Syntha Lab RU

Forecast and budget

How much you will sell in the coming periods and how plan meets actual.

both Planning and buying

Open interactively →

About the topic

A plan is not a prediction but an agreement about what we consider likely and what we will do if it turns out otherwise. The topic “Forecast and budget” combines three questions: how much will we sell, is there enough cash and profit for it, and where did the plan diverge from the actual. A retailer splits the plan by stores accounting for new points and closures, a brand reconciles the buyers' order book with the season target. It is read by the owner, the finance director, the commercial director and the analyst, and the results affect buying, hiring and promises to investors and the bank.

Start with the forecast. A forecast has an error and it is useful to measure it. WAPE is the sum of absolute deviations divided by the sum of actuals. Illustrative example: over four months actuals are 100, 120, 90 and RUB 110 million, forecast 105, 110, 95 and 100. The deviations are 5, 10, 5, 10, total 30; sum of actuals 420; WAPE = 30 / 420 = about 7.1%. Now methods can be compared honestly: simple average, smoothing, a method with seasonality. The winner is the one with lower error on held-out data, not on the data it was trained on. A seasonal forecast from one year of history cannot be built: it is unclear whether it is seasonality or chance.

A budget turns a forecast into money by month: revenue, gross profit, costs, EBITDA (profit before interest, taxes and depreciation). It is needed to see in which months the business goes into the red and to decide in advance how to cover it: a credit line, shifting payments, cutting the buy. It is important to separate fixed costs (rent, salaries) from variable ones (commissions, logistics): when sales fall by 10%, profit falls by more than 10% because fixed costs remain.

A brand has its own part: the order book. Season target RUB 80 million; confirmed orders of 48 million, negotiations for 30 million with a probability of 40%. Expected 48 + 0.4 x 30 = 60 million, that is 75% of target. If one buyer gives 24 million (30% of plan), concentration risk is high. Negotiations are not confirmed orders. For retail the separate difficulty is new stores: you cannot plan mature revenue from the first month. A typical curve: 50%, 70%, 85%, 100% over four months.

After the period, plan and actual are compared and deviations explained: price, volume, mix, markdowns, costs. The conclusion goes into the next plan. Traps: a seasonal forecast on short history, ignoring promotions and price changes, mature revenue for a new store, and trusting negotiations as orders.

Key ideas

Forecast error: WAPE

An average percentage over points misleads on small values. Two categories: one has actual RUB 2 million, forecast 4 (error 100%), the other actual 98, forecast 96 (about 2%). The average of the percentages is 51%, while WAPE = (2 + 2) / 100 = 4%. A small category must not dictate the verdict on the method, so the error is weighted by actuals.

Seasonality is needed but not from one year

Seasonal index = average sales of the month / average sales of all months. If December averages 1.4 and base revenue is RUB 10 million, we expect 14 million. But from one year you cannot tell season from a one-off event (a store opening, a promotion): you need at least two to three years.

Fixed and variable costs

The reverse question is useful too: how far can sales fall. Revenue RUB 80 million, variable costs 55%, fixed 24: profit 80 x 0.45 − 24 = 12 million. Zero comes at revenue of 24 / 0.45 = about 53.3 million, that is a fall of roughly a third. The larger the share of fixed costs, the closer this limit, and the more careful you must be with new leases.

Order book and probabilities

Instead of one figure, three scenarios are calculated. Target RUB 50 million, confirmed 28, in negotiation 22. Confirmed only: 56% of target. At a 25% closing rate the expectation is 28 + 5.5 = 33.5 million (67%), at 40% already 36.8 (74%). The 3.3 million gap between scenarios shows what the argument about probability is worth, and that buying should be sized to the lower one, not the upper.

A new store reaching maturity

Mature revenue is RUB 9.6 million a month. On a 50/70/85/100% curve the first four months give 4.8; 6.7; 8.2; 9.6 million, a total of 29.3 million instead of 38.4 on “mature from day one”: a gap of RUB 9.1 million.

Plan versus actual as learning

The deviation is split by cause: price, volume, mix, markdowns, costs. The goal is not to find a culprit but to understand what to change in the next plan: the markdown assumption or the traffic assumption. Without it a plan stays a ritual.

The value of a forecast stage: naive, model, adjustments

A forecast is made in stages: the last season as a starting point, a model, then manual adjustments at approvals. Each stage must make the forecast better than the previous one, and this is checked on the same groups and actuals. On six illustrative groups (actual 6,140 pieces) the WAPE of the naive is 11.4%, of the model 6.0%, after adjustments 8.1%. The value of the model is +5.4 pp, the value of the adjustments −2.1 pp: the adjustments raised the bias to +7.5%.

