecast is going to look for that cash issue to occur.
Because you're, hey, oh man, I've got 90 days or I've got 120 days.
I better go increase the cap on my line of credit because I'm going to need it.
Or I got to go talk to sales and say, hey, it's great that you're selling this much, but we've got to change our default terms from 90 days to 60 days for everything you sell from this point forward because that's the only way we're going to have enough cash.
Now, because I had that rolling forecast and I could see what levers I had available to play with, I was able to implement a temporary, hey, for the next two months, we've got to go 60 days.
And if they want 90 days, then you've got to come talk to me about it, even though your normal is 90 days.
Does that make sense?
So you have the ability to react.
And it's this idea that bad news early is good news because you have time.
If we go to the revenue piece of the puzzle, it's market chaos, right?
I mean, we've all been through tons of that chaos for the last couple of months.
And it's amazing how often you'll have people that say after something crazy happens in the market, people will come and be like, Oh man, I think I saw that starting to happen six months ago, but I just didn't recognize that that was what was going on.
And so the other reason for a rolling forecast is if you do a rolling forecast every quarter, you might have actually, when you get to say April, you might have actually had an indicator that the market was starting to soften in February.
But because you're not updating the rolling forecast, you've missed out on two months out of say six of time that you had to respond and correct and course correct as a business.
And so I don't, I think that what I would like to make sure we understand the separation between just because you did a rolling forecast doesn't mean you need to change your plan.
So do the rolling forecast.
It's going to tell you how well your plans are working, where you're most likely to end up.
And then you can make a conscious decision not to do any budget updates or not to do any plan changes because you're okay with where it's going to end up.
And to me, the point is to cause that discussion to happen.
It's actually very important that that discussion happens because if you're not separating the budget from the forecast, then there's a default answer of, no, we're not going to make changes or we're going to intentionally not choose to make that decision.
As you're talking through that, I'm just kind of going through all the budget recast items that I had.
So like I said, it seemed like the rolling forecast always became sort of the excuse deal for missing the budget and then nobody wanted to redo the budget.
And maybe sometimes you didn't have to, I think about all the frustrating things we had to do to satisfy because we didn't really separate this as the budget and this as the forecast.
So most of my career was in PE backed companies and they're just looking at everything, just straight numbers.
They're great modelers and they're doing their forecast based on what you have.
And then I found that the conversation we would get in all the time would be, well, okay, we're behind, but this is a timing thing.
See, we were going to close this big client and they're still going to close, but it got pushed into Q2.
And oh, and we had this capex project, but we had to move it up because we ran out of capacity or whatever it is.
So because you had budget as forecast, you couldn't just do this sort of the mathematical like, this is our trend.
This is where we're headed because then you'd have to go change everything and say, well, I know we budgeted this in June, but we actually pulled it up to end of Q1 or whatever.
But I think just differentiating between the two and keeping that clear, maybe that helps some management and maybe some, there's some unnecessary work when you try to jam the two together and make them the same thing.
Yeah, definitely. And I love that example.
I'm going to shift a little bit to leading market indicators and then we'll come back to why this is important.
So there's been talk for the last two or three years with the Federal Reserve interest rate and the rate coming down and all this like, oh man, are we going to hit a recession?
Are we not? Are we already in a recession?
I tend to try and stay away from global -based starting things.
And we really want to focus on, well, is my company's market getting soft?
And a couple of things to think about is that when money gets tight, the first behavior change is that people will wait longer to make decisions.
They'll try to hold onto their money and save up for a little bit longer.
The second behavior change is, okay, now we're going to go do price shopping.
We still want to buy the thing.
We still need to buy a new car, but we're going to try to maybe find it for cheaper.
And then the third behavior is, well, we're just not going to buy the thing.
Now, I can use that in consumer behavior, but it's the same in business.
And so what's interesting is when you use that, like, oh, it's a timing change, it's actually really important to update your assumptions on the timing of everything else downstream.
Because if the one first big project is delayed and you don't have a timing model set up in your forecast, okay, great.
That $6 million project got pushed back by a quarter.
But what happened to everything else that was going to be in Q2?
It's not necessarily a one -to -one, but it's actually really important from a modeling standpoint of how healthy is my pipeline.
One of the key pieces of the equation is, well, how long is it taking me to close?
Let me take this back to your example.
If timing happens and I have a forecast that's totally separate from my budget, then I can plug in that change of assumption to my forecast.
