Revenue Forecasting
Revenue forecasting models future sales from historical trends, cohorts, seasonality, and planned campaigns. Here's the formula, accuracy benchmarks, and how to choose a method.
Revenue Forecasting
Projecting future revenue from historical data, cohorts, seasonality, and planned demand drivers — used for budgeting, hiring, and inventory.
Revenue forecasting is the practice of estimating the sales an online store will generate over a future window — typically the next month, quarter, or year — by combining historical trends, customer cohort behaviour, seasonality patterns, and the expected lift from planned campaigns or product launches.
A forecast isn't a prediction in the lottery sense; it's a planning instrument. Its job is to be directionally right so you can commit to inventory orders, hiring plans, and paid-media budgets with a defensible range. As a sub-discipline of revenue intelligence, it converts what already happened in the store into a credible view of what's likely to happen next.
Most stores in the €1M-€15M range run one of three forecast styles: a trend extrapolation (last 12 months projected forward with a growth rate), a cohort model (orders per acquired customer, decayed by month), or a bottom-up channel build (sessions × conversion rate × AOV per channel). Each has a different failure mode.
Trend models break when the market shifts — a new competitor, a paid-channel cost spike, an iOS privacy change. Cohort models break when acquisition mix changes faster than the cohorts mature. Bottom-up models break when you forget that channels cannibalise each other. The fix is rarely a better single model; it's running two methods in parallel and watching where they disagree.
Forecast Revenue = (Sessions × Conversion Rate × AOV) × Seasonality Index + Campaign Lift
Sessions
Forecast sessions
Expected store sessions in the period, based on trended channel traffic.
Conversion Rate
Site conversion rate
Trailing 90-day session-to-order conversion, adjusted for known UX changes.
AOV
Average order value
Trailing-period AOV, adjusted for planned price or bundle changes.
Seasonality Index
Seasonality multiplier
Period's share of annual demand vs. the trailing baseline (e.g. 1.35 in November).
Campaign Lift
Campaign lift
Incremental revenue from planned promotions or launches not captured in baseline.
An apparel store forecasting November revenue. The baseline is 420,000 sessions at 2.4% conversion and €78 AOV. November historically runs 1.35× the trailing baseline, and a Black Friday email push is expected to add €120,000 of incremental revenue.
Sessions: 420,000
Conversion Rate: 2.4%
AOV: €78
Seasonality Index: 1.35
Campaign Lift: €120,000
→ ≈ €1.18M forecast November revenue
Baseline produces ~€786k; the seasonality multiplier pushes it to ~€1.06M; Black Friday adds the final €120k. If actuals land between €1.06M and €1.30M, the forecast was useful — that's a ±10% band the planning team can buy inventory against.
Accuracy expectations should scale with horizon. A 30-day forecast for a store with steady traffic should land within 5-10%; a 12-month forecast is doing well at ±20%. Forcing tighter precision on a long horizon usually means hiding assumptions inside spreadsheet cells rather than improving the model.
Typical forecast accuracy by horizon and method (mean absolute percentage error)
| Horizon | Trend extrapolation | Cohort model | Bottom-up channel build |
|---|---|---|---|
| 30 days | 4-8% | 5-9% | 6-10% |
| 90 days | 8-14% | 7-12% | 9-15% |
| 6 months | 12-20% | 10-18% | 14-22% |
| 12 months | 18-30% | 15-25% | 20-35% |
Two practical rules. First, always forecast a range, not a number — a P10/P50/P90 band tells the team how much risk lives in the plan. Second, review forecast variance every month: the gap between forecast and actual is where you learn which assumption was wrong, and that's the only way the next forecast gets better.
Revenue forecasting FAQ
Revenue intelligence is the broader discipline of understanding revenue drivers — cohorts, channels, retention, margin. Revenue forecasting is one output of that intelligence: a projection of where those drivers point next. You need the intelligence layer first; otherwise you're forecasting numbers you can't explain.
Run three horizons in parallel: a rolling 30-day for cash and ad spend pacing, a 90-day for inventory and hiring, and a 12-month for board planning. Each uses a different method and a different tolerance for error — don't try to make one model serve all three.
For stores under €5M revenue with stable traffic, a cohort model usually wins on horizons beyond 90 days because it captures repeat-purchase behaviour. For shorter horizons or stores driven by paid acquisition, a bottom-up channel build tends to be more accurate because it ties directly to controllable inputs.
Calculate a seasonality index — each month's share of trailing annual revenue divided by 1/12 — and apply it as a multiplier on baseline. Use at least two years of history if you have it; one year of seasonality can be skewed by a single anomalous event like a stockout or a viral moment.
No. Keep baseline revenue and campaign lift as separate components. If you bake a Black Friday spike into the baseline, next year's model will assume that spike happens every week. Forecast the underlying business, then add named campaign lifts on top.
It's three scenarios: a pessimistic (P10), expected (P50), and optimistic (P90) revenue outcome. Inventory and hiring decisions reference different points — you order safety stock against P90 but commit fixed costs against P10. A single-point forecast hides this risk structure.
Refresh inputs weekly, rebuild the model monthly, and rebase the assumptions quarterly. Weekly updates catch traffic and conversion drift early. Monthly rebuilds prevent stale cohorts. Quarterly rebases force you to re-examine the structural assumptions — pricing, channel mix, return rates.
Usually one of three things: you're under-modelling repeat purchase from existing cohorts, you're using a too-old conversion rate that hasn't reflected recent UX wins, or you're not crediting campaigns that have become 'always-on' (like retargeting) as part of baseline. Look at the variance composition, not the headline gap.
Yes, but you'll lean harder on order data from Shopify, WooCommerce, or Magento — sessions and channel attribution become assumptions rather than measurements. A platform that imports historical analytics on day one gives you a much faster path to a credible forecast than waiting 6-12 months to accumulate fresh data.
Finance owns the number that goes to the board; marketing and e-commerce own the inputs that drive it. The healthiest setup is a single shared model where marketing edits traffic and campaign assumptions, e-commerce edits conversion and AOV, and finance reconciles the output. Two separate forecasts always diverge — and the conversation becomes about whose spreadsheet is right rather than what's actually happening.
See Metricuno on your data
Bring your stack — Google Analytics, Stripe, a CRM, anything — and we'll walk through the metric tree that turns your funnel into one number.