# Revenue Modeling > Builds SaaS revenue models - a reconciling MRR/ARR bridge, cohort retention forecasting, a driver tree from leads to new ARR, and base/upside/downside scenarios. Use when someone asks "model my ARR for next year", "what will MRR be if churn doubles", "build a revenue forecast for the board", "why doesn't my MRR bridge tie out", or "what NRR do we need to hit our plan". Do NOT use for sizing the market opportunity - use market-sizing instead; for CAC, LTV, and payback analysis use unit-economics; for a full P&L or headcount-driven operating model use fpa-model; for cash timing and runway use cash-flow-forecast. - Skill: `skillmedev/revenue-modeling` (Agent Skill) - Install (CLI): `npx skillmds@latest add skillmedev/revenue-modeling` - Raw SKILL.md: https://api.skillmd.com/api/skills/skillmedev/revenue-modeling/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Data & Analytics - Author: SkillMedev (https://skillmd.com/u/skillmedev) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/skillmedev/revenue-modeling --- # Revenue Modeling A revenue model that does not reconcile is a story, not a model. The costly failure mode is a forecast built from a single blended growth rate: it hides which lever (new logos, expansion, churn) is actually moving, so the plan cannot be managed and the miss cannot be diagnosed. Build the model from a bridge that must tie out and drivers that someone owns. ## Operating procedure Steps run in this order because the bridge defines the accounting, the drivers feed the bridge, and scenarios only mean something once base-case drivers are pinned. ### Step 1: gather inputs Collect, and label every unsourced number a guess: - Beginning MRR (or ARR) and customer count, tied to the billing system or GL - not a spreadsheet someone remembers. - Trailing 6-12 months of the bridge components: new, expansion, reactivation, contraction, churned MRR. - Funnel history: leads (or pipeline created), lead→opportunity rate, opportunity→win rate, average deal size (ACV). - Cohort retention: revenue retained by acquisition month at months 1, 3, 6, 12. If unavailable, start with logo churn and note the model is weaker for it. - Billing mix (annual prepay vs monthly) - needed later to separate bookings, billings, and recognized revenue. Defaults when history is thin: mid-market SaaS commonly sees 15-30% lead→opp, 15-25% opp→win, and 2-4% monthly gross revenue churn for SMB versus 0.5-1.5% for enterprise. Use these only as placeholders and say so. ### Step 2: build the MRR bridge Every period: ``` Ending MRR = Beginning MRR + New MRR (newly acquired customers) + Expansion MRR (upgrades, seat adds, cross-sell) + Reactivation MRR (returning churned customers) - Contraction MRR (downgrades) - Churned MRR (cancellations) ``` Validate every period: components must sum exactly to the change in MRR, and the bridge must reconcile against the general ledger. If it is off by even 1%, find the leak before forecasting - the usual suspects are mid-period upgrades booked as new, or reactivations netted into churn. ### Step 3: compute the key rates - Gross revenue churn = (churned + contraction MRR) / beginning MRR. - Net revenue retention (NRR) = (beginning + expansion − contraction − churned) / beginning. NRR above 100% means the existing base grows with zero new logos; durable public-SaaS benchmarks cluster at 100-120%, with under 90% a red flag for the model's whole premise. - Logo churn = customers lost / customers at start. Track it separately from revenue churn - losing many small logos and one big logo look identical in revenue terms but demand different fixes. ### Step 4: build the driver tree for new revenue Model New MRR bottom-up, never as a single growth rate. Decompose top-down until every leaf is a number one team owns: ``` New ARR ├── New logos won │ ├── Leads (marketing owns) │ ├── Lead → Opportunity rate (marketing/SDR owns) │ └── Opportunity → Win rate (sales owns) └── Average deal size / ACV (sales + pricing own) New MRR = leads × lead_to_opp × opp_to_win × avg_deal_size / 12 ``` Procedure: (1) list the leaves; (2) assign each an owner and a historical baseline; (3) check the tree multiplies back to actual new MRR for the trailing two quarters - if it misses by more than 10%, a driver is mis-measured; (4) forecast each leaf, not the product. This is what connects marketing spend and sales capacity to the revenue line. ### Step 5: forecast the existing base with cohorts Forecast