And it's say like, well, if this one's changed and that's say 30 % of our business or 30 % of sales plan, I have to evaluate the scenario of, well, what if 30 % of our entire pipeline for the rest of the year is delayed by a quarter?
That's actually a really important piece of analysis to go to.
And you wouldn't be able to do that if you're not keeping the two separate.
Yeah. And that's so just kind of walking through building the budget every year and thinking about that.
So one of the things we talked about before the show was using the kind of the snapshot from the rolling forecast to be the basis for starting the budget.
And I think looking back at my career, I mean, there are times where you know, you know, sort of where the trends are and you're growing fast or whatever the big macro trend is going on and you use that, but then I think about budget season and we had the idea of run rate, but run rate was an average
of the however many months prior, maybe it's 12 months or whatever that has, there's no seasonality, no trend information.
So take, you know, going away from the revenue forecast, just looking at expenses that that's where whether we were, you know, it was always some mix of kind of top down and bottom up budgeting, you know how that always works out.
But when we give people the run rate, there's no there's nothing talking about the trend and what that's showing.
It's just, well, here's the average.
This is where you start your budget.
So maybe talk a little bit about really a better way by using that forecast rather than just doing the average run rate and how that could be used to start the budget process.
Yeah, yeah, definitely.
So this is when I'm looking at doing this, I'm any number that we're going to use and project or this is where we get into the inferential part of statistics, right?
So descriptive statistics would say, I'll use accounts receivable, because that's a pretty that's a bit of an easy one, right?
So we're all going to use DSO or day sales outstanding and say, well, we have an average of 47 days outstanding.
And then you'll project that forward in your infer what's going to happen in the future.
Well, the issue is there's DSO is actually seasonal.
If you if you look if you take a look at it, there's actually seasonality associated with DSO.
There's quite often, well, it depends on which line of business.
So you'll have one line of business that's 90 days, the other line of business that's 30 days.
And if you have a change in mix, then okay, that's modelable.
But the other part of this is well, if the cash gets tight, or if interest rates skyrocket, which is what we just saw happen.
Now, all of your customer CFOs are having it, you all had this discussion.
But all of your customer CFOs and FP &A teams are saying, well, if we just hold on to our money a little bit longer, we can pay down our line of credit and we won't pay as much in interest.
Well, your customers are doing the same thing.
And so it's a huge mistake to just assume like, oh, well, we're going to use the last five years average DSO with 47 days.
So we'll project that into the future.
Well, there's actually been a trend for the last 18 months where we've gone from an average of 47 up to say, 52 days.
And that trend is heading in that direction.
And that can figure out what the market drivers are for that trend.
But it's pretty common to see that...
And let me take this back to the very beginning.
Is that any ratio that I'm looking at from the past, I'm always going to ask the question of like, okay, well, is there a way I can figure out how this ratio has been changing and use that going forward?
So DSO is a really easy example because everybody's always working on it.
But if I see a trend that DSO has gone from 42 to 47 days for the last 12 months, it would be prudent for me to assume it will grow by another five days over the next 12 months.
Now that doesn't mean that would just be the trend.
That doesn't mean we think the trend is going to happen.
But it's important to have that as a starting point.
FP &A today is brought to you by DataRails, the world's number one FP &A solution.
DataRails is the artificial intelligence powered financial planning and analysis platform built for Excel users.
That's right, you can stay in Excel, but instead of facing hell for every budget, month -end close or forecast, you can enjoy a paradise of data consolidation, advanced visualization, reporting and AI capabilities.
Plus game -changing insights giving you instant answers and your story created in seconds.
Find out why more than a thousand finance teams use DataRails to uncover their company's real story.
Don't replace Excel, embrace Excel.
Learn more at DataRails .com.
So, you know, 12, trying to figure out DSO, it is, you know, that's the biggest to me was always the hardest part of the three statement model is, you know, you know, when you're selling and you know what you're kind of where you are with your, your collections on your AR and everything, but trying
to make it all tie out to the cashflow statement, and you know, it's one thing, 12 months is hard enough, but even beyond that, say over like 24 months, you're guessing at that point, you know, you have to factor in the exogenous factors and all the stuff you were just talking about.
But so really a two part question to this one, why is long range cash forecasting, you know, over that extended period, so hard and, and then maybe an example of how bad it can be if you have that forecast wrong.
