How to close the gap between a top-down and bottom-up budget
Episode 2· 27 min· Charlie Liu and Eugene ChengBudgeting · Top-down vs bottom-up
Chapters click to jump
From this episode
Rebuild it on your own numbers
The build guide is a short PDF of what we built on air. It covers:
- The problem: a 10.1% margin promised against 7.0% submitted, and the four kinds of gap
- Three templates: top-down targets, the service line budget, and the alignment report with the CFO's decisions
- Two Claude prompts, ready to paste: one reviews top-down against bottom-up, one writes the board brief
- What they return: the answers from the live demo, so you can check your own run
Free. One email with the PDF, no newsletter unless you ask.
The problem
A made-up outpatient health care group (about $56M revenue, five service lines, one board meeting). Senior management's FY2027 budget promises a 10.1% EBITDA margin. The service line managers send back 7.0%. Some of the gap is real, some is plugs and sandbags, and some is the target itself. In spreadsheets, sorting that out takes weeks.
- Promised
- 10.1%
- Submitted
- 7.0%
- Gap
- $1.76M
How we fixed it
- 1Top-down targets and the service lines' bottom-up budgets side by side in one model, with the gap on every line.
- 2An AI referee reads visits per clinician, denial rates, wages and two years of history, and sorts each gap: plug, sandbag, real gap or top-down optimism.
- 3Two plugs worth $1.0M come out, Urgent Care's cushion goes back in and the wage assumption is fixed: from 10.1% promised to 8.5% the CFO can stand behind.
- 4The CFO accepts, holds to target or returns each line, and every decision is saved with the budget.
Want this on your own numbers?
We build fixes like this on real data for finance teams every week. Talk to our team about yours. Not ready for a call? Send us your problem and it could be the next episode.
Transcript
Word for word, split by chapter. Lightly edited for reading.
Eugene Cheng and Charlie Liu, live on LinkedIn. Brightwell Alliance is made up, and every number in this episode is generated.
0:00 Welcome and Remembrance Day
Hello everyone. All right. Charlie, are you good to go? Yeah. Hi everyone. Terribly sorry for the, for the technical difficulties. Some of us switched to a Mac and, we didn't really test this, but, probably just starting a little bit late. We do want to start with a minute of silence for everyone here. We want to acknowledge that today is Remembrance Day here in Canada. I know we started a little bit late, but, want to just give a minute of silence for everyone.
And immediately after that we'll get started. Everyone. Thank you everyone. For the 11th hour of 11th day of the 11th month, we pause to remember and honor those who served and sacrifice for the freedom that we enjoy today. And, for everyone globally joining us today, thank you for sharing this moment with us. Thank you. Now we'll get into the speed run finance speed run by MCC, our second episode. And today we're pivoting a little bit with Eugene actually taking the wheel and leading the session.
I'll be around. And, today is really, I'm just here just for, for my smile. Perfect. Thank you, Charlie.
1:21 Top-down 10% vs bottom-up 7%: which number goes to the board?
And yes, as Charlie mentioned today is Remembrance Day. So we want to take this moment to remember everyone that has fallen and sacrifice for the lives that we have now. So with that said, let's get started on what we talk, what we want to talk today, Charlie, you've kind of seen this in lot in your implementation and your time as fractional CFO and helping other organizations. We generally have this mentality or this, project where we start off with working with the board.
We work with all the, all the fine, all the senior leaders. They come up with a what we call the top down planet, where they look at the environment and look at the industry. And say, Hey, you want, this is where we want our business to go in the next two years. So they come up with what we call top level planning, but that doesn't necessarily always the same as what the, what the scene, what the managers at frontline workers are seeing as well, too.
So today I actually want to talk about one of the, our sample companies. We call them, the Brightwell Alliance. They specialize in rehab health, home care, behavioral health, mental health, and some of the primary care as well, too, with about 56 million in revenue, five main lines of business, and then one board meeting that they need to go to. And so with that said, the senior manager, they started doing some of the top-down planning. They came up with about 10% of, of, of EBITDA margin.
And then as the, as we go through the planning process, the bottoms up managers, they also did the same exercise, but they did it bottoms up. So they would get a lot of detailed planning. And for them, they came up with just 7%. So just a gap between, between the two numbers. The question now here is which one is the real number? Which one is the one that goes to the board? Which one is, is someone that we should trust. And the challenge here is that no one really knows what the number is.
And the uncomfortable part is that neither of these numbers are exactly correct. And for a couple of form forming reasons, right?
