Top-Down vs Bottom-Up Budgeting and How to Reconcile the Two

Top-down vs bottom-up budgeting is a choice between two forecasts. Top-down sets the totals from a growth target and benchmark ratios; bottom-up adds up what each team says it needs, hire by hire. Each errs in a predictable direction, so the practice that holds up is to build both and budget the gap.

Decades of forecasting research explain why, and a public-sector rulebook shows how to act on it.

Exhibit 1. Two forecasts with opposite blind spots

	Top-down	Bottom-up	Both, reconciled
Starts from	A growth target and spending ratios	Each team’s hires, costs and capacity	Guidelines and an envelope from the top, a proposal from each team
Set by	Founders, finance, the board	Budget owners	Both, line by line
Kind of forecast	An outside view, if built from a reference class	An inside view	Both, with the gap written down
Typical error	Carries the ambition of whoever sets the target	Optimistic about timing, padded where pay or approval depends on it	Smaller, and visible
Blind spot	What the work costs	What the company can afford	Any gap nobody labels

Source: CX Cash synthesis of the research cited below.

Top-down vs bottom-up budgeting: two kinds of forecast

Daniel Kahneman and Dan Lovallo drew the distinction in a 1993 paper in Management Science. An inside-view forecast “draws on knowledge of the specifics of the case, the details of the plan that exists”, while the outside view “is essentially statistical and comparative, and involves no attempt to divine future history at any level of detail” (Kahneman & Lovallo, 1993). Their conclusion was blunt: “when both methods are applied with equal intelligence and skill, the outside view is much more likely to yield a realistic estimate.”

A bottom-up budget is an inside view by construction, since each manager starts from a plan and prices its steps. A top-down budget can be an outside view, but only when its numbers come from a reference class of comparable companies. When they come from the fundraising story, the top-down number is an inside view too: the board’s.

Kahneman’s own illustration dates from the 1970s, when he assembled a team to write a curriculum and textbook on judgment and decision making for Israeli high schools. A year in, he asked each member to write down how long the finished draft would take. The answers ranged from 18 to 30 months.

He then asked the team’s curriculum expert, Seymour Fox, how comparable teams had fared, and heard “40 percent failure and seven to ten years for completion”. The team carried on anyway (“Facing a choice, we gave up rationality rather than the enterprise”), and the book was completed eight years later and never used (Kahneman, McKinsey Quarterly, 2011). Kahneman himself called the expert’s informal survey “surely not up to scientific standards” and the closeness of the match “surely a fluke”. The direction of the miss is what matters, and it is the direction in which the research below finds bottom-up estimates lean.

Bottom-up budgets are optimistic about time and cautious about money

The cleanest measurement of the timing error comes from outside business. Roger Buehler, Dale Griffin and Michael Ross asked honours-thesis students at the University of Waterloo for their best estimate of when they would finish. The 33 who finished within the study had predicted 33.9 days on average and took 55.5, and only 29.7% of the group finished by their own best estimate (Buehler, Griffin & Ross, 1994). Students estimating their own theses are not managers estimating a hiring plan, so the study shows the bias exists rather than its size in a budget; for hiring, there is direct data.

In a startup, the timeline that carries the most cash is the hiring plan, because people are most of the budget. Kruze Consulting’s analysis of its clients’ spending put payroll-related costs, contractors included, at 76% of operating expenses for venture-backed startups, with wages alone at 51% (Kruze Consulting, 2024). A build that starts every new engineer on 1 January ignores how long hiring takes: Gem’s 2025 benchmarks put the average time to hire for engineering roles at 53 days (Gem, 2025). A search opened on 1 January at that pace fills on 23 February, before any notice period (our arithmetic). The late start lowers the year’s spending and delays the output the hire was meant to produce, and unless the revenue plan moves with it, the budget books the saving and keeps the output.

Money runs the other way. Where approval or pay depends on a number, the people submitting it build in room. A standard managerial-accounting textbook presents top-down allocation as a guard against exactly this, “to ensure that expenses are not padded to create budgetary slack” (OpenStax, 2019). In Theresa Libby and Murray Lindsay’s survey of North American budgeting practice, negotiating easier targets by sandbagging occurred occasionally or frequently in 77% of Canadian and 86% of US business units; the figures are self-reported, from samples with response rates of 13.6% and 1.5% (Libby & Lindsay, 2010). Michael Jensen opened a 2001 Harvard Business Review article with “Corporate budgeting is a joke, and everyone knows it”, and blamed budgets tied to pay for “lowballing targets and inflating results” (Jensen, 2001).

