Every organisation spends money, and every organisation has to decide in advance roughly how much and on what. That decision, written down, is a budget. The two questions that trip people up are what kind of spending each line is, day-to-day running costs (OPEX) or long-lived investments (CAPEX), and what to do when reality, month by month, refuses to match the plan.

The quick version

  • OPEX (operating expenditure) is the day-to-day cost of running the business, salaries, rent, cloud bills, materials. It is expensed in full the year you incur it.
  • CAPEX (capital expenditure) is money spent on long-lived assets, buildings, machines, sometimes major software. It is capitalised on the balance sheet and the cost is spread across the asset's life as depreciation, not booked all at once.
  • Annual planning turns strategy into a year of numbers: expected revenue, OPEX and CAPEX by department. Through the year you compare actuals (what really happened) to the plan, and the gap is a variance.
  • The honest debate: a fixed annual budget is a forecast made once and obeyed for twelve months. Many finance leaders now pair or replace it with rolling forecasts, because the world moves faster than the budget cycle.

OPEX vs CAPEX: the same money, two very different rules

The first thing budgeting forces you to do is classify spending, and the classification changes how the cost lands in your accounts. The cleanest plain-English split comes from the Corporate Finance Institute's CapEx vs OpEx reference: CAPEX is money spent to acquire or improve long-lived fixed assets, so it is capitalised, recorded on the balance sheet as an asset and then written down a slice at a time, year after year, through depreciation. OPEX is the ongoing cost of operating, so it hits the income statement immediately and reduces this year's profit in full.

The textbook example is a photocopier: buying the machine is CAPEX; the toner and paper you feed it are OPEX. Of any sizeable spend, then, ask one question, does this create a lasting asset, or does it keep the lights on this year? The answer determines which budget it belongs in, how it shows up in profit, and often who has to approve it. Most organisations guard CAPEX far more tightly, because it ties up cash now for a return that arrives slowly.

This matters beyond accounting tidiness. Because CAPEX is depreciated rather than expensed, a capital-heavy choice flatters this year's profit and loads cost onto future years; an OPEX choice does the reverse. That is a large part of why "rent it" versus "buy it", cloud subscriptions versus your own data centre, leasing versus owning, is a budgeting decision and not only a technical one.

flowchart TD
  A(["A planned spend"]) --> B{"Creates a long-lived
asset?"}
  B -->|"Yes"| C(["CAPEX
capitalise on balance sheet"])
  C --> D(["Spread cost over years
via depreciation"])
  B -->|"No, keeps the
lights on this year"| E(["OPEX
expense in full now"])
  E --> F(["Hits this year's
profit immediately"])
The fork every line of spending goes through, and why the same cash lands so differently in the accounts. Leaders Loop

An honest limitation. The line is fuzzier than the textbook admits, and the rules vary by jurisdiction and accounting standard. Whether internally developed software, major repairs, or research can be capitalised is genuinely debatable and policed by accounting standards and tax law, so treat the OPEX/CAPEX split as a real decision with real boundaries, and check the specifics with your finance team or auditor rather than guessing. For how the standards themselves decide what counts, the companion piece on accounting standards & revenue recognition is the place to go.

Annual planning, and the actuals that test it

A budget is the financial expression of a plan. The annual cycle, as the Corporate Finance Institute's overview of the budgeting process describes it, typically begins four to six months before the financial year starts: leadership sets the strategy and targets, departments build bottom-up estimates of their revenue and costs, finance consolidates and challenges them, the numbers are negotiated, and a final budget is approved, OPEX and CAPEX, line by line, for twelve months ahead.

Then the year happens. Each month, the accounts produce actuals, what was really earned and spent, and the discipline of budgeting is comparing actuals to plan and explaining the difference. That difference is a variance: spending more than budgeted is an adverse (unfavourable) variance; spending less, or earning more, is favourable. The point is not to file the budget and forget it but to run a monthly rhythm: review the variances, ask which are timing (a cost that will still arrive, just later) versus structural (a permanent miss), and decide what to do, reforecast, cut, or reallocate. A variance you cannot explain is a control problem, not just a number.

A budget is a forecast the organisation agreed to be measured against, not a prophecy, and not a cage.

Here is where the discipline tips into dogma, and where the most interesting research lives. Treat the budget as a fixed contract and you change behaviour for the worse. In Beyond Budgeting (Harvard Business School Press, 2003), Jeremy Hope and Robin Fraser argue that the annual budget, fixed targets, tied to incentives, pushes managers to "make the numbers" rather than make good decisions: to sandbag their targets, hoard slack, spend the remaining budget in December so it isn't cut next year, and defend a plan that the facts have already overtaken. The deeper point in their critique, drawn from work with the Beyond Budgeting Round Table, is that a number set last autumn cannot also be a realistic target, an accurate forecast, and a resource allocation all at once, yet the traditional budget asks it to be all three.

So what do you do instead, or as well?

The practical answer for most leaders is not to abolish the budget but to loosen its grip. Two moves do most of the work. First, separate the budget's jobs: keep a target to aim at, but maintain a rolling forecast, an honest, regularly updated best guess of where the numbers are actually heading, with no reward for gaming it, so the organisation is steered by a current view rather than a stale one. Second, hold some resource back from the annual carve-up so you can fund what emerges mid-year, rather than forcing every decision through a once-a-year gate.