Formula: The value of a stage = the error before the stage − the error after (the term “Forecast value added”)

What to do: If the FVA of a step is below zero, the step is removed or kept only where there is new information behind it: a contract, a promotion, a change of channel.

Think first: The WAPE of the model is 6.0%, after manual adjustments 8.1%. What did the adjustments give?

Answer: A value of −2.1 pp: the adjustments worsened the forecast, and in 4 groups of 6 the result is worse than the model.

The error in roubles is asymmetric

One piece of over-forecast costs the markdown on the excess and holding: 6,000 × 40% + 2,400 × 12% = RUB 2,688. One piece of under-forecast costs lost margin: (6,000 − 2,400) × 70% = RUB 2,520. The price of the error is RUB 1,811,040 for the naive, RUB 954,240 for the model, RUB 1,340,640 for the result: a smaller error in pieces is not always cheaper, so compare the stages both in percent and in roubles.

Formula: Price = over-forecast × (price × markdown + cost × holding) + under-forecast × (price − cost) × share lost

What to do: Set the price of a unit of over- and under-forecast from your own past seasons and assess the forecast stages in roubles.

The season is recalculated by the actual with a weight by weeks

On illustrative data the season forecast is 8,142.9, below the plan; the maximum seasonality index is 1.88.

Formula: κ = (1 − w) + w × actual ÷ plan to date

What to do: Recount once in two to four weeks and exclude an anomalous year from the indices.

How to work through it

  1. Collect monthly sales history for at least two to three years and clean it of one-off events.
  2. Compare several forecast methods by WAPE on held-out data.
  3. Add adjustments: promotions, price changes, new and closed stores.
  4. For a retailer split the plan by stores; for a brand reconcile the order book with the target.
  5. Build the annual budget by month: revenue, margin, costs, EBITDA.
  6. Find the months where the budget goes into the red and decide how to cover it.
  7. Run a stress check: what happens if sales fall 10–20%.
  8. After the period review plan versus actual by cause and carry conclusions forward.

Where to start

Retail: “Sales forecast”, then “Sales plan by store” and “Annual budget”. Brands: “Order book versus the plan”, then “Sales forecast” and “Annual budget”. After the period — “Budget: plan versus actual”. Calculations of the topic in order: “Sales forecast: method, horizon, error”, “Sales plan by store”, “Brand order book versus the season plan”, “Annual budget by month”, “Income and expense budget: plan versus actual”, “Zero-based budget: cost packages, priority and the effect of cut-off by level”, “Seasonal buying: critical ratio and the optimal quantity under normal demand”, “Season Monte Carlo: the probability of meeting margin and profit”, “Holt-Winters: a forecast with trend and seasonality, error on held-out periods”, “Forecast value added by stage and the price of error in roubles: the season assortment”, “In-season reforecast”, “Seasonality indices from history”, “Weather adjustment of demand”, “Top-down vs bottom-up forecast reconciliation”. See also (their primary topic is another one): “Management balance sheet from the P&L and turnover figures”, “Overheads by product and responsibility center: three allocation keys”.

Calculations in this topic

  • Sales forecast: method, horizon, error

    Builds a forecast from sales history using four methods, shows each method’s error on past data and suggests which is more reliable for your series.

    How much you will sell in the coming periods and how reliable the forecast is.

  • Sales plan by store

    Splits the chain’s revenue plan across stores: growth of existing stores, new stores ramping up, closures and partial years. Shows the total against last year and against the target.

    What revenue plan per store gives the required chain plan and how much it relies on new stores.

  • Brand order book versus the season plan

    Checks confirmed and expected buyer orders against the season target: plan coverage, weighted forecast, gap and dependence on the largest customer.

    Whether orders and negotiations cover the season target and how far the plan hangs on one customer.

  • Annual budget by month

    From last year’s revenue, growth and cost structure builds a monthly income and expense budget: revenue, gross profit, EBITDA and the months in which the business goes into the red.

    What revenue and EBITDA result by month and in which months the budget goes into the red.

  • Income and expense budget: plan versus actual

    Combines plan and actual of the profit and loss statement, shows variances by line and what drove the profit.

    How far actual deviated from plan and which line affected profit most.

  • Zero-based budget: cost packages, priority and the effect of cut-off by level

    Takes cost packages with priority, cost and effect, funds the mandatory ones and then by priority up to the limit, shows the cut-off line, the money saved and the effect of the cut packages; the table compares budget levels from 100% to 70% of the request.