So the same document is optimistic about when things happen and cautious about what they cost. Asking the people who do the work still pays on average: a 2016 meta-analysis found the overall link between participative budgeting and performance “positive and significant”, although its link to the performance of the organisation as a whole varied too much across settings to generalise (Derfuss, 2016). A purely bottom-up plan can also lack a shape. Lenny Rachitsky and Nels Gilbreth, drawing on their careers at Airbnb and Eventbrite, describe a plan built by combining each team’s plan into one. It produced “five times as many priorities as there were teams” (First Round Review, 2019).

Top-down targets carry their own optimism

Jensen’s 2003 paper extends the charge to the top of the hierarchy: when budgets drive pay, “both superiors and subordinates lie in the formulation of budgets” (Jensen, 2003). In software companies, plans as a whole run ahead of outcomes. Benchmarkit’s 2026 survey of B2B SaaS companies found a median planned growth rate of 26% for 2026 against 20% achieved in 2025, the fourth year running in which plans outran results, and fewer than 20% of companies formally track actual against planned growth with root-cause analysis (Benchmarkit, 2026). The survey does not say how the plans were built, so it shows optimism in plans generally, targets included.

A target becomes an outside view when it is built from a reference class. Bent Flyvbjerg, who turned Kahneman’s idea into a working method, sets out three steps: identify a reference class of similar past projects, establish the distribution of their outcomes, and place the new project within that distribution (Flyvbjerg, 2006). A SaaS company has two reference classes within reach. One is a benchmark. SaaS Capital’s 2026 survey of more than 1,000 private B2B SaaS companies puts a typical company with $3M to $5M of ARR at 12% of ARR on selling, 8% on marketing, 24% on R&D and 15% on G&A (SaaS Capital, 2026). The other is the company’s own record. Kahneman and Lovallo proposed the ratio of actual to planned spending as “a convenient unit that permits meaningful comparisons across diverse projects”. A company that has tracked it for three or four years owns the reference class that fits it best.

Governments already add the bias back in

The UK Treasury turned this research into a rule. Its 2003 supplementary guidance to the Green Book opens with a finding. “There is a demonstrated, systematic, tendency for project appraisers to be overly optimistic.” Appraisers are told to “make explicit, empirically based adjustments to the estimates of a project’s costs, benefits, and duration”, using uplifts drawn from Mott MacDonald’s 2002 review of large UK public procurement (HM Treasury, 2003). In 2004 the Department for Transport and the Treasury adopted reference class forecasting for large transport projects (Flyvbjerg, 2006).

Exhibit 2. HM Treasury’s recommended optimism-bias uplifts, % added to the estimate

Project type	Works duration, upper	Works duration, lower	Capital cost, upper	Capital cost, lower
Standard buildings	4	1	24	2
Non-standard buildings	39	2	51	4
Standard civil engineering	20	1	44	3
Non-standard civil engineering	25	3	66	6
Equipment and development	54	10	200	10

Source: HM Treasury, Supplementary Green Book Guidance: Optimism Bias (2003), Table 1. The upper figures are the average historic optimism bias at outline business case stage; the guidance says earlier estimates may need larger adjustments.

The category closest to a startup, equipment and development, explicitly covers the “development of software and systems”. At its upper uplifts, a twelve-month plan is appraised at about 18.5 months and a cost estimate at up to three times its bottom-up figure (our arithmetic). A public-procurement table should not be applied unchanged to a seed-stage hiring plan. What transfers is the design: name the bias, measure it against a reference class, and add it to the estimate in the open instead of negotiating it in private.

The reconciliation has a name: the countercurrent method

German management accounting calls the practice the Gegenstromverfahren, literally the countercurrent procedure; English texts call it mixed planning or top-down/bottom-up planning. The Gabler business lexicon defines it as the combination of top-down and bottom-up planning in which the managers responsible for carrying out a level’s plans also draw them up. A top-down pass breaks the plans down to the lowest level of the hierarchy, and a bottom-up return pass corrects the higher-level plans and builds consensus step by step. Gabler’s case for it is a chicken-and-egg problem the method removes: lower-level goals cannot be set without knowing the higher-level ones, and the reverse holds too (our translations; Gabler Wirtschaftslexikon).