The strongest practitioner case for this comes from Bjarte Bogsnes, who led the implementation of these ideas at Equinor (formerly Statoil) and chairs the Beyond Budgeting movement; his talks (linked below) are the clearest applied account of running a large company without a traditional fixed budget. Even if you keep your annual budget, stop treating its line items as twelve-month promises and start treating the monthly variance review as a steering wheel, adjusting the forecast as you learn, not defending the plan as you drift.

An honest limitation. Beyond Budgeting is a coherent philosophy with persuasive case studies, but it is not a universally proven, controlled result, and most of its evidence is practitioner and case-based rather than randomised. Rolling forecasts cost real effort, and abandoning fixed targets unsettles boards, lenders and incentive plans that were built around them. The defensible position is the modest one: a fixed annual budget is a useful planning artefact and a poor master. Keep it for the planning; refuse to let it stop you thinking.

A worked example

Take a mid-sized firm, call it Harbour Tools, planning next year. (Illustrative figures throughout; this is a teaching example, not real accounts.) The budget splits into OPEX and CAPEX. OPEX: salaries of £4.0m, software and cloud £600k, rent £400k, marketing £500k, call it £5.5m of running cost. CAPEX: a new £900k production machine the operations team has been asking for.

Watch how differently those two lines behave. The £5.5m of OPEX hits next year's profit in full. The £900k machine does not: capitalised and depreciated over, say, an illustrative nine-year life, it lands as roughly £100k of depreciation a year. So the cash leaves now, but the profit-and-loss hit is spread, which is exactly why the machine needs a harder approval and a clearer return than the marketing budget does.

Now run the year. By June, cloud spend is tracking at £780k annualised against a £600k budget, a £180k adverse OPEX variance, because usage grew with the customer base. Marketing, meanwhile, is £120k under, because two campaigns slipped. The naive read is "marketing saved us money." The better read, from the variance review: the cloud overrun is structural (it scales with growth and will persist), while the marketing underspend is timing (the campaigns are merely delayed, the money will be spent). So you reforecast cloud upward, hold the marketing budget rather than claim it as a saving, and decide deliberately whether the extra cloud cost is healthy (growth) or wasteful (no one turned off idle environments).

flowchart LR
  A(["Plan: cloud £600k
marketing £500k"]) --> B(["Mid-year actuals:
cloud £780k, mktg £380k"])
  B --> C{"Explain each
variance"}
  C -->|"Cloud +£180k:
scales with growth"| D(["Structural →
reforecast upward"])
  C -->|"Marketing −£120k:
campaigns delayed"| E(["Timing →
hold, don't bank it"])
The same review, two opposite conclusions, variance analysis is about why, not just how much. Leaders Loop

That is the whole craft in miniature: classify the spend correctly, then let the actuals argue with the plan every month and act on what the argument reveals. The budget that never changes all year is not disciplined, it is ignored.

Frequently asked questions

What is the simplest way to tell OPEX from CAPEX?

Ask whether the spend buys a long-lived asset or just runs the business this year. A delivery van is CAPEX (it lasts years and is depreciated); the fuel and the driver's wages are OPEX. The classic line: buying the photocopier is CAPEX, the paper and toner are OPEX. Where the boundary is genuinely unclear, software you build, a major refurbishment, let the accounting standards and your finance team decide, because the rules are specific and vary by jurisdiction.

What is a variance, and which ones should I worry about?

A variance is the gap between budget and actual. Worry less about the size and more about the cause. A favourable variance from a delayed cost is not a saving, the money is still coming. An adverse variance that scales with growth may be healthy. The ones to chase are the ones nobody can explain, and the structural misses that will repeat every month unless you act.

Is the annual budget really worth the effort?

The planning is worth it; the rigidity often is not. Building the budget forces useful conversations about priorities, trade-offs and what each team is actually for. The failure is treating last autumn's numbers as a binding contract all year. The widely adopted fix is to keep the annual plan but add a rolling forecast you update as you learn, so decisions are made on a current view rather than a stale one.

What's the difference between a budget and a forecast?

A budget is the plan you commit to and are measured against; a forecast is your best current estimate of what will actually happen. Early in the year they match. As reality diverges, the budget stays fixed (it is the yardstick) while the forecast moves to track the truth. Confusing the two, quietly editing the budget to hide a miss, destroys the variance signal that makes budgeting useful.

Why do teams spend their whole budget at year end?

Because of the incentive a fixed budget creates: an unspent line is often read as evidence the budget was too big and gets cut next year, so teams spend to protect their allocation. Hope and Fraser identify exactly this behaviour as a symptom of the annual performance trap. The cure is structural, stop punishing underspend, and judge teams on outcomes rather than on hitting a spend number.

Related in the Toolkit

Budgeting sits on top of the accounts themselves, the OPEX and CAPEX splits only make sense once you can read a profit-and-loss and a balance sheet, and it runs straight into the discipline that turns variances into decisions, FP&A and variance analysis.

Where to go next