3:25 Budget games: plugs, sandbags, real gaps, top-down optimism
So we see this quite a bit where a number of managers, we just plug a number to say, you know what, we don't know what that is. Let's just put a number out of thin air, nothing behind it to basically close, close the gap. Or you actually have instances where it manages to actually want to send back the numbers, set the bar a bit lower on purpose. So it's easier to beat. That way that, you know what we can say, Hey, we beat our, we beat our budget by 10 by 10%.
Where's my, where's my, where's my bonus or where's my compensation. This is especially true when you have more compensation or results based reward system where they compensate you on how well you did compared to your target, but then also, you know, the real gaps where the business just can't do what the target assumes. So say for example, we want to serve that many patients or have them in the thesis, the market may simply doesn't have that demand or you may not have the resource to do that.
And then lastly, as we've seen a lot, some just top, top, top-down optimism to say, the business would like to go in this direction, but the market doesn't serve it. Say for example, we want to say, you know, we're going to assume that we just goes up by 2%. But realistically, the wages in the market has gone up by more than four and a half percent due to recent inflation or whatnot. And to get to these numbers, usually it takes weeks and weeks where the CFO will work with the team and also with rest of finance team, go through what we call spreadsheet loop help.
Everyone goes in, updates their budget, sends up a new updates version of the numbers. And at the end of the day, whichever numbers survive or which one numbers get the loudest voice stays without any backings and Swiss basics, basically wasting a lot of people's time just to get this number. Is there something that you've seen?
5:11 Why sandbagging usually isn't a bad actor
Yeah, I've seen is over and over again. And I think the, the conversation is a little bit different from kind of setting up tools and systems about how to actually create better budgets. And I do want to highlight sometimes the term like sandbox sandbag sounds negative. And I think all of these can have negative connotations to it, but I don't think generally it's not a bad actor. It sometimes is competing priorities or a lack of data clarity, right?
So for example, sandbag, if you take that one, you may have a, a performance in the last, year or because of specific reasons that you cannot control, that's maybe temporary. That's not doing well. And then you just put it slightly above that when it's very realistic for things to come bit conditions to bounce back up and you can absolutely overachieve that. I think every single one of these can be, yes, sometimes you do have bad actors, but I really do think a lot of it just comes down to actual analysis and trying to actually build the budget in a correct way.
That's useful for everyone and having that alignment. Cause they, a good budget really sets attainable, strong target that drives everyone. And often because of what this process looks like and everyone, especially for the younger folks that's watching us that it's kind of early in their career. Just remember like the folks that's working with you, your head of sales, head of marketing, head of HR, they have their day to day jobs as well while trying to build us with you, right?
So it's competing priorities. It's lack of just, analytics understanding where things are. And because of that you often end with, some of these different gaps that can lead us maybe to a number that we have, or maybe not, but even if we get a number we have, it might not be a good plan to set the, ambitious goals that we're trying to achieve with a, with a business. Yup. A hundred percent. And we actually working with a customer right now
7:16 Goal seek budgeting: shoot and hope
Where the senior leadership, they will actually set a specific KPI. And then when we, when the bottoms up approach, when bottoms up approach doesn't match number, what they do is they do something for simplicity, for lack of vision. They just increase everything that they have by that percentage until we hit that number. So it's almost like a very simple go seek. Now we're hitting that management KPI now in our, in our budget. But what does that mean?
Does that really translate into actionable insight insights? What does action translate into real results? It's TBD. So there's really almost like a shoot and hope that you will hit that number. And then let's see what happens after. So it doesn't really translate into real anything that can be tangible. And I think the goal is to create a plan that works for everyone where you want to make the investors, you want to make the board happy. You want to have these targets, but we can't hit them.
What's the point of these ambitious kind of goals that we have. I do understand some of his expense and budget controls as well. And Eugene will go through some of these best practices to help you improve that budget process. I know a lot of you are in that right now. So this could be something that could be immediately used for helpful.
8:30 Excel versions vs one planning model
And so for a lot of you that are in this right now, you've probably experiencing this right now firsthand where you have Excel, where your plan lives, a couple of workbooks per service line or business unit, a lot of email back and forth, a lot of version 1.0, 2.5, final version 2.6. And then you compile a lot of these by hand, and then compile a lot of these by hand. And then to prove your evidence or do your analysis is whatever time you have left behind.
And then who finally makes the final decisions, whoever argues the longest, whoever argues the last loudest. And the time it takes for you to compile these can be weeks or sometimes even months before your budget comes together. So this is where a tool can help. And it's one of the tools where you can actually create a top-down planning and a bottom-up timeline at the same time and compare side by side. So you can see the line, the gaps line by line at any levels.
