A model built on one growth number and one churn number feels precise, but it hides the two things that decide whether investors and you can trust it: how each group of customers actually behaves over its lifetime, and how wrong your key assumptions can be before the plan breaks. Layering cohort behaviour answers the first. Sensitivity analysis answers the second. This lesson shows you how to add both without turning your spreadsheet into a science project.
Why one blended number lies
A blended metric mixes brand new customers with loyal old ones. If you write "churn is 5 percent," you may be hiding the reality that month one churn is 15 percent and month twelve churn is 2 percent. The average either flatters a weak product or scares you about a healthy one. Cohorts fix this by separating customers into groups so you can see behaviour change over time.
Building a cohort view
A cohort is a set of customers who share a start point, usually everyone acquired in the same month, or through the same channel. Lay them out as a triangle: cohorts as rows, and months since acquisition as columns.
- Retention curve: the percent of a cohort still active at month 1, 2, 3 and so on. Healthy curves flatten instead of falling to zero.
- Net revenue retention: the revenue a cohort generates over time. With upsells and expansion this can exceed 100 percent, which is the strongest signal a model can carry.
You do not need years of data. Even three to six months of real cohorts beats an invented single average.
Feeding cohorts into the model
Instead of "revenue equals customers times ARPU," project each month's new cohort, apply your retention curve to every living cohort, and sum the survivors. This bottoms-up revenue line naturally reflects decay and expansion. One India note: quote revenue net of the 18 percent GST on software and SaaS services. GST is a pass-through you collect and remit, not your revenue, so keep ARPU and CAC GST-exclusive to keep cohort economics clean.
Sensitivity analysis: stress what matters
You cannot know CAC, churn, conversion and price precisely, so sensitivity analysis asks a simple question: if this assumption is worse than I hope, what happens to runway, revenue or burn? Three tools cover almost every founder need.
- One-way table: vary a single input and watch one output. For example, months of runway against monthly churn from 3 to 10 percent.
- Two-way table: vary two inputs together, such as CAC and conversion rate, against your cash-out date.
- Scenarios: bundle assumptions into base, downside and upside cases, then confirm you survive the downside.
The discipline is focus. Stress only the two or three drivers that actually move the outcome, usually churn, CAC, sales conversion and price. Testing every cell produces noise, not insight.
India inputs worth stressing
- Grant timing: the Startup India Seed Fund Scheme offers DPIIT-recognised startups up to ₹20 lakh as a grant, but it is disbursed in milestone-based instalments, not a lump sum on day one. Model the tranches arriving later than promised.
- Collections: B2B receivables in India commonly stretch across 60 to 90 days. Stress cash-in timing separately from revenue recognition, because a profitable model can still run out of cash.
- Price: test willingness to pay net of the 18 percent GST your customer sees on the invoice.
A simple routine to follow
- Build a cohort retention triangle from whatever real data you have.
- Project revenue bottoms-up by summing surviving cohorts each month.
- Pick your top three assumptions and build a one-way table for each against runway.
- Build one two-way table for your two biggest drivers.
- Write base, downside and upside scenarios, and check the downside still leaves you enough months of cash to react.
Do this and your model stops being a single confident guess. It becomes a range you understand, which is exactly what a sharp investor, and a calm founder, wants to see.