Forecasting SaaS Revenue Bottom-Up:
The MRR Waterfall of New, Expansion and Churn

Short answer

Forecast SaaS revenue bottom-up with a monthly MRR waterfall: starting MRR plus new MRR plus expansion MRR minus churned and contraction MRR equals ending MRR. Drive each line with operational inputs — leads, conversion rates, ARPA, retention by segment — rather than a market-share percentage, and validate the output against your net revenue retention.

"We'll capture 1% of a $5B market" is not SaaS revenue forecasting — it's a wish with a spreadsheet attached. The forecasts that survive contact with reality (and with investors) are built the other way: from the bottom up, one driver at a time, in a monthly MRR waterfall. G Squared Partners' December 2025 analysis notes that most SaaS forecasts break by Q2 for one reason — churn and expansion get rough guesses while new bookings get all the rigor. Here's how to build the waterfall properly, in five steps.

How do you forecast SaaS revenue bottom-up?

The spine of the model is one equation, repeated month after month:

Starting MRR + New MRR + Expansion MRR − Churned MRR − Contraction MRR = Ending MRR

Each month's ending MRR becomes the next month's starting MRR. Multiply ending MRR by 12 and you have your ARR trajectory. That's the whole waterfall. The craft is in what feeds each line.

Step 1: Build New MRR from your funnel, not from ambition

New MRR = new customers × average revenue per account (ARPA). New customers, in turn, come from whatever your motion actually is:

• Sales-led: qualified pipeline × stage conversion rates × average deal size, constrained by rep capacity and ramp time
• Product-led: signups × activation rate × free-to-paid conversion × ARPA
• Hybrid: both, tracked separately — the conversion profiles are nothing alike

Every input here is something a team member owns and can be asked about. That's the test of bottom-up: when an investor asks "why does new MRR double in March?", the answer is "two reps finish ramping and webinar-sourced pipeline converts at 22%" — not "momentum."

Step 2: Forecast churn from trailing cohorts

Use trailing three-to-six-month averages for churn, as G Squared Partners recommends — long enough to smooth seasonal noise, short enough to reflect the current product. Split voluntary churn (cancellations) from involuntary churn (failed payments); they have different causes and different fixes, and lumping them together hides both.

Then respect the compounding. A cohort losing 5% of revenue per month is down 46% after twelve months — G Squared Partners' worked example puts a $100,000 January cohort at $53,944 by December. Small-sounding monthly rates do enormous annual damage, which is why a half-point churn improvement is often worth more than a big top-of-funnel push.

Step 3: Model expansion by mechanism, not by percentage

The laziest line in most SaaS models is "expansion: 2% of MRR per month." Real expansion comes from countable mechanisms:

• Seat growth as customer teams expand
• Tier upgrades when usage crosses plan limits
• Usage-based overages
• Cross-sell of add-on modules

Model each with its own driver — for instance, "18% of accounts add seats each quarter, at +1.8 seats on average." G Squared Partners lists blanket expansion rates among the four most common forecasting mistakes for good reason: a percentage can't tell you what to do when expansion slows. A mechanism can.

Step 4: Segment, or you're guessing

One churn rate across your whole base is a fiction. Enterprise accounts, SMB accounts, annual contracts, monthly contracts, discount-campaign cohorts — each retains and expands differently. Benchmarkit's 2025 benchmarks show retention rising with contract value: enterprise contracts retain best, small annual-contract-value accounts worst. Segment the waterfall by at least customer size tier and acquisition cohort, and forecast each segment with its own rates.

Step 5: Validate against NRR

Net revenue retention = (starting MRR + expansion − churn − contraction) ÷ starting MRR, for existing customers only. It's the single best sanity check on your waterfall. Benchmarkit's 2025 data puts the median at 101% and the 75th percentile at 110%. If your forecast quietly assumes 125% NRR while your trailing actuals say 98%, the model isn't optimistic — it's broken. NRR is also one of the metrics that move your valuation most, so investors will run this check whether you do or not.

A worked MRR waterfall

Twelve months compressed to four checkpoints. Inputs: starting MRR €80,000; new MRR growing from €6,000 to €9,500/month as two reps ramp; gross churn 3.5%/month; expansion mechanisms worth ~1.5%/month.

Ending ARR: ~€1.68M, up 75% year over year — every euro of it traceable to a driver someone owns. Change any input and the whole waterfall recalculates; that's what makes the model useful for scenario and sensitivity analysis rather than a one-shot slide.

The one number to watch before the waterfall moves

Our rule of thumb: your forecast fails in the leading indicators before it fails in MRR. Login frequency falling, support tickets climbing, seats hitting license limits, payment retries spiking — G Squared Partners' data suggests these signals surface churn and expansion shifts 30–60 days before the revenue line moves. Wire the top three into the same dashboard as the waterfall. A forecast you can act on beats a forecast that's merely accurate — and the same discipline underpins budgeting vs forecasting done properly.

What this means for you

Build the waterfall once, bottom-up, with segmented churn, mechanism-level expansion and funnel-driven new MRR — then refresh it monthly as actuals land. Hold it to two standards: every driver has an owner, and the implied NRR matches what your cohorts actually do. Do that and revenue stops being the least credible tab in the financial model VCs actually read and becomes the most defensible one.