How to Close a Budget Gap to the Board's EBITDA Target Without a Flat Cut
By Charlie Liu ·
The owners' budgets added up to $9.0M of EBITDA. The board wanted $12.05M. In Episode 1 of Finance Speedrun we closed that gap without a flat cut: one model, AI drafting the options, and people deciding every line.

Budget season, made-up company. An HVAC and plumbing business with about 690 people, 5 entities and 50 budget owners. The owners send in their numbers for FY2027. They add up to $9.0M of EBITDA. The board wants $12.05M.
That's a gap of about $3M, and everyone in finance knows what happens next. This is the problem we picked for the first episode of Finance Speedrun. Here's how we closed it, where AI helped, and where a person had to step in.
How do you close a gap to the board's EBITDA target?
Don't cap everyone by the same percentage. Put every owner's request in one model, let AI draft a few ways to close the gap, and test each one against what actually happened last year. Then owners revise their own lines, executives settle the rest, and the CFO signs one version, with a record of who approved what.
Why does the budget round take so long in spreadsheets?
Because every owner has a file, and every file has versions. With 50 owners across 5 entities, closing the gap means 100+ files over two rounds, and 3 to 4 weeks per round. Most of that time goes to chasing people, not to deciding anything. At the end, there's no clean record of who approved what.
Eugene calls it Excel hell. Someone sends "budget version 2", then a 2.1 an hour later, and the wrong one ends up in front of the board.
None of this is Excel's fault. Investors still get a model in a spreadsheet, not a login. The trouble starts when Excel is also the database, the approval trail and the email system.
What are the options for closing the gap?
There are three. A flat cap is fast but blunt. One shared model fixes the version problem but still leaves the options to be worked out by hand. AI drafting inside that model does the math in hours, and people still make every decision. Most teams are on the first one.
The flat cap. Cut everyone by the same amount. It's quick, and it cuts 21 of the 50 owners no matter what they asked for. The people with good reasons get the same haircut as the padded budgets.
One model, one version. Every hire, project and vendor request, with its share of the gap, sits on one report. Owners work in the same template, filtered to their department or entity, and every change is tracked. Better, but someone still has to work out the options by hand, and that can still take weeks.
AI drafts, people decide. The model's data goes to the AI of your choice (Claude on the episode, but GPT or Copilot would work). AI reads every line, drafts options that hit the target and explains each change. People pick, change and sign.
How do you test the AI's options before trusting them?
Check every option against last year's actuals. If an option only works because a cost runs well below anything you've ever spent, it isn't an option. Then ask the AI to check its own work. It often catches its own mistakes on a second pass, which saves you time before your own review.
The AI came back with three options that all hit the target exactly. Only one held up.
- People first hit the number by pushing overtime down to 6.3%. Overtime ran at 9.3% last year. Not happening.
- Spend first hit it by paying vendors less than last year. Also not happening.
- Balanced got overtime to 6.9%. A stretch.
The recommended budget landed at $12.11M, a little above the target, and kept every technician hire. Each line carried its own reasoning, so anyone could see why a number moved.
That's the point of the demo. AI is fast at the math. It doesn't know that your overtime has never been 6.3%. A person has to look.
Eugene added one habit worth stealing. Once the AI hands you a result, take 30 seconds to ask it to validate that result. On a client project, it once forecast revenue for a property that was closing for the rest of the year. A second prompt caught it.
What does the approval flow look like?
Six steps, in order, all in one place. Owners submit, AI drafts the options, owners revise their own lines, executives settle what's left, the CFO signs, and the board approves. Every decision, comment and sign-off is saved with a name and a date, so nobody has to rebuild the history later.
- Owners submit their budgets.
- AI drafts the options.
- Owners revise their own lines.
- Executives settle what's left.
- The CFO signs.
- The board approves.
For each hire and each project, the call is approve, reject or defer. Once those calls are made, the final budget is plain math that gives the same answer every time. On the episode, the approved budget came in at $12.08M, still above the target.
One thing AI should never do is write the first budget. Each owner's team knows things about next year that aren't in any data set, and the AI has no context for them. That knowledge has to come in first.
What should a CFO check before signing?
Three things. Every option was tested against last year's actuals, not just against the target. Every line that moved has a reason a person agreed with. And the approval record shows who decided what, and when. If any of those is missing, the number isn't ready for the board.
A few questions to ask your team before you sign:
- Which assumptions moved the most from last year, and why?
- Did any option hit the target only by assuming a cost you've never run at?
- Who approved each hire and project, and is that saved somewhere other than an email?
What would a real project add?
This was a made-up company and a build of a few hours. A real one connects to your actual ledger and payroll data, sets up the approval steps for your team, and gets checked line by line until it ties. The same approach works for the forecast, which makes a rolling forecast much easier to run.
Watch the episode
The full walkthrough, with chapters and the transcript, is on the Episode 1 page. You can also get the build guide there, with the input templates and the two prompts we used.
Got a problem for the next one?
Every Wednesday at 11 am ET, we fix one real finance problem live in about 20 minutes. Send us yours. If we pick it, the fix is free, and we never name you or your company on air.
Frequently asked questions
How do you close a gap between the bottom-up budget and the board's target?
Put every owner's request in one model, have AI draft a few ways to hit the target, test each one against last year's actuals, and let owners and executives decide line by line before the CFO signs.
Is a flat percentage cut a good way to hit a budget target?
It's fast, but it cuts owners with good reasons as hard as padded budgets. In Episode 1, a flat cap would have cut 21 of the 50 budget owners no matter what they asked for.
Can AI set the budget for you?
No. AI can draft options and do the math quickly, but it doesn't know what your business can actually deliver. People review every line, and the CFO signs the final version.
Last updated