    Which cost packages to fund within the limit, where the cut-off line runs and how much effect is lost per rouble saved.

  • Seasonal buying: critical ratio and the optimal quantity under normal demand

    Finds the buying quantity of a seasonal item that maximizes expected profit when demand is given by a mean and a spread: the critical ratio, the stock-out probability, expected leftovers and lost sales, the gain against your plan.

    How much of a seasonal item to buy so that expected profit is highest when demand is uncertain, and what that gives against your plan.

  • Season Monte Carlo: the probability of meeting margin and profit

    Plays the season many times with random demand and cost (with a fixed seed, the result is reproducible) and shows the probability of meeting the margin and profit targets, profit percentiles, the average tail of bad seasons, the stock-out probability and the expected leftover.

    With what probability the season meets the margin and profit targets and what a bad season looks like if demand and cost are random.

  • Holt-Winters: a forecast with trend and seasonality, error on held-out periods

    From a series of sales by period builds a forecast with level, trend and seasonal indices (additive or multiplicative), picks the smoothing parameters, tests the accuracy on the last held-out periods (WAPE, bias, a comparison with the naive seasonal forecast) and shows an approximate band.

    What sales will be in the coming periods if the seasonal cycle repeats and the trend holds, and whether the model beats the simple “as in the last season”.

  • Forecast value added by stage and the price of error in roubles: the season assortment

    For groups of the assortment compares three forecasts of one season: naive (a repeat of the last season), a model (statistics or AI) and the result after manual adjustments. Counts the WAPE and bias of each stage, the value added of each step (FVA) and turns the error into roubles: over-forecast costs a markdown and holding, under-forecast costs lost margin. The definitions of WAPE, bias and FVA are given in the “Forecast accuracy by week” calculation (fcdash); here the point is the price of error and the stages.

    Which stage of the forecast adds accuracy (the model or manual adjustments), how biased it is and how many roubles the error costs in the form of markdowns on the excess and lost margin.

  • In-season reforecast

    Adjusts the forecast of the rest of the season by how the actual is going: the more weeks have passed, the more weight the ratio of the actual to the plan to date gets. Shows the season forecast, the deviation from the plan and the change in the rest.

    What the season forecast is if the rest behaves like the actual to the plan but with caution at the start of the season.

  • Seasonality indices from history

    Counts 12 monthly seasonality indices from sales over 2–3 years: each month is divided by the monthly average of its year, the indices are averaged and brought to a sum of 12. Shows the span of seasonality and spreads the annual plan across months.

    What seasonal shape the sales have by your history and what share of the year falls on each month.

  • Weather adjustment of demand

    By weeks builds the link of category sales with the temperature deviation from the norm (a least squares line), estimates its reliability and recalculates the forecast for the expected temperature deviation.

    How much weekly sales depend on the temperature deviation from the norm and what adjustment to make to the forecast.

  • Top-down vs bottom-up forecast reconciliation

    Reconciles the category forecast (top-down) and the sum of the SKU forecasts (bottom-up): the scale, the bias of SKU forecasts and the reconciled SKUs at a given trust in the category level. Shows how much more optimistic the SKU forecasts are than the category.

    How much more optimistic the position forecasts are than the category and how to reconcile them with the chosen trust in the category.

  • Management balance sheet from the P&L and turnover figures

    From the P&L, the inventory, receivable and payable days and the opening balance sheet builds the closing balance sheet: cash by the indirect method, assets, liabilities, equity; checks that it ties, shows net working capital and net debt and compares the calculated cash with the actual.

    What the closing balance sheet will be from the P&L and turnover figures, whether it ties and how much cash is tied up in turnover.

  • Overheads by product and responsibility center: three allocation keys

    Splits the common pool of overheads across lines (products, categories, channels or financial responsibility centers) with three keys: by revenue, by units and by cost driver (orders, shipment lines, hours). Shows each line’s profit under each key and which lines change sign: the allocation key can make a line “loss-making” or “profitable” without a rouble of real saving.

    How the profit of each line changes with the way overheads are allocated and which conclusions do not depend on the key.

Self-check questions

  • Does the sales history cover at least two seasons?
  • Is the choice of forecast method justified by error on held-out data?
  • Are promotions, price changes and new stores accounted for by separate adjustments?
  • Is a ramp-up curve set for the new store?
  • Does the order book separate confirmed orders from negotiations?
  • Are costs separated into fixed and variable?
  • Are months with negative cash results visible together with a way to cover them?
  • Is there a sales-drop scenario and a threshold after which we act?
  • Do men’s and women’s lines have their own assumptions, not one common factor?