The name comes from physics, and the physics shows the shape of the method. When two streams exchange heat while flowing in the same direction, the exchange stalls as they converge: with equal flows, co-current exchange is “only capable of moving half of the property from one flow to the other, no matter how long the exchanger is”. Run the streams in opposite directions and the gradient holds along the whole length, so almost all of the heat can cross (Countercurrent exchange, Wikipedia). A budget run as a single handoff, with a target sent down and the submissions summed and split down the middle, is the co-current case. Exchanging guidelines and proposals at every level keeps the difference visible until it is resolved. Information, unlike heat, is not used up in the exchange, so the parallel holds for the shape of the process and no further.

The best direct evidence comes from a survey of German managers across 127 firms, in which top-down guidelines were associated with less budgetary slack and a bottom-up initial proposal with higher performance (Kramer & Hartmann, 2014). The study is cross-sectional, so these are associations, not proof of cause. The design itself is a century old. In 1922 James O. McKinsey, the University of Chicago accounting professor who founded McKinsey & Company, described a budget in which each department estimates its own activities and a committee of the principal executives revises the estimates (McKinsey, 1922).

The countercurrent method in five steps

1. Send guidelines down. The founders set the growth target, the priorities and the hiring rules. Finance builds a spending envelope for each line from a stated reference class: benchmark ratios, the company’s own plan-versus-actual history, or both.
2. Build up from the teams. Each budget owner proposes hires, start dates and costs from their own plan, without seeing finance’s envelope.
3. Write the gap ledger. For every line, record the difference between build and envelope. Give it one of four labels: ambition (the target needs capacity the envelope does not fund), timing (start dates or ramps faster than history or market data support), padding (contingency held inside a line) or missing (a real cost one side left out). gap = bottom-up build − top-down envelope
4. Adjust timing in the open. Apply an explicit uplift to start dates and revenue ramps, sized from the company’s own history or from external data such as time to hire.
5. Sign one number and keep the other two. Track actual results against the plan, the envelope and the build every month, so that next year’s reference class exists.

Exhibit 3. A gap ledger for a hypothetical SaaS company, $ thousands

A hypothetical B2B SaaS company ends the year at $4.0M of ARR, inside SaaS Capital’s $3M to $5M band, with a board target of 30% growth to $5.2M. Finance sets each envelope at the band’s median ratio applied to average ARR of $4.6M for the year; the teams build their own numbers.

Line	Envelope	Build	Gap	Label	Question for the meeting
Selling (12% of ARR)	552	760	+208	Ambition	Fund the 30% target with two more account executives, or lower the target?
Marketing (8%)	368	420	+52	Padding or missing	Which line items make up the difference?
R&D (24%)	1,104	1,380	+276	Timing (78), unexplained (198)	Will three engineers really start on 1 January, and what is the other 198?
G&A (15%)	690	640	−50	Missing?	Are the audit, insurance and legal costs in the build?
Total (59%)	2,714	3,200	+486

Source: hypothetical company. Envelope ratios: SaaS Capital (2026), median % of ARR for $3M to $5M ARR. Timing: three engineers at a hypothetical $180k loaded cost each, starting 53 days later than planned (Gem, 2025), cost about $78k less in the year. CX Cash calculated the rest.

The build asks for $486k more than the envelope, about 18% over. Argued as one total, that is a negotiation over a single number, and the easiest settlement is the midpoint, which resolves nothing. Split into four labelled lines, it becomes four decisions with different owners: the board decides the ambition gap, the hiring manager defends the start dates, and the line owners explain the remainder.

Wallander called budgeting an unnecessary evil

Svenska Handelsbanken is the standing counterexample. Jan Wallander became its chief executive in 1970, when the bank “was in a major crisis” (Wallander, 1999), and in the early 1970s introduced a model under which, in the words of a Beyond Budgeting Institute case study, “the bank does not operate with centrally controlled targets or budgets” (Beyond Budgeting Institute, 2013). Wallander called his 1994 book on the subject Budgeten – ett onödigt ont: budgeting, an unnecessary evil. In 2003 Jeremy Hope and Robin Fraser wrote in Harvard Business Review that “Budgeting, as most corporations practice it, should be abolished” (Hope & Fraser, 2003).

Wallander, Hope and Fraser are right about incentives, and their core point is worth keeping. When pay depends on the budget, both directions of it lie, as Jensen argued, so the target used to judge people should be kept apart from the forecast used to run the company. Their case against budgeting itself is thinner. Among the North American firms in Libby and Lindsay’s survey that use budgets for control, 94% were not planning to abandon them. The authors read the Handelsbanken story as “much more about promoting and controlling for radically decentralized employees than it is about managing without budgets” (Libby & Lindsay, 2010).

A venture-backed company has one more reason to keep a budget. Its spending runs ahead of its receipts by design, and between two funding rounds the budget is the plan for how long that can last.

