Variance analysis made useful: build the budget from drivers, then split the budget vs actual gap into volume, mix, price and FX. A worked monthly example.
By Sergey BruhPublished 6 min read
A free course for PMs and marketers: short lessons, real product data and exercises with instant feedback.
Start the free courseVariance analysis compares actual results with the budget and explains the gap, and a budget vs actual review is only useful if it tells you which assumption missed, not just by how much. Two habits make that possible. First, build the plan from drivers (customers, units per customer, price, unit costs) instead of "last year plus 10%". Second, when the month closes, split the gap into volume, mix, price and currency effects that add up exactly to the total. Then "we missed by 55 thousand" becomes "the discount cost 112 thousand, extra volume won back 52, the exchange rate flattered us by 17", and each line has an owner. Below: a one-month variance analysis example from a small online school, from plan to bridge.
A driver-based forecast starts from the operational numbers that cause the money and lets the sheet compute the money. For Klasna, an online tutoring school, one month is a chain:
(0.975 because the payment provider keeps 2.5% of what families pay.) Fixed costs come from drivers too: one support coordinator per started 400 active learners, plus salaries, office and the month's marketing.
September's plan:
The working: and . Sales are UAH. Fixed costs of 1,022,000 (three coordinators 114,000, team 600,000, office and tools 88,000, marketing 220,000) leave a planned operating profit of just 28,847 UAH. September is a thin month.
A total like "last year + 10%" can't be argued with, because nobody can say what has to be true for it. Every driver here has an owner: marketing for learners, pricing for the price, the tutor agreement for the payout.
September closes. The CEO's headline: "Sales 3,709,960 against 3,647,800, we beat the plan by 62,160!" The head of finance looks one line lower: contribution came in at 996,149.75, 54,697.25 below plan. If fixed costs landed on plan, the thin planned profit of 28,847 turned into a loss of about 25,850.
Selling more and earning less is exactly what variance analysis is for. Judge the month on contribution, never on sales alone.
Klasna sells English, maths and programming, in Ukraine (UAH) and to EU families (EUR). Four effects explain the gap:
The bridge closes exactly: . If yours doesn't close, a piece is double-counted or missing.
Separate performance from luck. Volume, mix and price together are −71,822. The weaker hryvnia hid about a quarter of that, and it can reverse next month.
Judge the promo on contribution. The discount cost 106,470 on all 3,640 English lessons, including families who would have come anyway. Even if every one of the 454 extra English lessons came from the promo, they leave at most . Net: −30,141.25 in the best case.
Feed the forecast. Replace September's assumed drivers with actuals (realized price, mix, rates), ask the owners whether the next months still hold, and add a month at the end. That's a rolling forecast: always twelve months ahead.
A budget is the plan you commit to and measure against; a forecast is your current best estimate, updated as months close. The variance between them shows which assumptions moved.
Actual minus budget, line by line, for the same period: Klasna's September contribution variance is 996,149.75 − 1,050,847 = −54,697.25 UAH. On revenue and profit lines a negative number is unfavourable; on cost lines a positive one is. The total tells you how much; the volume, mix, price and FX split tells you why.
The effect of selling a different combination of products at the same total volume. If more sales come from low-margin rows, contribution falls even when the number of units is on plan.
Monthly, as soon as the books close. Quarterly is too late to change anything in a seasonal business.
Module 7 of the free course Finance for Product People builds the whole loop on Klasna's data: a driver-based forecast, budgets people believe without padding, and the full plan vs actual case with the September dataset.