Knowledge check

Multiple-choice questions. The correct answer and the explanation follow each question.

  1. Two categories: actual RUB 10 million, forecast 14; actual 90, forecast 81. What is the WAPE?

    • A. 25%
    • B. 13%
    • C. 14.4%
    • D. −5%

    Correct answer: B. 13%. WAPE = sum of absolute deviations / sum of actuals = (4 + 9) / 100 = 13%. 25% is the mean of the percentages by category (40% and 10%): the small category with its 40% dictates the verdict. 14.4% comes from dividing 13 by the actual of one category (90) instead of the total. −5% is the bias (95 − 100) / 100: the sign is kept and errors of opposite signs cancel, so bias does not replace WAPE.

  2. Revenue is RUB 90 million, variable costs 60%, fixed costs RUB 28 million. A new lease is planned and fixed costs will rise to RUB 34 million. By what percentage can revenue now fall before profit reaches zero?

    • A. 22.2%
    • B. 37.8%
    • C. 6.7%
    • D. 5.6%

    Correct answer: D. 5.6%. The contribution per ruble is 1 − 0.6 = 0.4. Break-even revenue is 34 / 0.4 = RUB 85 million, the fall is (90 − 85) / 90 ≈ 5.6%. Profit is then 90 × 0.4 − 34 = RUB 2 million: there is almost no cushion. 22.2% was the figure before the lease (28 / 0.4 = 70, a fall of 20 million). 37.8% is 34 / 90, the share of fixed costs in revenue, not the safety margin. 6.7% is 6 / 90, the increase in fixed costs relative to revenue.

  3. The WAPE of the naive forecast is 11.4%, of the model 6.0%. What is the value of the model (FVA)?

    • A. −5.4 pp
    • B. +5.4 pp
    • C. 5.4%
    • D. 17.4 pp

    Correct answer: B. +5.4 pp. FVA = the error before the step − the error after the step. Plus means the model added accuracy. Minus results from subtracting the other way round, and percent must not be called percentage points. Covered in: The value of a forecast stage: naive, model, adjustments.

  4. The adjustments raised the error from 6.0% to 8.1%. Which conclusion is right?

    • A. The adjustments improved the forecast because the bias grew
    • B. Accuracy did not change, the difference is within noise
    • C. The model must be replaced by the naive forecast
    • D. The adjustments worsened the forecast; keep them only given new information

    Correct answer: D. The adjustments worsened the forecast; keep them only given new information. The value of the adjustments is −2.1 pp, in 4 groups of 6 the result is worse than the model. The model is better than the naive one (WAPE 6.0% against 11.4%), so it need not be replaced. Covered in: The value of a forecast stage: naive, model, adjustments.

  5. The price is RUB 6,000 excl. VAT, the markdown on the excess 40%, the cost RUB 2,400, holding 12% of cost. What does one piece of over-forecast cost?

    • A. RUB 2,688
    • B. RUB 2,520
    • C. RUB 3,600
    • D. RUB 2,400

    Correct answer: A. RUB 2,688. 6,000 × 40% + 2,400 × 12% = RUB 2,688. RUB 2,520 is the price of a piece of under-forecast: (6,000 − 2,400) × 70%. Covered in: The error in roubles is asymmetric.

  6. What does the weight of the actual do in the season reforecast?

    • A. Always equals one
    • B. Depends on the price
    • C. Grows with the weeks passed
    • D. Equals zero

    Correct answer: C. Grows with the weeks passed. The first weeks are random. Covered in: The season is recalculated by the actual with a weight by weeks.

  7. A span of seasonality indices below 15% means what?

    • A. There is no seasonality in the data
    • B. Seasonality is maximal
    • C. One more year is needed
    • D. The forecast is accurate

    Correct answer: A. There is no seasonality in the data. Do not spread the plan by such indices. Covered in: The season is recalculated by the actual with a weight by weeks.

Common mistakes

  • Mistake: Seasonality from one year

    What it costs: A one-off taken for a season

    How to avoid: Two to three years of history.

  • Mistake: Ignoring promotions and price changes

    What it costs: The forecast breaks in promotion months

    How to avoid: Month adjustments.

  • Mistake: Mature revenue for a new store

    What it costs: Plan overstated by 20–25%

    How to avoid: Ramp-up curve.