retained revenue per acquisition cohort, then layer new cohorts on top: ``` For each acquisition cohort: revenue_t = cohort_initial_mrr × retention_curve[t] × expansion_factor[t] Total MRR_t = Σ cohort revenue_t + new cohorts from the driver tree ``` Fit the retention curve from historical cohorts. Do not assume linear decay - SaaS retention curves flatten after months 3-6, and a linear assumption understates long-run revenue badly. ### Step 6: run scenarios Build base, upside, and downside by flexing win rate, churn, expansion, and ACV - one inputs sheet, scenarios one toggle away (structure it with spreadsheet-model-builder). Add a 2D sensitivity grid: NRR on one axis, new-logo growth on the other, ending ARR in the cells. Stress-test the downside explicitly: what happens to the plan if churn doubles. ### Step 7: connect to cash and economics - ARR = MRR × 12. - Keep bookings, billings, and recognized revenue separate - annual-prepay bookings inflate cash today but recognize monthly. Hand cash timing to cash-flow-forecast. - CAC payback = CAC / (ARPA × gross margin); LTV:CAC should exceed 3. The full treatment lives in unit-economics - reference it, do not rebuild it here. ## Worked example: one quarter of the ARR model Self-contained Python. Save as `arr_model.py`, edit the inputs, run `python3 arr_model.py`. ```python inputs = { "beginning_arr": 2_400_000, "beginning_customers": 200, "leads_per_quarter": 900, "lead_to_opp": 0.20, "opp_to_win": 0.22, "avg_deal_acv": 14_000, "quarterly_gross_churn": 0.03, # churned + contraction, % of beginning ARR "quarterly_expansion": 0.045, "quarterly_logo_churn": 0.025, } def quarter(i): wins = i["leads_per_quarter"] * i["lead_to_opp"] * i["opp_to_win"] new_arr = wins * i["avg_deal_acv"] expansion = i["beginning_arr"] * i["quarterly_expansion"] churn = i["beginning_arr"] * i["quarterly_gross_churn"] ending_arr = i["beginning_arr"] + new_arr + expansion - churn nrr_annualized = ((i["beginning_arr"] + expansion - churn) / i["beginning_arr"]) ** 4 customers = i["beginning_customers"] * (1 - i["quarterly_logo_churn"]) + wins return wins, new_arr, expansion, churn, ending_arr, nrr_annualized, customers wins, new_arr, exp, churn, end, nrr, cust = quarter(inputs) print(f"New logos won: {wins:.1f}") print(f"New ARR: ${new_arr:,.0f}") print(f"Expansion ARR: ${exp:,.0f}") print(f"Churned+contraction: ${churn:,.0f}") print(f"Ending ARR: ${end:,.0f}") print(f"NRR (annualized): {nrr:.1%}") print(f"Ending customers: {cust:.0f}") ``` Output: ``` New logos won: 39.6 New ARR: $554,400 Expansion ARR: $108,000 Churned+contraction: $72,000 Ending ARR: $2,990,400 NRR (annualized): 106.1% Ending customers: 235 ``` Read it: 900 leads at 20% and 22% conversion yield ~40 logos at $14k ACV - $554k of new ARR. The base contributes $108k expansion against $72k churn, an annualized NRR of 106%. Ending ARR of $2.99M is 25% quarterly growth, and the bridge decomposition shows 84% of it comes from new logos - so the plan lives or dies on the funnel, not retention. ## Deliverable Produce a revenue model workbook (or script) containing: a monthly or quarterly MRR bridge that reconciles to the ledger, the driver tree with an owner and baseline per leaf, cohort-based retained-revenue forecast, base/upside/downside scenarios with a single inputs sheet, an NRR × new-logo-growth sensitivity grid, and an assumptions page documenting every number's source. Show actuals vs forecast each period and track forecast accuracy over time. ## Do NOT - Do not forecast with a single blended growth rate - it cannot be diagnosed when it misses and no one owns it. - Do not let the bridge fail to reconcile "for now"; every downstream number inherits the error. - Do not blend contraction into churn or reactivation into new - the fixes for each are different teams. - Do not treat bookings as revenue; annual prepay makes them wildly different in-quarter. - Do not assume linear retention decay; fit the curve from real cohorts. - Do not bury assumptions inside formulas; every input lives on the inputs sheet with a source note. ## Quality bar Ship only when: the bridge ties to the GL within rounding for every historical period; the driver tree reproduces trailing actual new MRR within 10%; every assumption has a source or an explicit "guess" label; the downside scenario is genuinely uncomfortable (churn doubled, win rate down a third), not base-minus-5%; and a reader can trace ending ARR back to leads without opening a hidden tab.