zero_based_budgeting v1.0.0

---

name: zero_based_budgeting description: "Build a budget from zero using zero-based budgeting: define decision units, write decision packages at several funding levels, rank them against stated criteria, fund down the list to a cutoff, and govern the result. Covers organisational, team, project, government, and personal variants, plus hybrid and rolling forms. Use when someone asks for zero-based budgeting, a budget built from scratch, justifying every line, decision packages, cutting a budget without across-the-board cuts, or giving every dollar a job."

---

Zero-Based Budgeting

Purpose

Zero-based budgeting (ZBB) builds a budget from zero. Last period's spending carries no authority. Every activity has to justify itself for the new period, in writing, at a stated cost, against current goals.

Peter Pyhrr developed the method at Texas Instruments in the late 1960s and early 1970s. Georgia used it for the state budget under Governor Jimmy Carter, and the US federal government ran it from 1977 until it was formally ended on August 7, 1981. Modern corporate use revived it in the 2010s. An Accenture Strategy study published in 2018, reported by Investopedia, found adoption among the world's 85 largest companies grew 57 percent per year from 2013 through 2017.

ZBB is not a cost-cutting program with extra paperwork. It is a ranking system. The output is an ordered list of funded and unfunded activities, so cuts fall on the lowest-ranked work instead of being spread evenly across everything.

Two different things share the name. Keep them separate:

• Organisational ZBB: decision units, decision packages, ranking, a funding cutoff. This is Pyhrr's method and the main subject here.
• Personal zero-based budget: income minus allocations equals zero, so every unit of money gets an assignment before the month starts. No packages, no committee. Covered in step 7.

Workflow

1. Set the mandate before anyone writes a number

Write down, in one page:

• Scope: which units, categories, or cost lines are in. Which are out, and why.
• The funding envelope: the total available, or the range leadership will consider.
• The goals packages will be ranked against, in priority order. Three to five, no more.
• The calendar: package deadline, ranking session, decision date, appeal window.
• Who ranks, who decides ties, who can override a ranking and on what grounds.

No mandate, no ZBB. Without stated goals, ranking becomes a contest of who writes the best prose.

2. Define decision units

A decision unit is the smallest chunk of work one manager can describe, cost, and defend on its own: a program, a function, a project, a service line. Good units are independent enough that funding one does not silently fund another, and small enough that a weak activity cannot hide inside a strong one.

Rules:

• One accountable owner per unit.
• No unit so large that it bundles a must-have with a nice-to-have. Split it.
• No unit so small that the paperwork costs more than the decision is worth. Aggregate trivial lines into one housekeeping package and say so.
• List every unit, including the ones everyone assumes are untouchable. Untouchable is a ranking outcome, not a starting condition.

3. Write decision packages

One package per activity inside a unit. Each package states, on one to two pages:

• Activity and owner.
• Purpose: what it produces, for whom, and which stated goal it serves.
• Cost, built from drivers (headcount, volume, unit price, licence count), not from last year's total plus a percentage.
• Expected output and how it will be measured.
• Alternative ways to deliver the same result, with their costs. In-house versus vendor, automated versus manual, reduced frequency, shared service, stop entirely.
• Consequence of not funding it, stated factually. What stops, what degrades, what risk appears, and when.
• Funding levels (see step 4).

Template and a filled example: references/decision-package-template.md and references/worked-example.md.

Reject packages that argue from history ("we have always had this"), that price only one way of working, or that hide a cut's consequence. Send them back once, with the specific gap named.

4. Require several funding levels per package

This is the part that makes ranking possible. Each package offers levels, typically:

• Minimum level: the lowest funding that still delivers a recognisable version of the activity, normally below current funding. Federal ZBB required a minimum level of exactly this kind.
• Current-service level: what it costs to keep output as it is now, repriced from drivers.
• Enhancement level: extra funding, the added output it buys, priced separately as its own package so it competes on its own merits.

Each level is a separate line in the ranking. A unit can win its minimum level and lose its enhancement. That granularity is the whole point.

5. Rank all packages on one list

Rank across units, not inside each silo first. A consolidated list is what lets money move from a low-value activity in one department to a high-value one in another.

Score each package level on published criteria. A workable default, used by the ranking tool in bin/:

• Strategic fit: 40 percent. How directly it serves the stated goals.
• Benefit: 35 percent. Measured output or value per period.
• Risk of cutting it: 25 percent. Harm, exposure, or cost that lands if it is not funded.

Scores run 0 to 5. Legally or contractually required packages are flagged mandatory and placed first: they are not optional, and pretending to rank them wastes the session. Everything else competes.

Run it:

bash
python3 bin/rank_packages.py packages.csv --budget 500000

The CSV needs name,unit,cost,mandatory,strategy,benefit,risk_of_cut. The tool sorts, funds down the list, and prints the cutoff and the unallocated remainder. Ranking by hand in a spreadsheet is fine for under about 30 packages. Above that, use the tool or equivalent software so the arithmetic is auditable.

Then hold the ranking session. Managers defend packages, scores get challenged and adjusted in the open, and the final list is signed by the decision maker. Ranking done privately by finance alone produces a list nobody trusts.