Yeah, definitely. So I'll use I have another kind of stick frame company that I'll use for this as an example.
So let's say we've got a company that made, you know, again, $10 million in 2023.
And they've got about a 30 % gross profit.
That means their, their gross profits $3 million.
And then let's say their expenses are 2 .3.
Their net profit is $700 ,000, right?
So fairly typical, you know, 7 % net profit margin company.
And they've got 90 days receivables and 30 days payables and inventory is 20 % of COG.
So pretty standard, you know, we'll call it a manufacturing company, if they grow by 10%.
And we continue those ratios forward, then they're going to have a total change in cash of somewhere around $400 ,000.
Right? So just 10 % growth in revenue and everything else stays the same.
The issue is, is if I have a 5 % error in my revenue forecast, that would be $550 ,000.
Well, my total cash position change was $400 ,000.
So a 5 % error in revenue is more than my entire cash position.
Now, obviously, that's not a one to one relationship.
But it just highlights why even a 12 month cash flow forecast, how sensitive it is to very small forecast errors on the front end.
And so because of that, because we've spent so much time trying to do 12 month, 24 month cash flow, I'm always looking for that you guys have done sensitivity analysis, you can use Monte Carlo to do this, you can do it, you know, you can use one OFAT, it's called one factor at a time, you know, there's
all kinds of ways to go ahead and build a model for it.
But it's interesting because of how sensitive long range cash is to small changes in the model itself.
And so what do we do about it?
Right? And so the reason that in the first place, Nada is the CFO back before I started this company, the reason I started getting into machine learning and inferential statistics and all of this stuff is because I was trying to improve the accuracy of these items so that I could get a better picture
of what cash should be down the road.
And the reason I cared so much about cash is because if I had nine months preview to a cash problem, that gave me enough time to make structural changes for the company.
If I didn't have 12 months preview to nine months or 12 months preview to a cash problem, all of the contracts terms are already out the door.
All of the hiring plans are out the door, all of the project bids and the markup on the projects that's out the door, it's too late to make any changes.
And I didn't, I wasn't okay with that.
I wasn't okay with that level of I'll call it blindness to the future.
Another example, so DSO, I'll pick on that one one more time.
The best accuracy that I've seen for DSO is that they're able to get somewhere around 80 to 85 % of your actual receivables in using DSO.
If we switch to a machine learning based process, it can generally explain about 50 % of what's left over.
And so while it seems like, well, why do you care so much?
Keep it simple, Nate, why are you trying to make it so complicated?
I'm trying to make it complicated because I shouldn't say I'm trying to make it complicated.
I'm trying to take the complexity out of the way with the software.
And that was why I wrote it because I was sick of the complexity.
But the idea is, 5 % error on revenue doesn't sound like that big of a deal until you recognize that that 5 % error in revenue is over 100 % of my cash position.
And so 5 % error is not the issue.
It's well, what's my percentage of cash that is attributed to revenue forecast error, or to cost of goods sold forecast error, or to receivables forecast error.
And so I want to keep it simple.
But to me, it's worth adding complexity when it can explain that much of such a critical piece of the business.
So maybe this is a good way to wrap all this.
So we're in the middle of budget season.
Hopefully people are getting close to being finished with budgets for 25 right now.
But when we're doing budgets and we're arguing with Department and Division heads about their allocations and what they're going to spend next year versus last, it becomes there's the political part of it, and then just the math, and then how cool your model is.
But really what we're doing here, and this is where I see, and this is when I talk to businesses that are trying to really lean into data driven decision making and forward looking CFOs who are trying to be strategists more than just the bean counters, not the always looking backward.
Yes, we're reporting on this.
Yes, we're forecasting.
We're being accurate.
But there's another level of detail if you get this forecasting right, that you really are adding a level of value to the business.
And you talked about this before the show, but if you could elaborate a little more on how long range forecasts can help reveal these structural issues.
Because I think we talked about material cost sensitivity, misalignment and pricing and labor costs.
These are areas where, I mean, this is a very smooth running finance office who has this forward looking.
This is where you are providing value to the company.
You're not getting caught flat footed because you see this early because of the data you have.
So if you could walk through that a little bit, I think that that's maybe a good reminder of this time of year for all of our FVNA folks, too.
Yeah, definitely. Well, and I think it's the reason for the struggle here is always, well, how much time and effort and man hours or man years, depending on your situation, does it take to build a driver -based model?