  • Mistake: Negotiations as orders

    What it costs: Target looks covered but is not confirmed

    How to avoid: Probabilities from previous seasons.

  • Mistake: One annual plan without revision

    What it costs: By summer it does not reflect reality

    How to avoid: Quarterly revision.

  • Mistake: Manual adjustments over the model “just in case”

    What it costs: The forecast is systematically overstated, leftovers grow The adjustments added losses of RUB 386,400 per season (markdowns on the excess and holding).

    How to avoid: Record the author and the reason of each adjustment and measure the value of adjustments. An adjustment is acceptable given new information and with an author; after a season compare the FVA of adjustments by groups.

    What it looks like (illustrative figures): The model gave a WAPE of 6.0%, after approvals 8.1% with a bias of +7.5%.

  • One year in the seasonality indices

    What it looks like (illustrative figures): The indices repeat the accidents of the year.

Russia and the EAEU

When planning in Russia, money and taxes should be kept on one basis. Revenue in the plan is usually shown without VAT; the 22% rate from 01.01.2026 affects the shelf price and cash flows: the tax payment to the budget falls in a different month from the receipt of revenue. Corporate income tax is 25% for payers of the general system, USN provides rates of 6% on income or 15% on income less expenses (check rates and thresholds: they are revised; the revenue threshold after which the simplified system requires VAT is 20 million rubles under Law 228-FZ, with a number of special rates).

Timing of receipts matters for the budget: marketplaces pay in cycles, chains with a deferral, retail within days. A cash plan is built separately from the profit budget, because profit and cash diverge. Include the key rate in the cost of money in stock and credit: it changes, take the current one.

For a brand in Russia buyers' orders are often formalized by contracts with prepayment and deferral: a confirmed order means a signed document, not an oral agreement. Export supplies to EAEU countries bear VAT under union rules, and to third countries a 0% rate is possible with confirmed export: the time to collect documents affects the tax refund.

The planning method is the same; taxes, currency and the demand curve differ. As a guide, check with a local accountant: corporate income tax is 20% in Belarus and Kazakhstan, 18% in Armenia, 10% in Kyrgyzstan; VAT is 20%, 16%, 20% and 12% respectively. In Kyrgyzstan, benefits tied to the average wage apply to the garment industry from 01.01.2026 to 01.01.2030 (medium reliability). Keep a multi-currency budget with an explicit exchange rate: the Belarusian ruble, tenge, dram and som fluctuate differently, and the rate on the payment date changes margin. A brand’s order book by union country is useful for checking concentration: a buyer in one country may depend on a single tax regime.

Categories and audiences

Apparel
In apparel the forecast rests on strong seasonality and weather, and the plan is split by collections and sales stages. WAPE is higher for fashion items, lower for basics. Season markdowns must be built into budget margin, otherwise EBITDA is overstated.
Footwear
In footwear the forecast is by style and size, and the buying budget is predetermined by minimum quantities and factory lead times: error cannot be fixed by reordering. Size returns and reverse logistics belong in budget costs and are often underestimated.
Accessories
In accessories sales come in bursts with peaks around holidays: a monthly forecast gives a high error, better to look at weeks around dates. A small receipt makes revenue sensitive to traffic and the markup is higher, so budget gross margin is steadier than in apparel.
Kids
The children’s plan is tied to the school calendar and growth: a peak in June–September and before New Year, with troughs between. Seasonality in this category is sharper, so weekly indices are needed. So the plan is built by week and by age group, not as one number for the season.
Men
In the men’s plan the main risk is tied to the calendar: 23 February and year-end give a noticeable share of monthly revenue, and a forecast “for an average month” blurs them. For these months a separate season index and a separate stock depth for gift items (shirts, sweaters, accessories) are set. The men’s line can live as a separate budget; a common growth factor does not fit it.
Women
If planned markdowns are set as one share for the brand, they hide the difference between lines: for example, a common 15% can mean 20% for one line and 8% for another. The women’s line needs its own growth factor and its own markdown share, and the peaks of 8 March, spring and autumn are tied to it. Otherwise budget error accumulates month by month.

Questions this topic answers

  • How much will we sell in the coming periods and which forecast method is more accurate?
  • What plan does each store give and how much does it rest on new stores?
  • Do buyers’ orders cover the season target?
  • In which months does the budget go into the red and is there cash for it?

Key measures and terms

Study first

What next

Next step: “Scenarios and plan stress test”, “The 13-week cash plan”. Related: “Buying budget and cash”, “Stock and turnover”, “Store and chain”.

Documents on this topic

Open interactively →