Keep three numbers and a ledger through the year

If you are setting next year’s budget

• Build the envelope and the build separately, and keep both after the plan is signed.
• Write the gap ledger, and give every gap a label and an owner.
• Put an explicit timing uplift on new hires and on the revenue they are meant to produce.
• Keep bonuses off the bottom-up number, or expect it to be padded.
• Track actual results against the plan, the envelope and the build every month, and run a budget variance analysis that separates noise from plan error. The annual budget template and variance tracker below has a column for each.

If you approve that budget from the board

• Ask for three numbers and the ledger, not one reconciled total.
• Ask which reference class the growth target came from, and how last year’s plan compared with the result.
• Look at the shape of the hiring plan as well as its size. One finance chief’s rule of thumb, from the OnlyCFO newsletter: “Boards get nervous (rightfully so) when the plan has 70%+ of the hiring in H1.” The same writer keeps two plans, a board plan expected to be hit about 80% of the time and an internal plan closer to 60% (OnlyCFO, 2025).

Two biased forecasts check each other once the gap is written down

In top-down vs bottom-up budgeting, neither method wins on its own, because each fails in a direction the research has measured. The target carries the ambition of whoever set it; the build carries the optimism and the caution of whoever must deliver it. Built independently and reconciled in a written ledger, the two numbers check each other, and the budget meeting moves from whose total is right to which assumption is wrong.

The gap is the agenda.

Sources and notes

• Benchmarkit (2026). 2026 B2B SaaS and AI-Native Performance Benchmarks.
• Buehler, R., Griffin, D. and Ross, M. (1994). Exploring the planning fallacy: why people underestimate their task completion times. Journal of Personality and Social Psychology 67(3), 366–381.
• Beyond Budgeting Institute (2013). Handelsbanken: Consistency at Its Best (case report).
• Derfuss, K. (2016). Reconsidering the participative budgeting–performance relation: a meta-analysis regarding the impact of level of analysis, sample selection, measurement, and industry influences. British Accounting Review 48(1), 17–37.
• Flyvbjerg, B. (2006). From Nobel Prize to project management: getting risks right. Project Management Journal 37(3), 5–15.
• Gabler Wirtschaftslexikon. Gegenstromverfahren (G. Müller-Stewens).
• Gem (2025). 2025 Recruiting Benchmarks Report.
• HM Treasury (2003). Supplementary Green Book Guidance: Optimism Bias.
• Hope, J. and Fraser, R. (2003). Who needs budgets? Harvard Business Review, February.
• Jensen, M. C. (2001). Corporate budgeting is broken—let’s fix it. Harvard Business Review, November.
• Jensen, M. C. (2003). Paying people to lie: the truth about the budgeting process. European Financial Management 9(3), 379–406.
• Kahneman, D. (2011). Beware the “inside view”. McKinsey Quarterly, excerpted from Thinking, Fast and Slow.
• Kahneman, D. and Lovallo, D. (1993). Timid choices and bold forecasts: a cognitive perspective on risk taking. Management Science 39(1), 17–31.
• Kramer, S. and Hartmann, F. (2014). How top-down and bottom-up budgeting affect budget slack and performance through social and economic exchange. Abacus 50(3), 314–340.
• Kruze Consulting (2024). Startup Payroll Costs and Tax Budgeting for New Hires.
• Libby, T. and Lindsay, R. M. (2010). Beyond budgeting or budgeting reconsidered? A survey of North-American budgeting practice. Management Accounting Research 21(1), 56–75.
• McKinsey, J. O. (1922). Budgetary Control. New York: The Ronald Press Company.
• OnlyCFO (2025). Presenting an Annual Plan for Board Approval.
• OpenStax (2019). Principles of Accounting, Volume 2: Managerial Accounting, section 7.1.
• Rachitsky, L. and Gilbreth, N. (2019). The secret to a great planning process: lessons from Airbnb and Eventbrite. First Round Review.
• SaaS Capital (2026). 2026 Spending Benchmarks for Private B2B SaaS Companies.
• Wallander, J. (1994). Budgeten – ett onödigt ont. Stockholm: SNS. And (1999), Budgeting: an unnecessary evil, Scandinavian Journal of Management 15(4), 405–421.
• Wikipedia. Countercurrent exchange.

The figures come from the studies, surveys and guidance listed above, with survey limits stated where each is cited. Dates, uplifts and gaps that cite no source follow from those figures by simple arithmetic, and the company in Exhibit 3 is hypothetical. Translations from German are ours.