6. Fund down to the cutoff, then write the budget

Fund in rank order until the envelope runs out. The first unfunded package is the cutoff line. Everything below it is not funded this period, full stop. Do not shave 10 percent off everything to squeeze two more packages in: that converts a ranking back into incremental budgeting with extra steps.

Build the operating budget from the funded packages only. Each funded package becomes a budget line with its owner, its promised output, and its measure. Unfunded packages go on a published reserve list, in rank order, so mid-year money (a cancellation, a windfall) funds the next package in line instead of going to whoever asks loudest.

7. Personal variant

For a household or solo budget, skip packages and ranking. The method is an assignment pass, done before the period starts:

1. List net income for the month, all sources.
2. List every expense, including irregular ones divided by their cycle (annual insurance divided by 12, for example).
3. Assign amounts category by category: housing, food, transport, debt payments, savings, then discretionary, until income minus assignments equals zero.
4. Savings and debt payoff are assignments, not whatever is left over.
5. Track against category balances during the month, not the bank balance. Move assignments between categories when life changes, and re-zero the plan when you do.

Common failure points: too many categories at the start (begin with a handful), forgetting irregular expenses, and quitting after the first messy month. Expect the first two or three months to be rough.

8. Govern during the year

• Track actual spend and output per funded package, monthly or quarterly.
• Variance rule: an overrun needs the owner's explanation and either a reserve-list trade or a formal re-ranking, never a quiet top-up.
• Mid-year changes go through the same test as the original ranking: what gets defunded to pay for this?
• Keep the packages. Next cycle starts from zero again, but last cycle's packages, scores, and outcome data are the evidence base managers update instead of rewriting from memory.
• Audit a sample of packages after the year ends. Did the funded activity deliver the promised output? Did an unfunded activity cause the predicted harm? Feed both answers into next cycle's scoring credibility.

Choosing the form

Full ZBB on everything, every year, is heavy. Most organisations that sustain it use a lighter form:

• Rolling ZBB: put a slice of the organisation through the full process each year so every unit is zero-based once in a rotation. One documented case, A.T. Still University, rotates its six colleges so each is zero-based once every five years. Common practice reported by vendors is reviewing roughly a third of the organisation per year. Pick a rotation the finance team can actually run.
• Hybrid ZBB: zero-base the discretionary and fast-changing categories (contractors, software, travel, marketing programs, projects) and budget stable, regulated, or contract-locked costs incrementally.
• Periodic ZBB: run the full process every three to five years, or after a merger, a strategy change, or a margin crisis, and budget incrementally in between, using the last ZBB ranking as the reference.
• Pilot first: one department or one cost category, one cycle, then decide. A pilot that works beats a mandate that collapses.

Choose full annual ZBB only when leadership will staff it, the cost base is genuinely in question, and the ranking decisions will actually be enforced.

Output Contract

Deliver, in this order:

1. Mandate summary: scope, envelope, goals in priority order, decision maker.
2. Decision unit list, with owners.
3. Decision packages, one per activity, each with funding levels, alternatives, and consequence of not funding.
4. Ranked list of all package levels with scores, the funding cutoff marked, and mandatory packages flagged.
5. The budget: funded packages with owner, amount, promised output, and measure.
6. Reserve list: unfunded packages in rank order.
7. Governance notes: variance rule, review cadence, and the rotation or hybrid plan for the next cycle.

For the personal variant: income list, category assignments that sum to income exactly, irregular-expense conversions, and the tracking cadence.

Operating Rules

1. Every number comes from a driver or a quote, never from last year's figure plus a percentage. If a manager cannot show the build-up, the package is not ready.
2. History is evidence, not entitlement. Past spend may be cited to prove volume or price. It never justifies continuation by itself.
3. Alternatives are mandatory. A package with only one way of working has not done the analysis. At least one cheaper alternative and the stop option must be priced.
4. Consequences must be factual and checkable. "Morale will suffer" is not a consequence. "Response time goes from 4 hours to 2 business days" is.
5. One consolidated ranking. Separate departmental rankings that are merged by quota destroy the method, because money can no longer move to its best use.
6. The cutoff is real. If leadership funds below-the-line items ad hoc, say plainly that the process is advisory and stop calling it ZBB.
7. Do not game-proof by secrecy. Publish the criteria and weights before packages are written. Managers will optimise for the criteria either way; open criteria make them optimise for the stated goals.
8. Expect gaming anyway: inflated minimum levels, hostage packages (putting a popular activity low to force funding of the whole unit), and benefits stated without measures. Counter them by repricing minimum levels against drivers, ranking package levels separately, and rejecting unmeasured benefit claims.
9. Time is a real cost. Budget the hours the process takes and weigh them against expected savings. If the process costs more than the decisions are worth for a category, move that category to the hybrid or rolling track.
10. Protect long-horizon work explicitly. ZBB scoring favours benefits that show up inside the period. If research, maintenance, or capability building matters, give it a stated goal in the mandate and let it compete under that goal, or ring-fence it and say so. Do not let it lose silently on a short-term benefit score.
11. Never invent figures. Adoption rates, savings percentages, and case results must come from a named source or be cut. Typical first-year savings claims vary widely by source and are often vendor-reported; present them as claims, not promises.
12. Style for all output from this skill: no em dashes. Plain, direct sentences.