Written by The Pragmatic CFO. 15+ years running P&Ls and budget cycles across restaurants and portfolio companies.

An annual budget cycle, done properly, is a forcing function to align capital, headcount, and priorities before the year starts. Most cycles fail because they stretch three months, produce a binder, and get shelved by week two of Q1. The version that ships in three weeks is smaller, more argumentative, and gets used. Below: the five phases, what to cut, the failure mode at each step, the LLM prompt that shortens the work, and why Fixed Charge Coverage belongs inside the model rather than a compliance check after the fact.

Phase 1: Prior-year actuals and the baseline

Every budget starts by cleaning last year. Not summarizing, cleaning. Run actuals net of one-time items, normalize for anything that will not repeat (an insurance recovery, a lease abatement, a one-off legal expense), and rebuild the P&L on a run-rate basis. This baseline is what everything else rolls against.

Cut in a three-week version. Do not normalize every account. Focus on the five to eight lines that move the model: revenue, COGS, labor, occupancy, marketing.

Failure mode. Building the baseline off a Q4 annualization. Q4 is almost never representative. Use a TTM average, then adjust for seasonality.

LLM prompt (baseline scrub).

You are an FP&A analyst. Below is [PRIOR YEAR P&L] by month.
Identify:
1. Non-recurring items (one-time gains, losses, credits, refunds, settlements).
2. Line items whose Q4 differs from the trailing 3-quarter average by more than 15%.
3. Categories that appear "stepped" (a jump to hold flat going forward, not annualize).

Output: a table with columns Category, Reported FY, Adjustment, Normalized FY, Reason. No preamble.

Phase 2: Assumptions gathering

Assumptions are the actual budget. Everything else is arithmetic. Three buckets:

• Revenue drivers. Units, price, mix. Put the driver in the model, not the revenue number.
• Cost drivers. Labor as a percent of sales, food cost by category, occupancy at contractual step-ups, marketing at a defended ratio.
• Capex. Maintenance versus growth, split out. Do not blend them.

Cut. The assumption interview round. Instead of a 45-minute meeting with each department head, send a two-question memo: “What is your top revenue lever, and what is the one cost line you cannot flex?” 80% of the value in 20% of the time.

Failure mode. Departments setting assumptions in isolation. Marketing assumes 12% growth. Ops assumes 4%. Nobody reconciles until the CFO does it during Phase 4, and by then the story is inconsistent.

LLM prompt (assumption stress test).

You are a skeptical operator reading budget assumptions.
Assumptions: [PASTE]. Historical range for each driver: [PASTE 3-year min/median/max].

For each assumption:
1. Flag any that fall outside the 3-year range without a stated reason.
2. Name the single scenario that would break this assumption.
3. Suggest a more defensible number and one sentence of rationale.

Return only the flagged rows.

Phase 3: Bottom-up build versus top-down target

Two numbers get built in parallel. Bottom-up rolls up from department plans and driver assumptions. Top-down comes from the board, the debt covenants, or the equity story. They rarely match on the first pass.

The gap is the actual conversation. If bottom-up says $48M EBITDA and top-down says $54M, the CFO’s job is to name the specific initiatives that close $6M. Not to “sharpen the pencil.” Name the initiatives, sized, with an owner and a date.

Cut. More than two iterations. Bottom-up submission day 8. Reconciliation day 10. Final day 15. If the gap will not close in two rounds, you have a strategy problem, not a budget problem, and pretending otherwise wastes six weeks.

Failure mode. Treating the top-down as negotiable. Sometimes it is (a stretch target from the board). Sometimes it is not (a covenant floor from the bank). Know which before you start.

Phase 4: Iteration and executive review

Two working sessions with the leadership team. Not five. The first presents the reconciled draft and the named gap-closers. The second locks the number and cascades departmental targets. If a third session is needed, someone is stalling. Name it.

Cut. The pre-read. Send a five-slide summary at the meeting, walk it live, take questions. A perfect 40-page pre-read nobody reads adds a week and improves the decision by zero.

Failure mode. Letting the CEO “reserve the right” to revisit a number after sign-off. That is a placeholder, not sign-off. Push back.

LLM prompt (executive summary from raw numbers).

You are writing a one-page executive summary of next year's budget for a board.
Inputs: [PASTE bottom-up P&L, top-down target, gap, named initiatives to close the gap].

Structure:
1. Headline: revenue and EBITDA, growth vs. prior year (one sentence).
2. Three assumptions the plan rests on.
3. The gap and how it closes.
4. Two biggest risks and what triggers them.
5. What you are asking the board to approve.

Max 350 words. No adjectives.

Phase 5: Sign-off and cascading to owners

The budget is not real until every departmental owner has a signed target and a monthly reporting cadence. This phase determines whether the budget shapes behavior in February or ends up in a drawer.

Cut. Nothing. This phase compresses but does not disappear. Days 16 through 21: one 30-minute meeting per department head, agree the target, the leading indicator, the monthly review.

Failure mode. Cascading a P&L target without a leading indicator. “Grow revenue 12%” is not actionable. “Grow qualified pipeline 20% by end of Q1 to support 12% revenue growth” is.

LLM prompt (departmental brief).

You are writing a one-page brief for the head of [DEPARTMENT] on their piece of the budget.
Inputs: [PASTE the departmental slice: revenue lines, cost lines, headcount, capex].