I think you guys have all heard the term, here's a driver -based model for my business.
It takes a lot. It is really difficult to do and build.
And I have ruled examples from my career where it was a team of 12 people that took three man years and nine calendar months to build a forecast.
It's really hard to build a good one.
And so oftentimes, the decision is, well, it's too complicated to do that.
It's too hard to go do that.
And I think that that is true unless you have a tool that helps make it easier.
And I think that's why I'm so excited about this AI, ML space.
And we're not the only ones that...
Obviously, there's other ways to go about doing this, but that's the advantage.
Now, I'm talking about it because, hey, it's cool, but let's look at the ideal state of if you have a driver -based model for your company, and you're truly doing a 24 -month forecast for the entire business.
What that means is, I would actually have, okay, I can look at my balance sheet in September of 2025.
And if there's something I don't like for that, I have time to fix it.
I have time right now to forecast it.
Now, obviously, the critics are going to say, oh, well, your budget is always wrong anyway, so why even bother?
It's like, well, yeah, it is.
But you can actually quantify how wrong it's going to be.
It's actually possible to say, through sensitivity analysis, and there's a bunch of other methods we're not going to get into, yes, you can have a defense for that.
But the real reason for it is because...
And this is what I started running to.
A 13 -week cashflow forecast is very tactical.
It's very useful from a tactical perspective.
But if I'm seeing a cashflow issue that's 12 or 13 weeks from now, that's hardly enough time to fix it.
What levers do I have available if I only have 13 weeks preview to cash?
I basically can get a line of credit.
I can hold on to some cash for my payables.
But that's not really what I want to do.
If I have a 12 -month or 24 -month preview to my cash, I can start to recognize like, hey, we might be an expense reason or might be a cost -to -get -sold reason.
We're running into a cash issue because of how much we're expecting to grow next year.
It's great that we want to grow, but we've got to switch to 60 -day terms, not 90 days.
You have the ability to change things before they happen.
And I'll use a word picture for this one.
But it's this idea of like, you've probably all heard this, right?
I think it's called upstreaming.
I can't remember the official name for it.
But if you keep seeing people...
There's somebody drowning in the river that's flowing down the river.
You can jump in and save the person.
That's a tactical thing.
But if 7 people came down, yes, you could go save all 7.
But what you actually need to do is go up the stream and figure out who keeps throwing people in the river.
You got to go upstream, further upstream of your business.
And this is one of the key ways to do that.
That's great. And I think that that's really the kind of insight that we want.
And that's what we hope to get out of our models.
And I think you mentioned using AI and machine learning.
I know you guys recently launched your AI -powered forecasting tool.
I guess maybe tell me a little bit about that and what gaps you saw in the existing tools that are out there and how your product addresses those.
All of the examples that I've seen...
So our tool is anybody that comes in with a set of financials, it will build you a forecast from those financials very quickly.
And then we get into, okay, yes, we need to add some and tweak some pieces of the picture.
So every business is different.
So we always have to do a little bit of tweaking or training, as we're calling it.
We do all of this nerdy statistics stuff that we talked about a little bit.
And I apologize if we put you to sleep with that.
But think about it as we're handling all of that on the back end.
The only other examples I've seen...
I've seen a lot of people...
I love that it's happening.
But I've seen, okay, well, if you want to get better forecast accuracy, you've got to learn Python and you've got to go get these libraries and here's some code.
Or you've got to get chat GPT to write the Python for you.
That's another way to do it.
But then you're still learning how to code.
You're still having to learn how to write Python code.
Now, I don't have a problem with anybody doing that.
I love writing code in Python.
Now you're kind of stuck on the straight off of, well, it's going to take me a year to build a machine learning model in Python, the amount of effort it takes.
We make that something that's available today.
So I think that the gap that I'm really trying to solve for is when you get to a certain size, that 1%, 2%, 3%, 4 % forecast accuracy is a big number.
It makes a lot of...
It makes a really big difference.
And there are, with Excel, as much as I love Excel, getting an extra 1 % or 2 % forecast accuracy on a gravity model using Excel is a lot of work.
It is just an incredible amount of work to go do it.
It's fun. I would enjoy it.
Some people, I'm sure, would enjoy it, but you just don't have time to do it.
And so the gap that we saw that we're going and trying to solve for is we want to take what you're already doing.
I'm going to flip the script and say, I'm not trying to get more accuracy.