And those people draw the details and explain how all these are calculated, what these comparisons come from, and who put these numbers in. So actually you have the accountability and audibility in one single system. And then just going through a system like this can take days to actually get all of these because everything's now automated and consolidated together. Now we want to kick it up a notch where while you have this model, how can
9:49 AI makes the suggestion, a human makes the call
We then layer AI on top? And then now you actually use AI to do the analysis. One of the common themes that we've been seeing a lot over and over again is that there's still a bit of hesitancy on using AI. And this is why where AI comes in as almost like a suggestion to say, AI is not here to tell you what your budget is based on information it has. As smart as it is, it is still not human. It still does have all the context of the business. So what AI does in here is that it helps you go through all the gaps, sort through which one is, and help you classify it and make recommendations on how we can close the gap.
And we'll see this shortly, but at the end of the day, AI is just there as a suggestion. It still requires a human to finally approve and make sense of those numbers and basically present a board tag that is, while it is assisted by AI, the final decision is to rest in the human hands.
10:41 No planning platform? Start with a database
But if you don't have a platform, you can often mimic this with creating a quick database, saving the data from the different budgets, different versions in a database as well. And that's how you're going to be able to leverage AI better as well. Yes. You can push Excel files into AI, but if you have 30 different versions, 30 different Excel balls, probably more than 30, because you have different schedules for different departments. Trust me, your whatever you're using is not going to have the context when a big enough to do any of that correctly.
And especially if you don't have the semantic layers definitions laid out and with that many Excel files, you're going to have a lot of hallucination. Yeah. Cool. And in that case, let's switch gears and let's take a look at
11:26 Demo: top-down targets
What this could look like. In here, we have what we call a task contributor field, where your users can come in and basically see what kind of tasks they have been assigned to them. In here, we have three different types of tasks, top-down planning, service level planning, the basic bottoms up, and then seeing what the alignment report looks like. Let's start with very simple top-down planning. Top-down planning, your users be able to actually, for us as senior level managers or senior level executive, you can see very quickly how you can update or model your business depending on the entity, department, and also the scenario you want to look at.
So in here, we're looking at the base scenario, but we can create as many scenarios as you need to. So you can do the best case, worst case, what if, what if one, what if two. And one of the beauty of this is that now creating a new scenario is simply a click of a button as opposed to weeks and weeks of analysis. And now you can actually get your answers right in the board meeting without waiting for weeks to do the analysis. In here, we have our 26 forecasts.
First, simply define what the growth is on an annual basis. Do you see what your 27, 28, and 29 looks like? In this case, we've decided to, for simplicity, model your growth on an annual basis. But we can also put it at a monthly level as well, too. It's just so that you get that seasonality. So now we have, we have a web of management field that basically to say for each line, each line item, the salaries, revenue, how many patients are coming in.
You can model that very quickly, very easily on a very high level. This is what the senior management manager sees. And this is where the first top-down planning looks like. Now let's switch gears a little bit and see, now that we've seen the top -down level, let's look at it from the bottom-up, from the boots on the ground managers. What are they, how would they model this?
13:36 Demo: the service line budget
So now let's also open it up from the bottom-up approach. I won't bore you with the details here, but for example, in here, we have the ability to choose which department again, what scenario to look at. So now I'm going to look at the base case again. But now from a bottom-ups, bottoms-up perspective. In here, very simple model where we model the number of patient visits based on patients that are coming in. How many visits visits per that type of treatment?
What is the rate for, what is the rate for, for those treatments? And also what is the revenue that will be generated as a results of those treatments and what are the FTE staff or what are the costs that need to support that model, because you can see this is much more detailed, much more bottoms-up, much more demand-driven and also driven off by a lot of operational metrics and TPRs as well, too. So now we have the top-down and the bottoms-up. Now, now that we have these two, let's combine and see what does those two, those two look like.
This way we open up our last piece of our final report
14:45 Demo: the alignment report
Here, which is an alignment report. Let's look at 2027. Let's look at our base case. In here very quickly, we can actually see for each business unit, which ones are the ones that are actually aligned with the senior leadership expectations. So we have, let's say outpatient rehab, pretty aligned with what the top -down approach looks like. And then we also have some other ones where you can see that there's obviously differences in how the senior business is seeing the business versus how the boots on the ground staff is looking at the business with consistently negative variance between how they are seeing the business as well, too, with some specific outliers.