Include:
1. The target with prior-year comp.
2. The three assumptions embedded in that target.
3. The one leading indicator they report monthly.
4. What flex they have and what is fixed.
5. Date of the first monthly review.

Direct, plain English, 300 words max.

Where LLMs actually help

Four places, in order of ROI:

First-draft variance narratives. Every month you explain why actuals differed from budget. The first draft is mechanical. Have an LLM produce it, then edit for what it got wrong.

You are a CFO writing a variance narrative for [MONTH]. Data: [PASTE budget vs. actual by line].
Explain the three largest positive and three largest negative variances.
For each, name the likely operational driver (do not just restate the number).
Format: one paragraph per variance. No preamble.

Stress-testing assumptions. Feed the model your key drivers and ask “what would break this budget.” It surfaces scenarios you missed.

Departmental briefing docs. The Phase 5 prompt saves a full day of cascading work.

Writing the exec summary. The Phase 4 prompt produces a defensible first draft in two minutes.

The Fixed Charge Coverage covenant angle

If the business carries senior debt, the budget must model Fixed Charge Coverage month by month. FCC is typically (EBITDA minus unfunded capex minus cash taxes) divided by (interest plus scheduled principal, plus rent if the covenant includes it). A common minimum is 1.20x, tested quarterly on a trailing-twelve basis.

Anonymized illustrative table (generic mid-market operator, generic senior lender):

Line	Q1	Q2	Q3	Q4	TTM
EBITDA	2.4	3.1	3.4	2.9	11.8
Less: unfunded capex	(0.5)	(0.7)	(0.6)	(0.4)	(2.2)
Less: cash taxes	(0.3)	(0.4)	(0.4)	(0.3)	(1.4)
Numerator	1.6	2.0	2.4	2.2	8.2
Interest	0.9	0.9	0.9	0.9	3.6
Scheduled principal	0.6	0.6	0.6	0.6	2.4
Denominator	1.5	1.5	1.5	1.5	6.0
FCC (min 1.20x)	1.07x	1.33x	1.60x	1.47x	1.37x

Figures in $ millions. Illustrative only. Numbers and lender identity anonymized.

Q1 is 1.07x, below the 1.20x floor. On a quarterly test that is a breach; on a TTM test it clears. This is why the covenant belongs inside the budget model. Check it at the end and you find out in November that Q1 trips. Build it in as a constraint from day one and you catch it during assumptions, then adjust capex or the interest hedge before you commit.

Three rules for debt-financed operators:

1. Model FCC monthly, aggregate to the tested period. Quarterly and TTM both, if the covenant is one and management reporting is the other.
2. Solve backwards. If minimum FCC is 1.20x and the denominator is fixed by the loan schedule, the numerator has a floor. That floor sets the acceptable range for capex and taxes, which sets the growth spend the plan can carry.
3. Keep a 15% cushion. Not because the covenant asks for it, because actual will miss plan and 15% is roughly the average variance in the first year.

The 5 traps

1. Building the budget in the same file as last year’s actuals. You will overwrite something. Start clean, link back to actuals as a reference tab.
2. Confusing bottoms-up detail with accuracy. A model with 1,400 rows is not more accurate than one with 140. It is just harder to change.
3. Skipping the covenant check until after approval. Now you re-do the plan under pressure with the bank watching.
4. Letting the sales team set the revenue number alone. They will hit a ceiling that maps to their bonus, not to what the business can actually do.
5. Not reforecasting after Q1. Every plan is wrong. A live reforecast in April, tied back to original assumptions, keeps the budget useful for the other nine months.

FAQ

How long should a budget cycle take?

Three weeks of concentrated work, spread over four to five calendar weeks for leadership availability. Longer is stalling. Shorter probably skipped the cascade.

Bottom-up or top-down first?

Both, in parallel, from day one. Do one first and then the other, and the second pass just rationalizes the first. Building them simultaneously forces the gap conversation earlier.

Can I use ChatGPT or Claude to help build a budget?

For variance narratives, assumption stress tests, and departmental briefs, yes, and it saves real time. For the actual number-building, no. The model does not have your driver data or your covenant. Use it as an analyst who never gets tired of drafts, not as a replacement for the model.

What if my board wants a number my ops team cannot hit?

Write the gap explicitly. “Bottom-up produces $48M. Board expects $54M. To close the $6M we would need X, Y, and Z, which carry the following risks.” Do not pretend bottom-up is $54M. That is how you lose the team.

How do I make the budget useful after Q1?

Monthly variance review against the original assumptions, not against last month. A rolling reforecast every quarter. And a written retro at the end of Q1 on which Phase 2 assumptions were wrong and why. That retro is what makes next year’s cycle shorter.

What to run next

Once the budget is set, the weekly review is what keeps it honest. The 5-prompt Friday cadence is the shortest way to spot a variance in week three instead of month three. See The 5-Prompt Weekly Financial Review for CFOs.

Building a budget for a restaurant portfolio? The unit-level version, with prime cost, four-wall EBITDA, and per-store capex mechanics, lives at restaurantbottomline.com.

Sources and references

Structure informed by AFP annual budgeting practice and ACG mid-market operator conventions. FCC covenant construction consistent with standard mid-market senior debt agreements. All figures in the illustrative table are constructed for teaching, not drawn from any specific lender or operator.