Our CFO is asking for a revenue forecast. With the unpredictability of trials and churn, how does anyone create a model that isn't complete fiction?

Payments & Revenue
1

Trial conversion rates swing wildly month to month and churn seems random despite our best efforts.

Every forecast I build feels like I’m just making up numbers. What actually works for predicting subscription revenue when everything feels so volatile?

2

Focus on what you can control - don’t try to predict chaos.

I tackle acquisition first. Set a trial budget, track cost per trial for a few months, then multiply by your conversion rates from the same timeframe.

Match your forecast period to marketing cycles. Running weekly campaigns? Monthly forecasts will look insane since spending isn’t even.

For churn, check user behavior patterns instead of just percentages. Users who finish onboarding stick around way longer than those who bail early.

3

I break my forecasts into two chunks - what I’ll keep from current subscribers and what I’ll add from new customers.

Baseline’s easy. Take current MRR, factor in churn rates, done. Way more predictable than trying to guess new signups.

For growth, I flip it around and start with ad spend. Spending $50k next month with a $25 CAC? That’s about 2000 trials. Then I just use conversion rates from when I spent similar amounts before.

Here’s the thing - match your forecast timing to how you actually acquire customers. Running big quarterly campaigns? Don’t forecast monthly or your numbers will be garbage.

And always add 15-20% buffer. CFOs would rather you be conservative than surprised.

4

Try three-month rolling averages instead of monthly data. They’ll smooth out the wild fluctuations and give you more realistic numbers to work with.

5

Use worst month numbers and add buffer.

6

Skip monthly averages - build your model around cohorts instead. Track each month’s new trial users separately and see how they convert and churn over time. You’ll catch patterns that get buried in monthly totals. Need at least 6 months of cohort data to start. Use median conversion rates from your cohorts, not averages. Medians won’t get skewed by weird outlier months. Forecast conservatively and update monthly when new data rolls in.