Revenue Modeling

Forecasting produces a number; modeling produces a number plus the structure that explains it. A forecast says 'Q4 will be €1.2M.' A model says 'Q4 is 850k sessions × 2.1% CR × €68 AOV, and the CR assumption carries the most risk.' Modeling is what makes a forecast defensible and improvable.

Revenue intelligence is the broader discipline of using data to understand and grow revenue — it includes attribution, cohort analysis, retention curves, and experimentation. Revenue modeling is one specific practice inside it: building the quantitative driver model that other revenue-intelligence work plugs into.

Bottom-up (drivers × rates) is almost always more useful because it tells you which lever to pull. Top-down (market size × share) is faster and works for early-stage businesses with no operating history. Mature stores should run bottom-up as the primary model and use top-down as a sanity check.

Granular enough that each input is something one person on the team owns. Splitting by channel (paid social, paid search, organic, email, direct) is usually the right first cut because acquisition teams own each one. Going further — by campaign, by product — adds noise faster than insight unless you're already running channel-level models cleanly.

Re-baseline the inputs monthly using the last 30-90 days of actual data, and revisit the structure quarterly. Inputs drift constantly (CR moves with traffic mix, AOV moves with promo cadence); the model structure changes only when the business changes — new market, new subscription tier, new channel.

Flexing each input by a plausible range — say ±20% — and recording how much the revenue output changes. The input with the biggest swing per percentage point of change is your highest-leverage lever. On most Shopify stores, conversion rate beats traffic on a per-point basis because CR compounds across all channels at once.

GA4 gives you the historical inputs — sessions, conversion rate, AOV by channel — but it isn't a modeling tool. Teams typically pull GA4 data into a spreadsheet or a BI layer and build the model there. Metricuno's GA4 import shortcuts this by importing 13 months of historical inputs on day one, so you're not building a model from a six-week sample.

Anchor the inputs to comparable products you already sell or to public benchmarks for the vertical (see the table above). Build three scenarios — conservative, base, stretch — and document the assumption that separates them. The goal isn't a perfect forecast; it's a model you can revise weekly once real traffic starts hitting the page.

New signups, activation rate, retention curve (cohort survival by month), and ARPU. Revenue at month N = sum over cohorts of (cohort_size × retention_at_N × ARPU). Retention is the dominant input — a model with crisp retention curves and rough acquisition assumptions beats one with the opposite split every time.

Using a single point estimate for each input instead of a range. The model spits out one revenue number, the team treats it as a target, and then nobody updates it when reality drifts. Always model a range, always track actuals against the range weekly, and always flag which input is breaking the model when it does.