ARR forecasting is the number your board fixates on, your next round gets priced against, and your own team plans hiring around — and most of the forecasts behind it are built on a single growth-rate assumption typed into one cell. A founder at a $6M ARR SaaS company told us his "forecast" was last year's ARR times 1.4, because that's what the deck needed to show. Then a single enterprise logo churned in Q2 and the whole year unwound. Good ARR forecasting is not a multiplier. It's a model that builds annual recurring revenue from its moving parts — new business, expansion, contraction, and churn — so that when one part moves, you see it coming instead of explaining it afterward.
This is for founders and finance leads at $1M–$50M ARR companies who are tired of forecasts that look great until they don't. We'll cover the method, the five places it breaks, and what "good" looks like.
Start with the ARR bridge, not a growth rate
The single most useful artifact in ARR forecasting is the ARR bridge — the waterfall that takes you from starting ARR to ending ARR through four movements:
- New ARR — net-new logos closed in the period.
- Expansion ARR — upsell, cross-sell, and seat growth from existing customers.
- Contraction ARR — downgrades and seat reductions that don't fully churn.
- Churned ARR — customers who left entirely.
Ending ARR = Starting ARR + New + Expansion − Contraction − Churned. That identity is the whole model. A growth-rate forecast hides all four movements inside one number; the bridge forces you to forecast each one with its own driver. When the actual lands, you don't ask "why were we off" — you look at which of the four lines missed.
If you can't build the bridge for the last eight quarters from your billing data, that's the first problem to fix. Everything downstream depends on it.
Forecast new ARR from pipeline, not from hope
New ARR is where most forecasts inflate. The disciplined version ties new business to two things you can actually measure: sales capacity and pipeline conversion.
For sales-led motions, build it bottom-up: number of ramped reps × average quota attainment × the share that lands as ARR. A rep carrying $600K of quota at 75% blended attainment is $450K of new ARR — not the $600K the board deck wants. For product-led motions, build new ARR from top-of-funnel signups × trial-to-paid conversion × average new-customer ARR.
Either way, anchor the conversion assumptions to your trailing numbers. If trial-to-paid has run 4% for six quarters, forecasting 7% because the new onboarding flow "should help" is how forecasts drift. Model the base case at 4%, put the 7% in an upside scenario, and label it.
Treat churn and contraction as their own forecast
Net revenue retention is the quiet driver that decides whether your ARR forecast compounds or leaks. It rolls up gross customer retention and the expansion that offsets it. A company adding $2M of new ARR a year with 85% NRR is running up a down escalator; the same company at 115% NRR barely has to sell to grow.
Forecast retention from cohorts, not a blended average. Pull gross and net retention by signup cohort and by segment — SMB churn behaves nothing like enterprise. A single blended churn rate buries the fact that your SMB tier is bleeding 3% a month while enterprise holds. Model the two separately, then roll them up. Your MRR and ARR forecasts are only as good as the retention curve underneath them, and the retention curve is almost always more segmented than the spreadsheet admits.
Build three scenarios, commit to one
A single-line ARR forecast is a guess wearing a suit. Build three:
- Base — your honest expectation. Conversion and retention at trailing rates. This is the number you manage to.
- Downside — a logo concentration risk fires, conversion slips 20%, a quarter of expansion doesn't materialize. This is the number that determines your runway and your hiring freeze trigger.
- Upside — the new motion works, a partnership lands. This is the number you're allowed to get excited about, and the one you never put in the board commit.
The point of three scenarios isn't false precision. It's that the gap between base and downside tells you how fragile the plan is. If a 20% conversion miss turns a fundable year into a layoff year, you have a concentration problem no forecast formatting will fix. Run the scenarios with a scenario planner so you can flex one driver at a time and watch ending ARR respond.
The five places ARR forecasting breaks
After enough forecasts, the failure modes rhyme:
- Booking vs. recognized confusion. A signed annual contract is bookings; ARR is the run-rate of recurring revenue. Counting a multi-year prepay as a single-year ARR spike overstates the number and wrecks next year's comp. Keep revenue recognition discipline separate from the ARR bridge.
- One blended churn rate. Covered above — it hides segment-level bleeding.
- Expansion modeled as a percentage of new. Expansion is a function of the installed base and product usage, not of how much you sold last quarter. Model it off the base.
- No ramp on new hires. Eight reps hired in Q1 are not eight reps of capacity in Q1. Most carry near-zero quota for two quarters. Forecasts that ignore ramp front-load new ARR that physically can't close.
- The forecast that never updates. A forecast built in January and untouched through Q2 is a historical document. Re-forecast monthly against the bridge.
What a trustworthy ARR forecast looks like
You'll know the forecast is working when these are true:
- Every number on the board slide reconciles to the ARR bridge, and the bridge reconciles to billing.
- A board member can ask "what happens if enterprise NRR drops 10 points" and you answer in 60 seconds, not next week.
- The monthly re-forecast moves by small amounts, because the base case was honest to begin with.
- New reps are loaded with a ramp curve, not full quota on day one.
- The downside scenario is specific enough to name the triggers that would put you in it.
When those hold, the forecast stops being a quarterly fire drill and becomes a steering instrument. That's the whole point — the board trusts the number because the method is visible, and you trust it because you watched it survive a miss.
This is also where an AI CFO layer earns its keep: pulling the bridge straight from billing, re-forecasting on real cohort retention, and flagging when an actual diverges from the model before the quarter closes — instead of three weeks after. For the broader tooling question, see our guide to revenue forecasting software and the deeper cut on MRR software.
FAQ
Q: What's the difference between ARR forecasting and MRR forecasting? A: Mechanically, almost none — ARR is MRR × 12 for pure-subscription revenue. The difference is framing. Boards and investors talk in ARR; finance ops usually models in MRR because monthly cohorts are where churn and expansion actually move. Forecast in MRR, present in ARR.
Q: How far out should an ARR forecast go? A: Run a detailed monthly model 12–18 months out and a lighter quarterly view to 24–36 months. Past 18 months the driver assumptions are guesses; don't pretend otherwise with monthly granularity.
Q: Should I forecast ARR top-down or bottom-up? A: Bottom-up from the ARR bridge for the number you manage to. Use a top-down market-size check only as a sanity test — if bottom-up says you'll triple but top-down says the segment can't support it, something's wrong.
Q: How do I handle a few large customers that dominate ARR? A: Model concentrated accounts individually, not in the cohort average. If your top five logos are 40% of ARR, their renewal dates and expansion paths belong in the model by name, with a downside scenario where one churns.
Q: How often should I re-forecast? A: Monthly for the rolling 12-month view, with a deeper quarterly reset. The monthly cadence keeps the forecast honest and catches drift while you can still act on it.
Q: What ARR growth rate is "good"? A: It depends on stage — early companies can post triple-digit growth off a small base, while a $30M ARR company growing 40% with strong retention is in great shape. The rate matters less than the quality of retention underneath it. High growth on weak NRR is a leak you're outrunning, not a business you're building.
Q: Can I forecast ARR in a spreadsheet, or do I need software? A: A disciplined spreadsheet with a real ARR bridge beats bad software every time. Move to a tool when the spreadsheet becomes a single fragile file one person maintains, or when you need cohort retention pulled automatically from billing rather than rekeyed.
The takeaway
Stop forecasting ARR with a multiplier. Build the bridge — new, expansion, contraction, churn — forecast each movement from its own driver, segment the retention curve, ramp your new hires, and run three scenarios so you know how fragile the plan is before the quarter, not after. The forecast that survives a board meeting is the one that survived a down quarter first.
Run your ARR forecast on real cohort retention, not last year times 1.4.