I'm trying to reduce uncertainty.
So best -in -class forecasting is going to have 10 % to 20 % uncertainty on all of your forecasts.
Some might say, well, 5%.
But if you're going to talk to me about average error, that's a different discussion.
But I'm saying actual total error.
If I can reduce my uncertainty by 50%, I'm going from 10 % error, maximum error, to 5 % maximum error.
That's huge. It's a huge amount of increasing confidence of what we can do as a business.
The problem is the amount of work it takes to go from 85 % to 90 % total error explained to that, getting that extra little bit is an incredible amount of work.
So our goal is to eat the complexity.
Our goal is to make it so that that's all handled.
It's less work than it would take for you to go add accuracy, add capability to the models you already have.
And it's going to do the absolute maximum level of capability that we know how to do.
This is so cool. And I love the automation of this kind of thinking.
And I keep going back to imagine if every FP &A professional had access to this tool and how efficient they would be.
But I also go back to my data robot example because I just...
And it's almost like if you've ever sat in a board meeting and gotten dragged over the coals by that one angry board member who's going to pick apart every model that you do.
And I think about that.
And I think about auditors, not that auditors in your forecast, but the equivalent here where if you're the one...
If I'm the CFO and my team has put all this together, and they don't have the full skill set to explain it, and I have to justify the numbers, I'm like the guy who didn't know what model I was using in data robot and having to justify it.
So I guess this is an important question to me because I travel around and I'm talking to people about using AI and machine learning and finance.
And like you said at the beginning, if you're going to be writing Python or to...
Honestly, I would say I go so far as if you're going to be using chat GPT for anything that you're doing in your business.
I'm not saying you've got to be a developer or a machine learning engineer, but you've got to have a basic understanding of what's happening.
Otherwise, you might as well be shaking up that magic eight ball, asking the question, shake up the eight ball and get the answer from that.
And I don't want to be the guy that's standing in front of the board and saying, yeah, these numbers are good that we put it into the black box and it came out.
So I think... I mean, I guess I would say super cool that tools like this are available, but I almost feel like it's incumbent on the users and this may go against your sales plan here because...
No, this is actually great.
Yeah. So you know where I'm going.
So I'll stop talking and I'll see the floor to you here.
Yeah. No, no, no. I totally agree.
And I had that question of like, okay, well, hey, we use our AR to forecast AR and here's how it does it.
And then I had somebody ask me, well, how does that compare to DSO?
We have to start with the...
I'll call them industry best practice methods and then demonstrate that we've added value beyond the industry best practice methods.
And so one of the pieces of our toolset that we've built in is to help with that.
And it's this idea.
So it's not just being able to get a good forecast, but it's being able to get a good forecast, explain the source of stuff that's happening.
Well, here's the seven drivers that are contributing to AR.
It's actually these invoices from three months ago, this much from 60 days, this much from 30 days.
But then it's also like that board member, we'll talk about that board.
Why didn't you just use DSO?
Great question. DSO got me 83 % accuracy and I added 7 % by using these other methods.
Would you like me to go into how that...
And if he wants to go into a further discussion, a lot of what we're building in the tool is the explainability side of things.
And that's again, I think it was Einstein that said it this way.
He said, if you understand something well enough, you'll be able to explain it simply.
And so it's not just, if you want to go learn it yourself, I love it, please do.
But you have to recognize that you're going to go on this journey of knowing and understanding, but not being able to explain it to a board member.
And then eventually you'll get to...
I'm not drawing the curve correctly, but get to the point where you can explain it simply.
What we're trying to do is we're trying to bypass that entire process.
We're trying to make it as easy as possible for you to...
If you get that board member that says, why didn't you use DSO, you can put a chart up that says, well, DSO explained 80%.
We had this a little bit left over.
We added value. Yeah, I love it.
And this way you can...
You were drawing the curve too.
And I was saying that when you were drawing the curve, I was thinking, and you can avoid the Dunning -Kruger effect as you think you understand it and are trying to explain it, and you're completely wrong.
I already went through that for you all.
So this has been great, and we are coming to the end.
And we've got some questions we like to ask everybody at the end of the show.
And the first one is something that we...
I've gotten some surprising responses.
I got an opera singer here a couple of weeks ago that don't normally picture opera and FP &A going together.
But what is something that people may not know about you?
Yeah, great question.
So I have an interesting one.