And now we can see very quickly which ones are the business that are deviating from the senior leadership and which ones are the ones that are aligned with senior leadership's vision. And then I'll get into this in a bit, but we can actually now also, the second piece of this is how we use AI to differentiate it or group these differences to say which ones are the ones where we historically we've seen a gap and management plug in the numbers, or which ones are the ones where we actually see the boots on the ground to be consistently sandbagging the numbers while to consistently set the bar lower, just so that it becomes easier to obtain the targets.
But these are some of the things that we can do within the system. And you can make a lot of these updates very quickly without the need to manually aggregate and recompile these numbers every time a number has changed.
16:34 Demo: the AI referee in Claude
So with that said, let's take a look at now that we've seen these numbers, let's do some analysis on it. And this is where we've managed to mention we can use some of the AI tools to help us understand where these differences are coming from. So with that, I'm going to open up my instance of Claude. And here our tool of choice is Claude. Doesn't mean that this is the only one that you'll be using. We generally like this one because of how our tech stack looks like.
You can do this with chatgpt, gemini, any type of tool that is at your disposal. And in here, we are actually going to see this as we created what we call a referee. Basically what a referee does is that this referee will then basically come in from a fairly neutral stance to look at all the budget that we've submitted, comparing historical actuals, comparing historical budget to see what are some of the trends that we have. So now I can say, hey, Brightwell referee, which is the company that we have, run the data, run the budget data for 2027.
So very quickly, it would then start to think about, based on all the patterns that we've seen, which ones are the common communities, what are some of the themes that we've been seeing in the manage this input versus what it's actually been doing in their actuals.
17:58 The AI read: plugs, a sandbag, a real gap
Cool. So back to where we are now. So we have ran the scale, took about maybe a half a minute to an hour to start to a minute to basically does all this full analysis. I won't bore you with the full analysis here, but then it also, what this does is actually create a very quick dashboard to say, these are the service line or business units that sees that we're see a pattern of plugging the numbers, and then there's actually a real gap in some of the behavior, behavioral health business, where we actually do see a gap between what the business is seeing and what the company, what the organization can actually offer.
Some of them, a little bit of sandbagging, some primary care, which is very aligned with what senior leadership is doing and what the boots on the ground, staff is seeing, but then we also on the, on a company-wide, we also see some top-down optimism as well too. This is where I mentioned where we just within expecting 2% increase, but from the managers where everyone's expecting more than four and a half percent increase. So this contributes to grab a gap of about 800,000.
So very quickly, you can now then see, how can we go from a top-down approach, what the top-down approach looks like, what the bottoms up looks like, and how can we sort of bridge the gap? What are some of those plus and minus that contributes to the, to the gap?
19:18 Why reading the drivers matters
Eugene, if you screw up a little bit, I just want to really highlight how powerful this was for everyone, right? So the fact that, and again, we're doing this with Claude, I assure you, it's going to work with other kind of, frontier LM tools as well as whether you're a co-pilot, GPT, Gemini, et cetera. It's going through every scenario. It's not just looking at the numbers, it's also looking at the drivers and the decisions and comments that each one of the budget owners are making is looking at by service line for you could be departments.
It could be entities, could be locations. It's comparing against kind of, industry benchmarks, inflation growth rates. It's also actually looking at historical data so that, for example, the sandbagging one, right? Highlights that that budget is clearly under the actual performance this year. And this is where if you're doing that, if you don't do this level of analysis, it gets really dangerous. Cause right. We were talking about earlier, right? Like with that target that the, the CFO, the board wants, what happens if you have a department where the number is exactly, that percentage or the EBIT that you're looking for, but when they can actually be doing better or worse, and you can't treat each of them kind of with the same kind of metric or divide or cut, it's a combination analysis for each of these departments, these each of these different budget owners, and then you collectively work towards that target that you didn't have, which then you can highlight in the second kind of page.
Right. It's both up and down by service line, right? It's not all of them hitting that one kind of metric or balance that we're trying to get to. It's collectively, we have to get to that, specific EBIT that we're called margin target. Yeah, you're absolutely right. And you know, the, the part that really amazes me and that what excites me about the future here is you're not looking at just simple analysis of budget versus actuals, you actually have to, with the help of your AI model, you can actually have, be able to run your budget versus analysis across all the data that you've ever had.