When I was at Cummins, I actually left Cummins and took a sabbatical.
So my wife and I sold everything and bought an RV, and we drove around the country for three years.
We actually went to all 48 contiguous states.
And it was awesome.
And so it's not... A lot of people know about it because I get interesting responses.
80 % of the time, people think it's cool.
And 20 % of the time, people look at me like, What's wrong with you?
Right? So it's like...
Why aren't you out there producing?
Yeah. What? What do you mean?
But it was amazing.
I'm so glad we did it.
It's actually when I started this company, because I was on the road driving around in an RV, and I had time to think.
And I was like, Oh, well, let's go do some forecasting.
So that was awesome.
If anyone's thinking about doing it, just ask yourself this question.
10 years from now, am I gonna wish that I had done it?
Or am I gonna wish that I hadn't done it?
Super cool. All right, the next one.
I'm actually very curious to hear what you say on this one, because being the big nerd that you are.
And I know you're in...
You said you love Excel, but you're also...
You're in R, you're in Python and all that.
But what is your favorite Excel function and why?
Yeah. So the one I use all the time is actually index match, which is seems fairly simple, but it's amazing how often sequencing data helps figure things out.
So index match, rank, combined all of those together.
Now, if I had to pick a more sophisticated one, I'd say linest, linear estimate.
But that's the one that broke Excel for me.
And that's, if I'm getting into Linest, then I should be in Python, not in Excel.
Now, again, if you're using that to do regression in Excel, more power to you.
That's awesome. I just found that you could only run so many of those before Excel just says, help.
So you still go to Excel a lot, but I guess you're working a lot more though in Python or I don't know, even doing data science stuff, I still...
Excel is always a great place to start for me, even for making charts or something.
I think of Excel as the original no -code or low -code solution.
And it's the best prototyping out there.
If I have to prototype it, like, okay, hey, we need to add this new algorithm or whatever, I'm probably going to prototype it in Excel and then move over to Python just because it's easier to put it together and visualize and do all of that stuff.
So I use it all the time.
So if I'm doing a prospectus, which is a totally different type of forecasting, I'm not going to try to do a prospectus in Python.
I'm probably going to do a prospectus in Excel.
And then the result of the prospectus we might pull into Python because it's more like an ad hoc and it's...
So I like the flexibility of it.
I love Excel. I probably use Python 60 % of the time in Excel, 40 % of the time, whereas before it was 100 % of the time in Excel.
Are you... And I have not explored this a lot.
I know enough Python to be dangerous and to be a mess.
And Chat GPT has really helped me there.
But are you using Python in Excel?
I haven't talked to a lot of people who are really diving into that yet.
But I have not. So Excel did a...
Or Microsoft did a demo where they did a forecast and they asked Office 365 co -pilot to use Python to do a forecast.
It was a really cool demo.
But I looked at the method that they used and they used Arima.
Arima Max is the methodology that they used.
And again, it gets to that like, do you understand what the shortcomings are of that function?
And so we actually don't use Arima Max, not because it's bad at forecasting, but because it doesn't help you explain how you got there.
And so we have a whole different methodology that we took to solve the downstream problem that that's going to create for you.
So it's a good starting point, but it's not going to help you with the...
Well, now I got to explain how Arima Max works to a board member.
That's never going to happen.
How you can get out the charts and show the residuals and de -seasonalize and de -trend.
That'll be an exciting board meeting.
We've just found that there's other ways to get to the just as accurate or more accurate forecast.
They're more work, but they do a better job of giving you something that's explainable at the end of the day.
So I haven't really dug into it.
I'm super excited that it's there because everybody here has probably written VBA scripts that all the problems that come with that.
I probably need to go play with it, but I just haven't yet.
Yeah, I'm in the same boat.
And with CoPilot, I keep thinking, you'll see the demos and they look great there.
But then when you go play around with it, it's like, well, they're not quite...
I feel like Microsoft's going to get there.
So, well, this has been great.
I guess before we let you go, how can our listeners connect with you and learn more about better forecasting?
Oh yeah, totally. So go to betterforecasting .com is our website.
And there's a way to get a hold of us there or find me on LinkedIn.
Those are the two ways to get a hold of me.
So the website, I think is probably the best way.
There's a form in there that kicks off a thing for our team to pay attention.
Well, Nate, I really appreciate you coming on the show and sharing your insights and wish you the best of luck.