So historically, how have each team, planned this out? Have they been consistently under actuals? Have they been consistently over actuals? Have they been consistently sandbagging? To do that type of analysis manually, it's nearly impossible due to the size of data and the amount of giant cells you need to do the analysis and not to mention how painful it would be for you to write all those formulas on giant Excel files, but with the help of AI models, you can feed all of them and actually even help you understand where are your gaps in your planning methods, there are some, there's a company I'm working with where they have over 4,000 hotels that we're planning for, we actually use the AI model to help them not only to the planning, but how to improve the model, what is the gap that they have in their specific brand, is there specific regions or different dimensions that we are underutilizing, how can we, use those extra dimensions and do a better plan?
And this is all done by running for every single combination and seeing what the gaps are that simply cannot be done by any human or manual Excel files. Well, at least in the speed that you're hoping for.
22:42 You need a semantic layer
Well possible, but I also want to highlight all this requires a good semantic layer that has good definitions, good control controls as well. Cause what I'm noticing in the industry is a lot of folks may have, individuals are really strong. But oftentimes those individuals are really strong have their documentation, their reference or MD files, but it's local, right? The whole idea is like, how do you build a semantic layer? So everyone is using the same source of truth, not just the financial or operational data, but also the definitions, right?
How are certain KPIs calculated? What does that definition look like? What dates do we use your charter account, et cetera, et cetera. And that allows us to get to similar, similar figures for everyone.
23:32 Decisions: accept, hold to target, return to manager
Yup. So one before specific conscious of time, make sure that we end on a powerful note. So let's take a look at one, one, one, some of the last pieces that we want to put in. So as we saw, not only can the data and the, AI model generate these analysis, they can actually feed it back into a platform of choice to highlight what these, what, what those gaps are. And finally, you can then actually even ask your AI model to come up with recommendations to say, these are, these are, these are the different five business units that we have.
What is the AI's evidence and what is the AI recommending? So now as a, as a, as a CFO or as someone in leadership positions or each of our managers, I can then say, okay, here's what the AI is suggesting. Do I want to, do I want to go ahead and do this or not? Do you want to say, okay, for these ones, we'll accept because we understand what the plug is. For some of these, I say, you know what? We actually want to return to managers and some of them actually want to hold the target.
So now as the finance lead or as the senior leadership, we are not letting AI just basically do the budget for us. It's giving us recommendations, but we can now put in the workflow and the governance in place to hold people accountable while still giving them the eight tools of recommendations of AI to either accept it or not. And we've seen this a lot in account recognitions or consolidations where AI simply gives you the recommendation of what, what it should look like, but still requires the final human touch or human approval to give it a comment and make a decision.
And then from there, once all the decisions are made up, you can now then finally feed it back into the model to have a finalized, finally consolidate model that you should take, usually take weeks or months to compile, put a governance and put a semantic layer on it for the AI to control and also give recommendations. So this is where we sort of see the future heading. And, you know, probably just to wrap this up is at the end of the day, we are trying to condense or compare two numbers.
We're trying to set expectations ourselves to say, to control the expectation of what the senior leadership is seeing in the market, the industry trends, compare it to finally what people are seeing on the boots on the ground, what's really happening in their locations in the clinics, merge the two, just so that the team comes with a number that is not too lofty, but still grounded on based on real facts and supported by AI analysis. So with that said, Charlie, I want to pass it back to you.
Anything else you want to add before we wrap this up?
26:10 Wrap-up: you sign off, not the AI
No, I think for the audience, we dropped a lot of information. This is our second one to run in this. My gut actually feels like we're trying to cover too much, but if you, if you don't agree with that, we just grew that love to leave a, see some comments and just feedback from everyone. Our goal is to provide these to be educational, highlight different solutions, questions, problems that you're solving and be able to show you that solution and share the solution with you as well.
But ultimately I do want to highlight, AI is going to make our jobs easier, but at the end of the day, you are the one signing off on that data on those decisions, which is why you need to have good data structure. You need to have what we keep seeing that semantic layer done properly. You need to have correct data. You'd need to eliminate any sort of, chances of hallucination, but also the final figures, the final decisions. That's you, my friend, you are making those decisions.
That's nine, the AI hands. That's why at the end of the day, it's going to do a lot more analysis for you. Even an analysis that you never had before, but it's ultimately you making those decisions, doing the fun work. I think it's through the strategy of decisions and then creating a good budget that is both realistic and ambitious to achieve targets from the, the board of the execs, but also attainable so that we all win together. Thank you. Oh, thank you everyone.
And I hope to, and I hope to see you next week. Again, if there are, if you're watching this, if any new challenges that you would like to see us speed run, please put it in the comments and that will incorporate it into a future episode. Thank you so much for watching and talk to you all soon.
Next live: Episode 4, Wednesday 10:45 am ET
Got a problem like this one? Send it in. If we pick it, the fix is free.

